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.1006790 | Mycobacterium tuberculosis-induced miR-155 subverts autophagy by targeting ATG3 in human dendritic cells | Autophagy is a primordial eukaryotic pathway, which provides the immune system with multiple mechanisms for the elimination of invading pathogens including Mycobacterium tuberculosis (Mtb). As a consequence, Mtb has evolved different strategies to hijack the autophagy process. Given the crucial role of human primary dendritic cells (DC) in host immunity control, we characterized Mtb-DC interplay by studying the contribution of cellular microRNAs (miRNAs) in the post-transcriptional regulation of autophagy related genes. From the expression profile of de-regulated miRNAs obtained in Mtb-infected human DC, we identified 7 miRNAs whose expression was previously found to be altered in specimens of TB patients. Among them, gene ontology analysis showed that miR-155, miR-155* and miR-146a target mRNAs with a significant enrichment in biological processes linked to autophagy. Interestingly, miR-155 was significantly stimulated by live and virulent Mtb and enriched in polysome-associated RNA fraction, where actively translated mRNAs reside. The putative pair interaction among the E2 conjugating enzyme involved in LC3-lipidation and autophagosome formation-ATG3-and miR-155 arose by target prediction analysis, was confirmed by both luciferase reporter assay and Atg3 immunoblotting analysis of miR-155-transfected DC, which showed also a consistent Atg3 protein and LC3 lipidated form reduction. Late in infection, when miR-155 expression peaked, both the level of Atg3 and the number of LC3 puncta per cell (autophagosomes) decreased dramatically. In accordance, miR-155 silencing rescued autophagosome number in Mtb infected DC and enhanced autolysosome fusion, thereby supporting a previously unidentified role of the miR-155 as inhibitor of ATG3 expression. Taken together, our findings suggest how Mtb can manipulate cellular miRNA expression to regulate Atg3 for its own survival, and highlight the importance to develop novel therapeutic strategies against tuberculosis that would boost autophagy.
| Mycobacterium tuberculosis (Mtb) is one of the most successful pathogens in human history and remains the second leading cause of death from an infectious agent worldwide. The major reason of Mtb success relies on its ability to evade host immunity. Autophagy, a cellular mechanism involved in intracellular pathogen elimination, is one of the pathways hijacked by Mtb to elude the control of dendritic cells (DC), major cellular effectors of immune response. Recently, it has become clear that Mtb infection not only alters cellular gene expression, but also controls the level of small RNA molecules, namely microRNAs (miRNAs), which function as negative regulators of mRNA translation into protein. In the present study, we observed that the infection of human DC with Mtb leads to a strong induction of host miR-155, a critical regulator of host immune response. By mean of miR-155 induction, Mtb reduces Atg3 protein content, a crucial enzyme needed for the initial phase of the autophagic process. Interestingly, miR-155 silencing during Mtb infection restores Atg3 level and rescues autophagy. These findings contribute to better elucidate Mtb-triggered escape mechanisms and highlight the importance to develop host-directed therapies to combat tuberculosis based on autophagy boosting.
| As estimated by the 2016 global WHO report, Mycobacterium tuberculosis (Mtb), the causative agent of tuberculosis (TB) in humans, infects approximately 10 million people each year, making TB the second leading cause of death from an infectious disease worldwide [1].
The success of Mtb principally relies on its ability to perfectly adapt to the host, by establishing latent infection and evading from the control driven by immune system. In this scenario, both innate and adaptive immune responses are required to control TB progression and pathogenesis [2].
In particular, the interaction of mycobacteria with antigen-presenting cells (APC) is a key feature in the pathogenesis of TB and the outcome of this interaction is pivotal in determining whether immunity or disease would ensue. Human and murine macrophages and dendritic cells (DC) were shown to be infected by mycobacteria and, in turn, to produce a specific immune response. As professional APC, DC display indeed an extraordinary capacity to stimulate naïve T cells and initiate primary immune responses [3, 4].
Once in contact with the pathogen, immature DC undergo maturation, modifying their phenotypic features and producing several pro-inflammatory and regulatory cytokines with the ability to tune the immune response by acting on different cellular populations [5]. These events result into a fine control exerted at transcriptional, post-transcriptional and post-translational level. In turn, host factors, utilized by mycobacterial molecules, play an important role in the regulation and progression of the infection [6].
In light of these observations, it is conceivable that Mtb infection might not only alter cellular gene expression but also control important regulators of mRNA translation, such as microRNAs (miRNAs). As a class of small non-coding RNAs, miRNAs are highly conserved within eukaryotic species and function as key regulators of gene expression by post-transcriptional targeting of mRNAs for translation arrest or for degradation [7]. These effects are mediated by imperfect binding of the miRNA-recognition elements within the 3' untranslated region (UTR) of the target mRNAs to the miRNA ‘seed’ region located between residues 2–8 at the 5' end [7].
In addition to the known role of miRNAs in the multi-level regulation of several cell processes and mammalian immune system [7], growing evidences show how pathogens profoundly perturb host miRNA expression. While the involvement of miRNAs in viral immune responses has been well investigated [8, 9], targeting of host miRNAs by bacterial pathogens is less characterized. Though, in recent years regulation of miRNA expression has been recognized as a molecular strategy exploited by bacteria to manipulate host cell pathways [10]. In particular, bacteria may redirect several host cell functions to sustain survival and/or replication. Most often, these effects require the delivery into the host cytoplasm of bacterial effector proteins through specific secretion systems [11]. Among closely related bacterial species, the mycobacterium genus is one of the most effective in inducing differential regulation of miRNAs [12, 13].
We set out to identify miRNAs perturbed by Mtb in infected human primary DC and understand their pathological relevance. We found that in human DC, Mtb induces miR-155 to negatively regulate ATG3, an E2-ubiquitin-like conjugating enzyme involved in autophagosome formation, thereby impairing autophagy.
To investigate whether Mtb modulates the immune response by acting on the host DC miRNome, we perfomed a global analysis of miRNA expression profile in human primary DC infected with Mtb.
RNA was prepared from DC infected in kinetic with live H37Rv Mtb for 3, 8 and 24 hours and microarray experiment performed. A hierarchical clustering was constructed by considering only miRNAs whose overall dynamic profile was significantly modulated in Mtb infected versus uninfected cells (Fig 1). The complete list of the 68 probes differentially expressed, corresponding to 43 different miRNAs is shown in Table 1. While only a few miRNAs were down-regulated as a consequence of Mtb challenge, the majority of them were found up-regulated in Mtb-infected DC (Fig 1). In particular, a high number of miRNAs showed a fold change (FC) greater than 2, starting from 3 hours post infection and in all the analyzed time-points, while only two miRNAs, namely miR-548m and miR-548n, were down-regulated with FC lower than -2 after 24 hours of Mtb infection (Fig 1).
To match our miRNA data with those observed in vivo during TB disease, the list of de-regulated miRNAs identified in infected DC was intersected with the miRNA profile found in patients affected by active TB [14–28]. Interestingly, among miRNAs de-regulated in TB patients and validated by real time PCR or similar techniques [14–28], seven miRNAs—namely miR-155, miR-155*, miR-29b-1*, miR-150, miR-146a, miR-212 and miR-483-5p –were found altered in Mtb-infected DC (Figs 1 and 2A). Furthermore, microarray data indicated that Mtb induced miR-29b-1*, miR-150, miR-212 and miR-483-5p at early time points (3 and 8 hours) (Figs 1 and 2A). MiR-150 and miR-212 induction was maintained over time, while 24 hour post-infection, miR-29b-1* decreased and miR-483-5p peaked (Fig 2A). Conversely, a different kinetic of expression was instead observed for miR-146a, miR-155 and its star filament, miR-155*, displaying an increasing level over time during Mtb infection (Fig 2A).
Microarray data were then validated by quantitative PCR analyses (q-PCR) in an independent cohort. According to microarray, miR-155, miR-155* and miR-146a were significantly induced by Mtb during human DC infection in a time dependent manner and, in particular, miR-155 and its star filament were among the top up-modulated miRNAs in our data set (Fig 2B).
On the contrary, although miR-29b-1*, miR-150, miR-212 and miR-483-5p induction was confirmed by q-PCR experiments, their kinetic of induction does not perfectly mirror that showed by microarray analysis (Figs 1 and 2). In particular, the levels of miR-212 and miR-483-5p induced by Mtb in infected DC and measured by q-PCR are weaker than those detected by microarray experiments while miR-29b-1* and miR-150 induction peaked at different time after Mtb infection when measured with the two techniques (Figs 1 and 2).
Since a single miRNA might regulate different mRNAs involved in distinct physiological processes, a functional gene ontology (GO) annotation study was conducted to identify those pathways and processes significantly represented within the list of targets of the seven selected miRNAs (for complete lists see S1 Table).
In particular, by taking advantage of the performed GO analysis, we noticed that, while miR-29b-1*, miR-150, miR-212 and miR-483-5p showed an enrichment in GO terms important for host-response to Mtb infection such as ubiquitin, proteasome degradation and endocytosis, only miR-155, miR-155* and miR-146a displayed a significant enrichment in biological processes linked to autophagy, a process deeply implicated in Mtb infection control [29, 30] (S2 Table).
Having previously demonstrated that live and virulent Mtb is able to interfere with the autophagy pathway in human DC [29], we compared the expression of miR-155, miR-155* and miR-146a in DC stimulated with live or heat-inactivated Mtb (HI Mtb), the avirulent BCG vaccine, as well as LPS, as positive control (Fig 3).
While the expression of the three analyzed miRNAs was indistinctly promoted by LPS treatment, we interestingly found a different picture upon stimulation with the mycobacteria (Fig 3). Indeed, BCG and HI Mtb poorly induced miR-155 expression in treated DC as compared to live Mtb strain, suggesting that the modulation of miR-155 requires stimulation with virulent and live bacterium (Fig 3A). In contrast, miR-155* expression was similarly induced in response to both live and HI Mtb or BCG (Fig 3B). A different scenario was observed for miR-146a, induced in DC by all the analyzed stimuli at the different time points with no significant differences (Fig 3C).
Since the expression of miR-155 was preferentially stimulated by live and virulent Mtb strain, we focused on this miRNA and further investigated its function in the context of Mtb infection of human DC.
Small RNAs, like miRNAs and RNA not associated with ribosome chains, principally reside in the light polysome fraction (constituted of monosome and single ribosome subunits), whereas actively translated mRNAs primarily occupy the heavy polysome fraction (rich in long ribosome chains). However, it has been recently shown that miRNAs can be co-purified with polysomes and their specific low (light fraction) or high (heavy fraction) polysome occupancy appears to reflect the strength of interaction with their target mRNAs [31]. Thus, we investigated whether the Mtb-driven miR-155 induction correlates with its localization in heavy polysome fraction by digital PCR approach (Fig 4). This allowed us to determine the absolute copy number of miR-155 in total as well as low and high occupancy polysome-associated RNA samples in the absence of a reliable housekeeping small RNA.
Interestingly, a 10-fold higher miR-155 copy number was found in total RNA samples from Mtb-infected DC and this increase was even higher (30-fold) in polysome-associated RNAs. Few miR-155 copies were instead associated to the light fraction of untreated and Mtb-infected DC (Fig 4A). As expected, the highest copy number of RNU6b - a commonly used housekeeping small RNA—was found in light fraction irrespective of Mtb infection (Fig 4B). Of note, after Mtb infection of human DC, the other two mycobacteria-induced miRNAs, namely miR-146a and miR-155*, mainly resided in light associated polysome fraction, although a low number of copies of those miRNAs was also present in heavy polysome associated RNA samples (S1 Fig). Taken together, these data indicate that upon Mtb infection the induced miR-155 is specifically associated with long ribosome chains where it may actively interact with its target mRNAs.
We performed an in silico target prediction analysis of miR-155 by employing and comparing five different algorithms: Targetscan (http://www.targetscan.org/), Pictar (http://pictar.mdc-berlin.de/), Diana-microT (http://diana.imis.athena-innovation.gr/DianaTools/index.php?r=microT_CDS/index), miRanda (http://www.microrna.org/microrna/home.do) and miRwalk (http://www.umm.uni-heidelberg.de/apps/zmf/mirwalk/index.html). In addition to several predicted and, in some case, validated targets of miR-155 [32–34], amongst the miR-155 targets, all 5 databases identified ATG3, an E2-ubiquitin-like-conjugating enzyme with an essential role in autophagosome formation [35, 36] (Fig 5A).
To validate the specificity of miR-155-ATG3 mRNA 3′-UTR pair interaction, a dual luciferase assay reporter vector containing the ATG3 3′-UTR sequence was transfected in HEK293 cells, together with a miR-155 mimic (hsa-miR-155-5p), a chemically synthesized, double-stranded RNA, which "mimics” the mature endogenous miR-155. A commercial random sequence miRNA mimic molecule was used as a negative control mimic. With this setting, a significant and selective reduction of luciferase activity was observed upon transfection with the miR-155 mimic, thus confirming a functional interaction between miR-155 and the ATG3 3’-UTR sequence (Fig 5B).
Next, we sought to correlate miR-155 expression with Atg3 protein levels, by studying the effects of the transfection of miR-155 mimic or negative control elicited in DC. Intriguingly, miR-155 mimic, but not the scrambled control, caused a dose-dependent reduction of Atg3 protein (about 30–50%), confirming that this enzyme is an authentic miR-155 target in human primary DC (Fig 5C).
As previously mentioned, Atg3 is an E2-ubiquitin-like-conjugating enzyme with a key role in the autophagy pathway, which, in a complex with the E1-ubiquitin-like-activating enzyme Atg7, transfers LC3 or GABARAP proteins to the phosphatidylethanolamine (PE) of the forming isolation membrane allowing their binding [37]. Accordingly, the low Atg3 protein abundance in miR-155 mimic-transfected DC correlated with the reduction of LC3-II and GABARAP (Fig 5C). Although the transfection procedure, requiring nutrient deprivation, per se promoted LC3 lipidation, a reduction of LC3-II was found only in response to miR-155 mimic transfection, while no effect was observed with the scrambled control. Interestingly, no variation in the abundance of Atg7 was observed, indicating that LC3-II reduction was strictly and specifically dependent on miR-155-mediated regulation of Atg3 expression (Fig 5C).
Based on these findings, we next tested if Mtb-induced miR-155 would impair autophagosome formation by reducing Atg3 protein level and, in turn, affecting the capacity of DC to convert the autophagosome marker LC3 in its lipidated form. To this end, immunoblot analysis showed a significant decrease in the level of Atg3 24 hours after Mtb infection, which lasted up to 32 hours (Fig 6A), in tandem with increasing Mtb-driven miR-155 expression (S2 Fig). The reduction in Atg3 abundance driven by the infection correlated with an impairment of the lipidated form of LC3 (LC3-II) and GABARAP, particularly at 32 hours post-infection (Fig 6A). Furthermore, while the reduction in LC3 conversion was evident 32 hours post-infection, Mtb promoted LC3-II accumulation in 24 hour-infected DC, given its well-known ability to block autophagosome-lysosome fusion [29]. The variation in the abundance of LC3-II detected by immunoblotting was further confirmed by the significant decrease in the number of LC3 puncta per cell, as enumerated by confocal microscopy analysis in DC infected for 32 hours with Mtb as compared to those present in 24 hours-infected cells (Fig 6B and 6C).
Since expression of ATG3, LC3 and GABARAP mRNAs was unchanged upon Mtb challenge (Fig 7A), it is likely that the miR-155 mediated reduction of Atg3 impacts at a post-transcriptional level on LC3 and GABARAP processing and, in turn, on autophagosome formation. In line with q-PCR data, no variation in ATG3 mRNA copy number was detected by using digital PCR on total RNA samples from Mtb-infected DC as compared to the untreated counterpart (CTRL) (Fig 7B). Conversely, association of ATG3 mRNA to high occupancy polysome chains was significantly induced by Mtb infection as evident from the 2.5 fold induction of ATG3 expression measured in polysome-associated RNA fraction (Fig 7B). Interestingly, our analysis highlighted that Mtb infection favors the association of both miR-155 and ATG3 mRNA on polysome chains (Figs 4A and 7B), thus supporting the miR-155-mediated regulation of ATG3 protein level in DC during Mtb infection.
To test whether silencing of Mtb-induced miR-155 impacted on autophagy, Mtb-infected DC were transfected with a specific inhibitor of miR-155 (α-miR-155 mimic) or with a non-specific miRNA inhibitor (CTRL mimic) (Fig 8). Silencing efficiency was assessed by analyzing the level of miR-155 expression in infected cells. An important (≈75%) reduction was observed only in the presence of the specific inhibitor (α-miR-155 mimic) (Fig 8A). As expected, by antagonizing Mtb-induced miR-155 expression in infected DC, the ATG3 protein content was restored at a level similar to that found in untreated and un-transfected control cells (Fig 8B). Moreover, following the specific inhibition of miR-155, DC recovered their capacity to form autophagosomes in response to Mtb infection, as demonstrated by the significant increase in the number of LC3 puncta per cell (Fig 8C and 8D). Notably, the selective silencing of miR-155 in Mtb-infected DC consistently stimulated the fusion of rescued autophagosomes with lysosomes as demonstrated by the increased co-localization of the LC3 signal with the lysosomal marker, LAMP-1, thus, allowing the recovery of the Mtb-blocked autophagic flux (Fig 9).
During their coevolution with the host, successful pathogens, like Mtb, have evolved a myriad of strategies to evade and, in some cases, subvert innate and adaptive immune responses [6]. These include targeting of autophagy, a cellular mechanism usually devoted to recycle damaged organelles or macromolecules, which also plays a crucial role in counteracting continuous microbial insults [37]. More specifically, the importance of the autophagic process in host immune response against Mtb has been demonstrated by the fact that virulent mycobacteria may impair phagosome maturation by altering the acidic, hydrolytic environment of the intracellular compartment needed to kill the bacteria, thus promoting their own survival [30]. Moreover, the impairment of autophagy machinery compromises antigen-processing capacity [38, 39] thus, affecting specific host-immune response. Lastly, in Mtb-infected human DC, the block of autophagosome maturation could be overcame by the autophagy inducer, rapamycin, which simultaneously boost a protective Th1 response [29].
In spite of the fact that autophagy has been a fast-growing research area in recent years, not much was known about immune-evasion mechanisms exploited during both viral and bacterial infections and acting on the E2-like conjugating enzyme Atg3, critical for LC3 lipidation during autophagosome formation [37]. Here, we show for the first time the capacity of Mtb to manipulate autophagy, by impairing autophagosome formation, through the host miR-155-mediated targeting of ATG3 in infected human DC.
The mechanism described in the present study is also in support of the involvement of the Mtb-triggered de-regulation of host miRNA profile, which, in turn, interferes at post-transcriptional level with the fine-tuned DC gene expression program, crucial for the establishment of an appropriate immunological outcome [13].
In addition to miR-155, we identified several others miRNAs induced in Mtb infected DC, as miR-155*, miR-29b-1*, miR-150, miR-146a, miR-212 and miR-483-5p, whose expression was also found altered in patients affected by pulmonary TB [14–28]. A mechanistic explanation concerning the function of some of these miRNAs along mycobacterial infection has been already proposed in previous in vitro studies [40–45]. In particular, Mtb stimulates the expression of miR-29b-1* to target IFN-γ in both NK and CD4+ T cells [40]. A negative regulation of the TLR2 signaling was instead demonstrated for miR-150 although in the context of BCG infection [41]. A dual function was shown for miR-146a, which acts on PTGS2, thus increasing macrophages ability to kill BCG, and reduces TNF-α release by targeting both IRAK1 and TRAF6 [42, 43]. Finally, several roles were proposed for miR-155 in the context of mycobacterial infection in both innate and adaptive immune cells spanning from maintenance of cell survival to interference with the autophagy pathway [44–46].
Given the critical role played by autophagy in Mtb infection control [47], we performed a GO analysis to get insight into the potential functions of Mtb de-regulated miRNAs with a particular focus on this cellular process. The analysis of putative miRNA target lists revealed a significant enrichment in autophagy-linked biological processes for three of the selected miRNAs, namely miR-155, miR-155* and miR-146a.
Among these miRNAs, we focused our attention on miR-155 because its expression was specifically induced by live Mtb, and not by HI Mtb or by the vaccine strain BCG. This feature is particularly interesting in light of our previous data showing that DC differentially respond to live, dead or attenuated Mtb by promoting the expression of a specific cytokine milieu and by modulating the autophagic process [5, 29, 48].
Although the potential correlation between miR-155 and mycobacterial infection was already proposed [13, 17, 32, 44], the presence of miR-155 into the polysome-associated RNA fraction is a novel finding supporting the idea that in this subcellular localization miR-155 can engage actively translated mRNA transcripts to specifically control gene expression in Mtb-infected DC.
Interestingly, data obtained so far in human DC are in support of the interaction between miR-155 and ATG3 indicated by our in silico target prediction analysis. Indeed, we found a decrease in Atg3 protein level at late time points of Mtb infection together with the impairment of LC3 puncta number per cell. Notably, LPS treatment does not alter LC3-II level in human DC in spite of its ability to strongly induce miR-155 expression (S3 Fig), thereby uncovering a previously unidentified Mtb-driven effect of miR-155 on ATG3 expression.
MiR-155-dependent control of ATG3 expression was confirmed by luciferase assay and by transfection experiments conducted in human primary DC with a specific miR-155 mimic and showing that miR-155 down-regulates Atg3. This impairment translates into a reduced ability of the complex Atg7/Atg3 to convert LC3-I in LC3-II and in a diminished capacity to process the LC3 homologue, GABARAP. Indeed, Atg7, which was not altered as a consequence of miR-155 mimic transfection, activates and transfers LC3-I to the E2-ubiquitin-like-conjugating enzyme, Atg3, essential for the final conjugation of LC3 to the PE group [49]. These events allow the assembling of the membrane-bound form, LC3-II, to the newborn autophagosome structures.
Notably, by using an anti-miR-155 mimic, we were able to restore Atg3 protein and its capacity to form autophagosomes during Mtb infection of human DC. Importantly, the silencing of miR-155 rescued autophagic flux commonly altered in DC as a consequence of Mtb infection. These data strongly support the role of miR-155 in controlling the autophagic process in human DC and suggest the miR-155/ATG3 axis as a critical component in the management of autophagosome formation as well as of the autophagic flux during Mtb infection.
In addition to the mechanism here demonstrated for Atg3 protein, it has been reported that another autophagy protein involved in autophagosome formation, namely Atg5, has an autophagy-independent protective function in neutrophil-mediated immunopathology during Mtb infection, however, its role seems to be dispensable in myeloid cells [50].
It was recently demonstrated that the miR-155-mediated regulation of the survival pathway in innate and adaptive immune cells leads to opposite impacts with regard to Mtb containment [44]. Accordingly, depending on cell type and stimuli, miR-155 might act as inducer as well as repressor of the autophagic process [45, 51–53]. Indeed, a pro-autophagic function for miR-155 was shown by its capacity to target multiple players within the mTOR cascade as Rheb, although in response to hypoxia or to infection with avirulent Mtb or BCG vaccine [45, 51]. Conversely, Holla and colleagues demonstrated that miR-155 limits IFN-γ-induced autophagy in murine macrophages infected with BCG or Mtb [52]. This is consistent with a negative role for miR-155 in human DC infected with Mtb, as proposed in the present study. In line with this view, it has been suggested that inhibition of this miRNA hinders Mtb survival [53]. This further underlines that Mtb might induce miR-155 to its own end. Moreover, it was also demonstrated that in addition to miR-155, Mtb engages other miRNAs, such as miR-33 and its star filament miR-33*, to limit the autophagic pathway, whose up-regulation in infected macrophages leads to repression of multiple autophagy effector molecules [54]. The notion that Mtb might use cellular miRNAs to interfere with autophagy process has been also confirmed by other recent studies showing that Mtb-mediated alteration of miR-125a, miR-17 as well as miR-144* expression levels undermines the anti-microbial activity of autophagy in infected macrophages or monocytes [55–57].
Based on these findings, it is clear that Mtb can manipulate the expression of cellular miRNAs in order to enhance its survival inside the host and, our study, highlights a novel Mtb-triggered escape mechanism involving cellular miR-155 by which Mtb antagonizes the autophagic process via a previously unidentified ATG3 reduced expression (Fig 10).
Taken together our data suggest that reconstitution of normal autophagy by antisense miR-155 molecules could represent an innovative approach for exploiting novel host-directed therapeutic strategy to combat TB.
Monoclonal Abs specific for CD1a, CD14, CD86 (BD Bioscience, San Jose, CA, USA) and Fixable Viability Dye (FvDye, eBioscience, San Diego, CA, USA) were used as direct conjugates to fluorescein isothiocyanate (FITC), phycoerythrin (PE) or eFluor780 as needed.
For immunoblotting analysis, goat anti-Atg7 (Santa Cruz Biotechnologies, Santa Cruz, CA, USA), rabbit anti-Atg3 (Cell Signaling, Danvers, MA, USA), rabbit anti-GABARAP (Santa Cruz Biotechnologies), rabbit anti-LC3 (Cell Signaling), mouse anti-actin (Sigma-Aldrich, St. Louis, MO, USA), and horseradish peroxidase-conjugate secondary anti-goat, anti-mouse and anti-rabbit Abs (Santa Cruz Biotechnology) were used.
Where indicated, LPS (Sigma-Aldrich) was used at 1 μg/ml.
Istituto Superiore di Sanità Review Board approved the present research project (CE/13/387). No informed consent was given since the data were analyzed anonymously. Peripheral blood mononuclear cells were isolated from freshly collected buffy coats obtained from healthy voluntary blood donors (Blood Bank of University "La Sapienza", Rome, Italy).
DC were prepared as previously described [5]. Briefly, DC were generated by culturing CD14+ monocytes with 50 ng/ml GM-CSF and 1000 U/ml IL-4 (R&D Systems, Minneapolis, MN, USA) for 5 days at 0.5x106 cells/ml in RPMI 1640 (Lonza, Basel, Switzerland), supplemented with 2 mM L-glutamine and 15% Fetal Bovine Serum (FBS) (BioWhittaker, #DE14-801F). At day 5 the cells were tested for their differentiation status by evaluating CD1a expression (>90% CD1a+) and lack of CD14 (>95% CD14-). Before infection, the medium was replaced with RPMI without antibiotics and supplemented with 2 mM L-glutamine and 15% FBS. Cytokine deprivation did not affect DC survival rate, which was >90%.
For gene reporter assay, HEK293T cells (5x105 for each condition; a generous gift of Dr. Kate A. Fitzgerald—University of Massachusetts Medical School, Worcester, MA) were re-suspended in 100 μl DMEM (Lonza) plus 10% FBS (BioWhittaker without antibiotics and plated in a 96-well plate.
Mtb H37Rv (ATCC 27294; American Type Culture Collection) was grown as previously described [5]. Logarithmically growing cultures were centrifuged at 800 rpm for 10 min to eliminate clumped mycobacteria and then washed three times in RPMI 1640. Mycobacteria were re-suspended in RPMI 1640 containing 10% FBS and then stored at -80°C. Bacterial frozen vials were thawed and bacterial viability was determined by counting the number of colony forming units. All bacterium preparations were tested for endotoxin contamination (<1 Endotoxin Unit/ml) by the Limulus lysate assay (Lonza). DC cultures were then infected with a multiplicity of infection (MOI) of 1 bacterium/cell as previously described [5]. Where indicated, Mtb was heat-killed at 80°C for 1 hour.
Total RNA was isolated from 1x106 infected or not infected DC using mirVana isolation kit (Ambion, Thermo Fisher Scientific, Waltham, MA, USA) and following the manufacturer’s recommendations. An additional phenol/chloroform extraction was performed to inactivate residual mycobacterial particles. RNA was quantified using a Nanodrop spectrophotometer (Nanodrop2000, Thermo Fisher Scientific) and quality assessed with an established cut-off of 1.8 for both the 260/280 and 260/270 absorbance ratios. RNA integrity was instead inspected by bioanalyzer analysis (Agilent Technologies, Santa Clara, CA, USA) considering a cut-off of 1.8 for 28S/18S ratio.
Microarray experiments were carried out on the Human miRNA Microarray platform (V3) (Agilent Technologies). 100 ng of total RNA of experimental sample (infected DC) and of a reference sample (uninfected DC) was labeled with cyanine Cy3 (miRNA Complete Labeling and Hybridization Kit, Agilent Technologies). The labeled experimental and reference samples were denatured and then hybridized on microarray slides containing microRNA capture probes targeting all human and viral miRNAs listed in the Sanger miRBase database, release 12.0. Each slide was placed in a hybridization chamber (Agilent # G2534A) and hybridized in a hybridization oven (Agilent # G2545A) for 20 hours at 55°C. Low- and high-stringency washes were carried out and the microarrays were dried following the Agilent washing protocol. A G2565C microarray scanner (Agilent Technologies) was used to acquire images and Feature Extraction software version 10.1 (Agilent Technologies) was used to quantify hybridization signals. Absent and marginal signals, flagged automatically by the software, were removed from the analysis. Probe intensities were log2-transformed, after adding an offset to accommodate zeros according to the formula [x′ = log2(x+1)], where x′ is the adjusted signal and x the ‘total probe signal’ calculated by Feature Extraction software. Normalization was executed according to the quantile normalization method [58]. Quality assessment was based on both the Feature Extraction Quality Control reports and the pre- and post-normalization MA plots, which allow a pairwise comparison of log-intensities of each array to a reference median array and also the identification of intensity dependent biases. Samples showing a poor reproducibility of the replicate probes or/and evident trends or saturation effects in the corresponding MA plots were excluded from the analysis.
To test for miRNA differentially expressed between CTRL and Mtb-infected DC we applied the “Bounded Area” method [59, 60] to identify the miRNA probesets whose overall dynamic profile is significantly modulated in Mtb.
Briefly, the area bounded by the miRNA profile with respect to its baseline value was evaluated, and the miRNA was considered differentially expressed if this area exceeded a threshold θ. θ is the confidence threshold determined in correspondence to a significance level α, based on the null hypothesis distribution. This latter was derived by repeatedly sampling data from an error distribution, empirically obtained from the available biological replicates as explained in [61].
As already shown, the method is quite robust to random oscillation since the entire expression profile is considered, thus diminishing both false positive and false negative rates with respect to standard tests when few replicates are available. Moreover, both the precision and the recall of the test are enhanced by the fact that the entire expression profile is tested, rather than single time points. In order to account for multiple testing, the significance level α was corrected for multiple testing so to control the global False Discovery Rate at 5%.
MiRNA microarray data have been deposited in Array Express public repository (https://www.ebi.ac.uk/arrayexpress/, accession number: E-MTAB-6083) according to the MIAMEstandards. Hierarchical clustering was performed using Euclidean distance and average linkage method. Bioinformatics analysis was implemented in the R statistical environment.
For RNA fraction isolation, 10x106 cells untreated DC (CTRL) and DC infected for 16 hours with Mtb (MOI 1) were used. Polysome-associated RNA were prepared as previously described [62].
Each gradient was fractionated in BSL3 facility by hand collection from the top of the gradient into 2 samples, light phase (low polysome occupancy) and heavy phase (high polysome occupancy). The hand collection method was previously tested to parallel with fraction collection using a fractionator coupled with an UV optical reader (A 254nm). The obtained fractions were immediately mixed with an equal volume of Trizol (Invitrogen, Thermo Fisher Scientific) for later RNA isolation following Trizol manufacturing instructions.
Total RNA was reverse-transcribed using the TaqMan miRNA reverse transcription kit (Applied Biosystems, Thermo Fisher Scientific) and miRNA-specific primers (Applied Biosystems, Thermo Fisher Scientific) for hsa-miR-155, hsa-miR-155*, hsa-miR-29b-1*, hsa-miR-150, hsa-miR-146a, hsa-miR-212 and hsa-miR-483-5p. MicroRNA expression levels were then analyzed using the appropriate TaqMan miRNA assay (Applied Biosystems) and TaqMan Universal Master Mix II (Applied Biosystems) according to the manufacturer’s instructions. The ubiquitously expressed U6b snRNA was also quantified as an endogenous control and used to normalize miRNAs expression by using the equation 2-ΔCt or 2-ΔΔCt as needed. The values are mean ± SEM of triplicate determinations.
For the analysis of ATG3, LC3 and GABARAP mRNA levels, total RNA was reverse transcribed as previously described [5] and then analyzed using the appropriate TaqMan assay (Applied Biosystems) and TaqMan Universal Master Mix II (Applied Biosystems) according to the manufacturer’s instructions. Transcript expression was normalized to the GAPDH level by using the equation 2-ΔCt; the values are mean ± SEM of triplicate determinations.
Digital PCR experiments were carried out on the BioMark HD System (Fluidigm, San Francisco, CA, USA) by loading hsa-miR-155 or U6b reverse transcribed samples prepared, as described previously, from total RNA, light and heavy phase-associated RNA, into Fluidigm's 37K Digital Array microfluidic chip. Similarly, cDNA samples from total RNA, light and heavy phase-associated RNA, prepared as previously described [5], were used for ATG3 copy number determination by digital PCR.
Fluidigm's 37K Digital Array consists of 48 panels, each of which is further partitioned into 770 reaction chambers. The reaction for each panel was set up with the specific TaqMan miRNA assay probe and with TaqMan Universal Master Mix II (Applied Biosystems) into a final volume of 5 μl. For each sample, six serial dilutions were loaded in triplicate reactions and the chip was then thermocycled and imaged on Fluidigm's BioMark HD real-time PCR system. Positive chambers that originally contained 1 or more molecules has been counted by the Digital PCR analysis software (Fluidigm) and only templates that yielded 150–360 amplified molecules per panel were chosen for technical replication in order to obtain absolute quantification of miRNA copy number.
For the identification of putative target mRNAs of the selected miRNAs, several prediction algorithms, namely miRanda (http://www.microrna.org) [63], Targetcan (http://www.targetcan.org) [64], Pictar5 (http://pictar.mdc-berlin.de/) [65], DIANA-microT (http://diana.imis.athena-innovation.gr) [66] and miRWalk (http://www.umm.uni-heidelberg.de/apps/zmf/mirwalk) [67] were used.
Only targets consistently present in all five target-prediction databases were taken into consideration for the study.
To functionally characterize differentially abundant proteins for biological interpretation, Gene Ontology (GO) analysis was performed. Gene Ontology annotation for the list of putative targets of miR-155, miR-155*, miR-29b-1*, miR-150, miR-146a, miR-212 and miR-483-5p was obtained from miRWalk (http://www.umm.uni-heidelberg.de/apps/zmf/mirwalk). A two-tailed Fisher’s Exact Test with Benjamini-Hochberg multiple testing correction was used to measure the significance of enrichment.
A 174 bp fragment of 3′-UTR of ATG3 mRNA, containing the seed sequence (GCATTAA) of hsa-miR-155-5p was amplified by PCR (forward primer 5′- GAGAGAGCTCGAAGAGAGCATAAAATCTATCCTAA-3′, reverse primer 5′-GAGACTCGAGTTTTTATTAAACAAGTAAGGCTGG -3′). The fragment was designated as ATG3 3′-UTR and inserted into the pmirGLO Dual-Luciferase miRNA target expression vector (Promega, Madison, WI, USA), between the SacI and XhoI restriction sites.
The hsa-miR-155-5p mimic was used for transient transfection in gain of function experiments. As a negative control, a random sequence miRNA mimic molecule (Ambion, Thermo Fisher Scientific), which has been extensively tested in human cell lines and tissues and validated to not produce identifiable effects on known miRNA function, was employed.
For Dual-Luciferase assay, HEK293T cells (5x105 for each condition) were re-suspended in 100 μl DMEM plus 10% FBS and plated in a 96-well plate.
Lipoplexes were prepared in Opti-MEM serum-free medium (Invitrogen, Thermo Fisher Scientific) by mixing the constructed plasmids together with hsa-miR-155-5p or negative control mimics and Lipofectamine 2000 (1mg/ml) (Invitrogen, Thermo Fisher Scientific). Briefly, 400 ng of the pmirGLO Dual-Luciferase miRNA target expression vector plus 5 pmol of specific miRNA or negative control mimics and 1 μl Lipofectamine 2000 were diluted separately in 25 μl of Opti-MEM medium. The two solutions were gently mixed and incubated for 5 minutes at room temperature and then added to HEK293T cells.
After 24 hours, reporter assay was performed using the kit Dual-Glo Luciferase Assay System (Promega), according to manufacturer’s instructions. Signal of firefly and renilla luciferase was detected using a Victor X2 reader (Perkin-Elmer, Waltham, MA, USA).
Predesigned miRNA mimic oligonucleotide for hsa-miR-155 was obtained from Ambion (Thermo Fisher Scientific). A negative-control miRNA mimic [control mimic (Ambion, Thermo Fisher Scientific)] was used to address the specificity of the observed effect to the specific miRNA sequence. SiRNA transfection efficiency was determined by transfecting the BLOCK-iT Fluorescent oligo [a FITC-labeled oligonucleotide (Ambion, Life technologies)] together with the specific miRNA or negative control at 1:2 ratio using Lipofectamine RNAiMAX (Invitrogen, Thermo Fisher Scientific). For each condition, 4x105 DC were pleated in a 12-well plate. Lipoplexes were prepared in Opti-MEM serum-free medium (Invitrogen, Thermo Fisher Scientific) at 1:1 ratio. Briefly, miRNA mimics plus fluorescent oligonucleotide and 10 μl of Lipofectamine (1mg/mL) were dilute separately in 250 μl of Opti-MEM medium.
The Lipofectamine solution was added to the miRNA solution and placed for 5 minutes at room temperature and then added to DC. Four hours after incubation, medium was replaced with fresh Opti-MEM medium supplemented with 15% FCS and 2 mM L-glutamine for at least 18 hours. After the indicated time, cells were harvested for transfection efficiency, protein extraction and RNA isolation from twin wells.
Providing that transfection by itself may perturb DC viability, maturation and activation [68], preliminary experiments transfecting an oligo-FITC as control of transfection were performed to monitor DC viability, transfection efficiency as well as expression of the co-stimulatory molecule CD86 by flow cytometry on a FACS Canto (BD Biosciences) (S4 Fig).
This preliminary setting indicated that transfection did not induce cell mortality (S4A Fig). Moreover, around 70–75% of viable culture DC resulted to be transfected and, as expected, CD86 induction was not affected by transfection (S4B Fig and S4C Fig, respectively).
For ATG7, ATG3 and GABARAP determination 10 μg of total protein extract were separated on 10% NuPAGE Bis-Tris gel (Invitrogen, Thermo Fisher Scientific) and electroblotted onto nitrocellulose membranes (GE Healthcare). For LC3 detection 30 μg of total proteins were used for separation on 13.5% NuPAGE Bis-Tris gel and electroblotted onto polyvinylidene fluoride membranes (GE Healthcare Bio-sciences, Pittsburgh, PA, USA). Blots were incubated with primary Abs in 5% nonfat dry milk in TBS plus 0.1% Tween20 (Sigma-Aldrich) overnight at 4°C. Detection was achieved using horseradish peroxidase-conjugate anti-rabbit, anti-mouse or anti-goat (Santa Cruz) secondary Abs and visualized with Clarity Western ECL Substrate (Bio-Rad, Segrate, Italy). ATG7, ATG3 and GABARAP quantifications were performed by calculating their ratio compared to the actin level by using ImageLab software (Bio-Rad). LC3-II/actin ratio was quantified by ImageLab software according to Klionsky et al [69].
Cells were fixed with 4% paraformaldehyde (Sigma-Aldrich) in PBS followed by permeabilization with 0.2% triton X-100 (Sigma-Aldrich) in PBS. Mycobacteria were stained with Auramine for 20 min and decolorized for 10 min. Cells were then labelled with primary antibody anti-LC3 (Sigma-Aldrich) for 1 hour at room temperature and visualized by means of Cy3-conjugated secondary Ab (Jackson Immunoresearch, Baltimore, PA, USA). For co-localization experiments, after fixation, cells were permeabilized with methanol/acetone (1:1) (Sigma-Aldrich) solution. DC were then labelled with anti-LC3 (Cosmo Bio, Tokio, Japan) and anti-LAMP1 (Abcam, Cambridge, UK) primary antibodies for 1 hour at room temperature and visualized respectively by means of Cy3- and Alexafluor 488-conjugated secondary antibodies (Jackson Immunoresearch).
Coverslips were mounted in SlowFade-Anti-Fade (Invitrogen, Thermo Fisher Scientific) and examined under a confocal microscope (Leica, Nussloch, Germany). Digital images were acquired with Leica software and LC3 dots were quantified with Leica Application Suite X (LAS X) using a set of defined intensity thresholds that were applied to all images. A minimum of 50 cells per sample was counted for triplicate samples per condition in each experiment.
After image acquisition, the co-localization of LC3 and LAMP-1 signals was examined with the ImageJ software. Quantification of co-localization, expressed in terms of Mander's Overlap Coefficient, was calculated using the JacoP plugin of ImageJ software.
Statistical analysis was calculated using a two-tailed Student’s t-test for paired data. A P value < 0.05 was considered statistically significant.
|
10.1371/journal.pcbi.1002319 | Recovering Protein-Protein and Domain-Domain Interactions from Aggregation of IP-MS Proteomics of Coregulator Complexes | Coregulator proteins (CoRegs) are part of multi-protein complexes that transiently assemble with transcription factors and chromatin modifiers to regulate gene expression. In this study we analyzed data from 3,290 immuno-precipitations (IP) followed by mass spectrometry (MS) applied to human cell lines aimed at identifying CoRegs complexes. Using the semi-quantitative spectral counts, we scored binary protein-protein and domain-domain associations with several equations. Unlike previous applications, our methods scored prey-prey protein-protein interactions regardless of the baits used. We also predicted domain-domain interactions underlying predicted protein-protein interactions. The quality of predicted protein-protein and domain-domain interactions was evaluated using known binary interactions from the literature, whereas one protein-protein interaction, between STRN and CTTNBP2NL, was validated experimentally; and one domain-domain interaction, between the HEAT domain of PPP2R1A and the Pkinase domain of STK25, was validated using molecular docking simulations. The scoring schemes presented here recovered known, and predicted many new, complexes, protein-protein, and domain-domain interactions. The networks that resulted from the predictions are provided as a web-based interactive application at http://maayanlab.net/HT-IP-MS-2-PPI-DDI/.
| In response to various extracellular stimuli, protein complexes are transiently assembled within the nucleus of cells to regulate gene transcription in a context dependent manner. Here we analyzed data from 3,290 proteomics experiments that used as bait different member proteins from regulatory complexes with different antibodies. Such proteomics experiments attempt to characterize complex membership for other proteins that associate with bait proteins. However, the experiments are noisy and aggregation of the data from many pull-down experiments is computationally challenging. To this end we developed and evaluated several equations that score pair-wise interactions based on co-occurrence in different but related pull-down experiments. We compared and evaluated the scoring methods and combined them to recover known, and discover new, complexes and protein-protein interactions. We also applied the same equations to predict domain-domain interactions that might underlie the protein interactions and complex formation. As a proof of concept, we experimentally validated one predicted protein-protein interaction and one predicted domain-domain interaction using different methods. Such rich information about binary interactions between proteins and domains should advance our knowledge of transcriptional regulation by CoRegs in normal and diseased human cells.
| CoRegs are members of multi-protein complexes transiently assembled for regulation of gene expression [1]. Assembly of these complexes is affected by ligands that bind to nuclear receptors (NRs), such as steroids, retinoids, and glucocorticoids [2]–[5]. CoRegs complexes exist in many combinations that are determined by post-translational modifications (PTMs) and presence of accessory proteins [6], [7]. To date, over 300 CoRegs have been characterized in mammalian cells [8] and it has been shown that CoRegs complexes control a multitude of cellular processes, including metabolism, cell growth, homeostasis and stress responses [6], [9], [10]. Many CoRegs complexes are considered master regulators of cell differentiation during embryonic and post-developmental stages [10], [11], and evidence suggests that malfunction of these proteins can lead to the pathogenesis of endocrine-related cancers [3], [12] and diabetes [13]. Importantly, it is believed that development of better chemical modulators of CoRegs will lead to a ‘new generation’ of drugs with higher efficacy and selectivity [14], [15].
To accelerate research in the area of CoRegs signaling, the Nuclear Receptor Signaling Atlas (NURSA) [16] have been applying systematic proteomic and genomic profiling related to CoRegs [17], [18]. Recently, the NURSA consortium released a massive high-throughput (HT) IP/MS study reporting results from 3,290 related sets of proteomics pull-down experiments [19]. The results from these experiments are protein identifications with semi-quantitative spectral count measurements, which can be used to approximate protein enrichment in individual IPs. Multiple IP experiments that sample different protein complex subunits can be integrated to gain a global picture of protein complex composition [20]–[22]. Several prior studies applied to human cells have proposed strategies to reconstruct protein complexes by combining results from HT-IP/MS [23]–[28]. Some of the results from such studies have been processed by algorithms that probabilistically predict binary protein-protein interactions (PPIs). In some cases, such predictions were validated using known PPIs from the literature, where in few cases predicted interactions were further validated experimentally. For example, Washburn and colleagues implemented the multidimensional protein identification technology (MudPIT) method to pull down complexes using 27 bait proteins from the Mediator complex to suggest 557 probabilistic interactions between the baits and their pulled preys [23]. They used the Jaccard distance to integrate protein co-occurrence in the different experiments, and compared their ‘high-confidence’ interactions with those listed in a literature-based database, the human protein reference database (HPRD) [29]. Experimentally, the study validated few predicted interactions using co-IP and western blots. In a follow up study, different clustering approaches to extract sub-complexes from related affinity purification (AP)-MS experiments using three distance measures: Manhattan, Euclidian, and Correlation Coefficient for clustering are described [30].
The aforementioned work, and other similar prior studies, ranked predicted associations and provided probabilities for interactions between baits and preys, building on the explicit nature of bait-prey relationship in epitope-based purifications. However, due to secondary cross-reacting proteins, bait-prey relationships are rarely explicit in IPs carried out with primary antibodies. Hence, here we developed and compared different ways, coded into mathematical functions, to score prey-prey interactions from a large, recently published, HT-IP/MS dataset. The equations predict direct protein-protein interactions between prey proteins without considering the specific baits. We also used the same equations to predict domain-domain interactions underlying the protein-protein interactions. We evaluated the performance of these equations using known protein-protein and domain-domain interactions from the literature and validated one protein-protein interaction experimentally, and one domain-domain interaction using computational docking. By combining the data from the 3,290 IP-MS experiments collected by NURSA we predicted binary interactions between prey proteins and their domains. We offer a global view of CoRegs complexes in human cells, and provide the predicted networks for exploration on the web through a web-based application with downloadable tables freely available at http://maayanlab.net/HT-IP-MS-2-PPI-DDI/.
A detailed description of the IP-MS procedure can be found in references [19], [26] and the list of experiments in Dataset S1. The data we analyzed is provided as supporting material tables for reference [19]. These supporting tables contain GeneIDs for identified protein products, as well as the spectral count (SPC) measurements, and ‘abundance’ values, defined as SPCs/MW, where MW is the molecular weight for the largest isoform of the gene product. The latter normalization approximately accounts for the number of peptides expected from a protein. Abundance is logically similar to the normalized spectral abundance factor (NSAF) scores previously proposed [30], except the values are not scaled per experiment.
To score prey-prey interactions from the HT-IP/MS data table, containing the ranks of proteins from the 3,290 IP-MS experiments, we evaluated existing and developed new equations implemented as algorithms in MATLAB and Java.
Sørensen similarity coefficient (Sor) provides a symmetric similarity coefficient for comparing two finite sets. The coefficient ranges between 0 and 1, where 0 denotes no similarity, and 1 denotes identical sets. The Sørensen coefficient is calculated as the ratio of the cardinality of shared members between two sets and the sum of the cardinalities of the same sets.(1)The Sørenson coefficient was applied to determine the likelihood that proteins A and B directly interact. MA and MB are the sets of all experiments that reported either protein A, B or both as present in the lists of pulled prey proteins. MA,B are lists where both A and B are present.
Pearson's Correlation coefficient (Pr) characterizes the linear dependency of two variables. Here we used the Pearson's Correlation coefficient to quantify the correlation the SPC scores of two proteins across all IP/MS experiments.(2)ρA,B is the Pearson's Correlation coefficient between proteins A and B where Q denotes the reported ‘abundance’ which is SPC/MW (MW, molecular weight). and are the column vectors of Q at indices and . is the covariance and and are the standard deviations of and .
Equation 3 (E3) was developed through an intuitive manual symbolic search for functions that perform well, based on benchmarking, using known protein-protein interactions. E3 calculates a ratio between the sum of the SPC scores in experiment j () and the difference between the ranks of protein pairs based on their SPC scores in the same experiment. The average E3 scores across all experiments is the final score that is used to quantify the likelihood that two prey proteins interact. The rationale behind the E3 equation is to reward pairs of proteins that have similar SPC scores and similar ranks across all experiments, rewarding pairs of proteins with high SPC scores that appear in the same complexes.(3)
The AB correlation was also developed through an intuitive manual symbolic search for functions that perform well based on benchmarking using known protein-protein interactions. The AB correlation computes the mean of the product of SPC scores normalized by dividing by the sum of mean SPC scores across all experiments.(4)The AB method also rewards pairs of proteins that have higher SPC scores in the same subset of experiments.
To evaluate the predicted prey-prey protein interactions using the four equations, we used an updated version of the human literature-based protein-protein interactome we developed for the program Genes2Networks [31]. The PPIs are from 12 databases: HPRD [29], MINT [32], DIP [33], MIPS [34], PDZBase [35], PPID [36], BIND [37], Reactome [38], BioGRID [39], SNAVI [40], Stelzl et al. [41], and Vidal and co-workers [42]. These databases contain direct physical interactions for mouse, rat, and human proteins containing 11,438 proteins connected through 84,047 interactions extracted manually from publications. We converted all IDs to human IDs using homologene (http://www.ncbi.nlm.nih.gov/homologene).
To identify domains for proteins, we used the Pfam domain database release 24.0. The file ‘Pfam-A.full.gz’ was downloaded from: ftp://ftp.sanger.ac.uk/pub/databases/Pfam/releases/Pfam24.0/on November 1st 2010.
Domain-domain interactions (DDI) were obtained from the Domine database [43]. The Domine database contains 26,219 domain-domain interactions. Among these domain-domain interactions, 6,634 were inferred from the protein data bank (PDB) and 21,620 were computationally predicted by one or more of 13 prediction methods. In order to score domain-domain interactions, we developed a prediction vector containing a combined score for all predicted PPIs that contain domain-pairs at each side of a scored PPI. We assigned the score of the predicted PPI to the DDI score.
Antibodies for STRN, also called Striatin, are polyclonal rabbit, and were purchased from Millipore Corp. Antibodies for CTTNBP2NL were purchased from GeneTex. MCF-7 cells were lysed in immunopreciptation buffer containing Hepes (50 mM, pH 7.4), NaCl (150 mM), EDTA (1 mM), Tween-20 (0.1%), glycerol (10%) and protease inhibitors. The lysates were pre-cleared in the presence of rabbit IgG and protein A beads. The input sample was collected after pre-clearing. Samples were rotated overnight with IgG or Striatin antibody and subsequently incubated for two hours with Protein-A beads. The washed protein-containing beads were denatured and analyzed by Western blot.
The MolSoft ICM software was used to perform the domain-domain docking simulation. ICM uses a two-step method: pseudo-Brownian rigid-body docking followed by biased probability Monte Carlo minimization of the ligand side-chains, to sample conformational space in order to identify the global energy minimum for a given interaction [44]. For this specific simulation, the protein PPP2R1A (PDB ID: 1B3U), the receptor, was kept rigid, while conformations of the ligand STK25 (PDB ID: 2XIK) were sampled around the receptor and corresponding docking scores were retrieved. Domains were then examined for interactions based on these scores.
We analyzed the experimental data from 3,290 IP-MS experiments targeting 1,083 antigens (bait proteins) using 1,796 different antibodies. These experiments detected 11,485 non-redundant proteins (Dataset S1). Some of the baits were pulled-down with several different antibodies. Some of the experiments with the same baits and antibodies were repeated several times but conducted under different conditions, i.e., stimulated/un-stimulated cells, or different cell types. Complexes are mostly isolated from nuclear fractions but some experiments use cytosolic fractions. Summary of the experimental conditions, cell types, antibodies and baits used, counts of normalized peptides identified in each experiment per protein, and size of the lists of proteins identified in each experiment can be directly obtained from the primary publication provided as reference [19].
IP-MS proteomics profiling have several known experimental challenges that need to be considered when applying functional global analyses on such data. First, it is well established that the proteins identified in such experiments are enriched for highly abundant and “sticky” proteins. This results in numerous proteins appearing frequently in almost all pull-downs regardless of the cell type, cellular fraction or experimental conditions. To address this we used a list of “non-specific” proteins to filter protein identifications that appear frequently in many pull-downs (Dataset S1). For all further analyses we removed these proteins from the results. Such a “non-specific” protein list can be useful as a guideline for filtering other IP-MS proteomics data applied to human cells. However, it should be noted that the concept of filtering IP-MS proteomics data based on a “non-specific” list is only meant as a guide. The sticky non-relevant proteins may play an important biological role that would be missed by removing them. In general, proteins that appear in the list are enriched in heat shock, ribosomal, and heterogeneous nuclear ribonucleoproteins (hnRNPs). Also, the majority of proteins on the non-specific list were selected based on the purifications from nuclear extracts, so some abundant cytosolic proteins may be over represented in the protein-protein and domain-domain interaction predictions since these may not have been removed. In order to integrate and visualize the results from the 3,290 IP-MS experiments, we first used the Jaccard Distance (JD) to construct a CoRegs complex similarity graph were nodes represent protein lists from each experiment and links represent overlap between experiments (Fig. S1). Nodes and links are preserved in the network if the similarity is greater than the Jaccard distance of 0.7. This retained 491 experiments and 2233 links between them, which are a small portion of all possible experiments and their similarities (Fig. S2A). On average, pull-down experiments reported the identification of ∼30–200 proteins but the distribution has a heavy tail with few experiments identifying over 1000 proteins (Fig. S2B).
Our aim in this study is to assign confidence scores to binary prey-prey protein-protein and domain-domain interactions by integrating information from the 3,290 IP-MS experiments. The rationale for this approach is that the experiments, reporting lists of ∼30–200 proteins for each pull-down, taken together, provide enough information to reconstruct high-fidelity, small-sized complexes and potentially enough to recover direct physical interactions between pairs of proteins and domains. We reasoned that if we use all the information across all experiments to score each pair of proteins for potential direct interaction, we will be able to identify novel associations in addition to recovering known interactions better than by chance. In contrast with most prior methods that focused on scoring bait-prey interactions, our equations predict interactions between prey proteins that commonly reappear together in different pull-downs. Although the data collected for this study was aimed at the recovery of interactions between the intended antigens (baits) and other proteins, the majority of primary antibodies cross-react with multiple secondary antigens and those antigens interact with other proteins. This complicates bait-prey scoring of HT-IP/MS data. Yet, logically, if two proteins reappear together at the top of lists in many different pull-downs, we can guess that they may physically interact regardless of which baits were used to pull them down, making it possible to predict likely binary interactions by utilizing the spectral counts, not just co-occurrence. To encode such logic into mathematical functions we devised four scoring schemes, each attempting to address the problem in a slightly different way. To evaluate the performance of the four scoring schemes we used known PPIs we consolidated from online databases [31]. The overall schema for this approach is depicted in Fig. 1.
To compare the performance of the different scoring methods we visualized the results as either receiver operator curve (ROC) (Fig. S3), random walks (Fig. S4), or a sliding window (Fig. S5). Visualization of overlap between a ranked list and a gene set using a random walk was borrowed from the popular Gene-Set Enrichment Analysis method [45]. The three equations AB, E3, and Pr can be combined with the Sørenson coefficient to slightly improve the predictions by the AB and E3 equations, and significantly improve the predictions made with the Pr equation. AB and E3 perform best when combined with the Sørenson coefficient because these equations take into account the quantitative levels of the peptides, rewarding interactions that appear on top of the same pull-downs and penalizing potential interactions where the two proteins are not present in the same pull-down, or when one protein appears at the top and the other at the bottom. The different methods recover different sets of interactions and in some cases complement each other, suggesting perhaps that a combined weighted score may provide better results than using a single equation (Fig. S6, Dataset S2).
Next, we used ball-and-stick diagrams to visualize the results across all experiments. We first visualized all overlapping interactions listed in the top 10% of predicted protein-protein interactions by each method (AB, E3 and Pr combined with Sor). This resulted in a network made of 2,509 proteins (nodes) and 28,886 interactions (edges) (Fig. 2). Using Cytoscape's organic visualization algorithm, the hubs of this network self-organize into an interesting hierarchical structure that may reflect their complex formation relationship. This network provides a global view of the CoRegs interactome, allowing zoom-in to view the identity of high confidence predicted protein-protein interactions and the complexes that these interactions form. To accomplish this zoom-in view, we increased the threshold to only include interactions from the top 1% of predicted interactions by all three scoring methods and include only three-node cliques. Three-node cliques are triangles in the network topology where three proteins are connected to each other with a maximum of three links. The resultant network contains 543 proteins and 1,893 interactions organized into 63 tightly connected protein complexes containing 3 to 25 proteins (Fig. 3). Many of the interactions and complexes that emerged are already known from low-throughput protein-protein interactions studies. However, some of the complexes within this network and many of the predicted protein interactions are novel. As a proof of concept, we focused on one predicted complex where most of the members of the complex were exclusively prey proteins in all experiments, and most interactions in the complex were not previously known (Fig. 4A). The complex contains ten densely connected proteins with the protein STRN in the center, predicted to interact with all other nine members. STRN, STRN3 and STRN4 are scaffolding proteins with a calmodulin binding domain. Interestingly CTTNBP2NL has been previously reported with STRN and STRN3 in another IP/MS study [46]. To experimentally validate one of the interactions within this complex we used IP and western blotting to demonstrate a direct interaction between STRN and CTTNBP2NL which is another member of the predicted complex (Fig. 4B). We chose this interaction based on antibody availability. Our experiment clearly shows that the two proteins interact. Such a demonstration of physical interaction experimentally does not prove that our prediction method works well, but it demonstrates how predicted interactions can be further validated experimentally. To prove that the predictions are of high quality, many such experiments need to be performed with appropriate controls to show statistically that the combined equations can predict, with high fidelity, physical interactions.
Before analyzing all of the 3,290 IP-MS experiments published by Malovannaya et al [19], we had access to a subset of the data before it was published. Therefore, we developed our analysis methods on a subset of 114 IP-MS experiments that are a fraction of the entire set of the 3,290 IP-MS experiments. In order to integrate and visualize the results from these 114 IP-MS experiments, similarly to the network shown in Fig. S1, we created the Jaccard Distance (JD) CoRegs complex similarity graph (Fig. S7). Most of these initial 114 experiments used Estrogen Receptor α (ESR1) and nuclear receptor co-activator 3 (NCOA3), also called SRC3, as baits in different cellular conditions. Both proteins play an important role in breast cancer, where SRC3 serves as the main co-activator of estradiol-dependent ESR1 [47], [48]. The experiments that used ESR1 and NCOA3 as baits resulted in similar protein lists (clusters in the subnetwork in Fig. S7) compared with the other pull-downs. Using the same prediction combined scores with the three equations, with lower thresholds, we identified five distinct high confidence complexes we named: SMARC, CSTF, RCOR, MBD, and SIN3A (Fig. S8). These five complexes have been previously reported in the Corum database [49] and some have been functionally characterized (Fig. S9). Specifically, the SMARC complex highly overlaps with complex IDs 238, 714, 803, and 806 in Corum, a database of reported protein complexes [49]. The CSTF complex is listed as complex number 1147 in Corum, RCOR is listed as 626, and MBD and SIN3A have associated IDs with highly overlapping entries for complexes in Corum. The SMARC and CSTF complexes were recovered mostly from ESR1 pull-down experiments, while the other three complexes are formed by combinations of many other types of baits. Notably, the SMARC and CSTF complexes are nearly mutually exclusive to two different antibodies targeting ESR1, and are recovered in the control experiment from HeLa cells that do not express ESR1. Thus, one antibody is likely cross-reacting with a member of the SMARC complex, whereas the other antibody cross-reacts with a member of the CSTF complex (Fig. S10). This result highlights the importance of protein complex reconstruction from HT-IP/MS based on prey-prey co-occurrence alone, independently of the intended baits.
Since PPIs are often the result of interactions between the structural domains of the interacting proteins, and since we know most of those domains for all pulled prey proteins based on their amino-acid sequences, we can use the scores for PPIs to also score and rank domain-domain interactions (DDIs). The scoring of domain interactions is slightly more complex since most proteins have several different domains and the domains can appear more than once within the same protein. To resolve this we used the score for PPIs containing domains between all possible domain pairs from each side of the PPI and normalized the score across all the domains (see methods). The aggregated score for all DDIs was accumulated across and within all 3,290 IP-MS experiments. The idea of predicting DDIs from PPIs is not new [50]–[52]. DDIs can also be predicted using structural biology methods or by evolutionary conservation of sequences across organisms [53]. To evaluate which PPI scoring method works best to predict DDIs, we compared the predicted scores for DDIs with reported DDIs from the Domine database. The Domine database contains both structurally observed and computationally predicted DDIs [43]. ROC curves and random-walk plots were used to evaluate DDI predictions, similarly to how we evaluated the PPI prediction methods (Fig. S11 and S12, Dataset S3).
The plots show that we can reliably recover known and predicted DDIs. In addition to the four equations used to score PPIs, we introduced another scoring scheme, λ, for scoring DDIs. λ is an index that counts the number of times two predicted interacting prey proteins have a domain on each side of the PPI. Such an index improves DDI predictions. In addition to the type of analysis we did for PPIs, we also attempted to further combine different prediction methods to optimize DDI predictions. Finally we visualize our predicted DDIs with known DDIs as a network diagram to visually explore interactions among all domains (Fig. S13) and within the STRN centered complex identified by the PPIs predictions (Fig. 5A). To further validate one of the predicted DDIs we pursued a computational structural biology approach. We attempted to dock the PKinase domain of STK25 to the HEAT domain of PPP2R1A. We chose these two proteins because they had a crystal structure in PDB. Although the DDI is already listed in Domine, the prediction of this DDI interaction is based on sequence and homology. Hence there is no direct evidence of such interaction between these two proteins and their domains. Using the Molsoft ICM software we obtained a docking score of −46.75 kcal/mol. This score is considered high and as such confirms the interaction. By examining the confirmation of this interaction it appears that the Pkinase domain of the STK25 protein binds to the HEAT domain of PPP21RA. The energy gap of approximately 2 kcal/mol (ICM score units) between the best obtained and next consecutive docking score clearly suggests strong recognition of the HEAT domain by the Pkinase domain (Fig. 5B–D).
In this study we combined results from 3,290 experiments that identified nuclear protein complexes in human cells using IP-MS. We implemented and evaluated four different equations assessing their ability to predict direct physical PPIs from the aggregated proteomics data using known PPIs from the literature. The highest scoring predictions were visualized as networks with many densely connected clusters that are likely made of real protein complexes. The prediction scores for potential interactions could be considered as surrogates to real affinity constants. However, since we do not know the exact quantities of proteins, it is not possible to compute exact dissociation constants. Such binding constants can be useful for dynamical simulations where we could stochastically trace the transient dynamics of CoRegs complex formation in-silico. Scoring PPIs by only using the prey measurements may become more robust as more IP-MS experiments are published. However, careful attention should be given to weighting the repetitiveness of experiments so interactions from similar pull-downs, if repeated, are not mistakenly given higher scores. Regardless of possible limitations, the ability to recover direct PPIs based on such a massive dataset is an important step toward utilizing HT/IP-MS datasets for reconstructing networks and generating hypotheses. In addition, we show that the equations can be extended to predict interactions between structural domains. We also demonstrated two ways to further validate predicted PPIs and DDIs, using experimental and computational approaches. In summary, our analyses explored new methodologies for scoring PPIs and DDIs using data from related IP-MS experiments, providing many hypotheses about mammalian CoRegs complexes formation, and allowing users to explore novel complexes, PPIs and DDIs online at http://maayanlab.net/HT-IP-MS-2-PPI-DDI/. This resource can help us advance the catalogue of transcriptional regulation by CoRegs in normal and diseased mammalian cells.
|
10.1371/journal.pcbi.1003514 | Probabilistic Approach to Predicting Substrate Specificity of Methyltransferases | We present a general probabilistic framework for predicting the substrate specificity of enzymes. We designed this approach to be easily applicable to different organisms and enzymes. Therefore, our predictive models do not rely on species-specific properties and use mostly sequence-derived data. Maximum Likelihood optimization is used to fine-tune model parameters and the Akaike Information Criterion is employed to overcome the issue of correlated variables. As a proof-of-principle, we apply our approach to predicting general substrate specificity of yeast methyltransferases (MTases). As input, we use several physico-chemical and biological properties of MTases: structural fold, isoelectric point, expression pattern and cellular localization. Our method accurately predicts whether a yeast MTase methylates a protein, RNA or another molecule. Among our experimentally tested predictions, 89% were confirmed, including the surprising prediction that YOR021C is the first known MTase with a SPOUT fold that methylates a substrate other than RNA (protein). Our approach not only allows for highly accurate prediction of functional specificity of MTases, but also provides insight into general rules governing MTase substrate specificity.
| Our approach is easily applicable to different organisms, because it does not rely on species-specific properties and uses mostly sequence-derived and other readily available data (e.g. isoelectric point or predicted structural fold). Tests on yeast MTases indicate that the accuracy of our predictions is ∼90%. We show that knowledge of substrate binding sites or corresponding motifs is not crucial for highly accurate general substrate specificity predictions of enzymes, and provide new insights into how such specificities are achieved at the molecular level. We predict substrate specificities not yet observed for a given class of enzymes, and experimentally verify our predictions.
| This is a PLOS Computational Biology Methods article.
Prediction of protein function from its sequence is an important goal of bioinformatics [1], [2], since the function of many proteins remains unknown, including more than 50% of human proteins. Because of its importance, a large-scale community-based Critical Assessment of Protein Function Annotation (CAFA) experiment is held biannually [3], to objectively evaluate and compare different methods and stimulate research in this area. One of the most difficult cases of protein function prediction is that of enzyme substrate specificity, which is essential for understanding its role in cellular processes. Even if the exact 3D structure of an enzyme is known, its substrate specificity is often not clear, as it depends on both local (e.g. active site) and global (e.g. protein structure) properties [2], [4], [5].
Many approaches have been proposed to predict enzyme substrate specificity. One example, applied to type II restriction endonucleases (REases), relied on the observation that connectivity of the secondary structures in the αβα structural core correlates with the angles between the secondary structure elements and the cleavage patterns of the REases [6]. Prediction of optimal substrate peptides (encompassing the phosphorylation site) for protein kinases was done taking only the amino acid sequence of a kinase as input [7]. Analysis of available crystal structures, molecular modeling, and sequence analyses of kinases and substrates led to extraction of a set of rules governing the substrate specificity of protein serine/threonine kinases. The method was used to analyze yeast cell cycle control and DNA damage checkpoint pathways. Combined genomic and functional context was recently used in Zhang et al. [8] to assign function of homologous proteins from the carbohydrate FGGY kinase family. However, homology alone is not sufficient to successfully predict protein substrate specificity [4], [9].
Several bioinformatics approaches have been applied to predict substrate specificity of yeast MTases. An attempt to infer the substrate of methylation from a hidden Markov model profile clustering analysis, applied to Saccharomyces cerevisiae Rossmann-like fold methyltransferases, revealed some grouping of MTases that correlated with their substrate specificity [10]. However, this method is limited and not capable of predicting substrate specificity for all studied proteins. In Wlodarski et al. [11] we proposed that fold, pI, temporal expression pattern and protein localization contribute to determining MTase substrate specificity.
The prediction methods discussed above typically rely on complex heuristics and in some cases require a detailed 3D structure of the protein, or are applicable only to some of the studied enzymes. Here, we propose a very general framework based on fundamental laws of probability that is applicable to all considered proteins (even in cases of missing data) and does not require any specific data type (e.g. known 3D structure, conserved sequence motifs). Moreover, our method is capable of correctly predicting substrate specificity from a combination of properties not yet observed among known enzymes. Our method has a much higher percentage of successful predictions (84–89%) than previous approaches and is not limited to a certain group of MTases [10], [11]. Since our approach is general and relies on features that are sequence-derived and not organism-specific, it should be easily applicable to other organisms and enzyme classes.
As proof-of-principle, our approach is employed to predict general substrate specificity of yeast MTases. MTases are present in all living organisms and involved in many important cellular processes such as signal transduction, transcriptional control, biosynthesis and metabolism [12]. MTases comprise a large and highly diverse group of enzymes that transfer a methyl group from a donor (typically S-Adenosyl-L-Methionine, SAM) to an acceptor (MTase substrate) [13]. In S. cerevisiae, there are 86 MTases and their substrates are either proteins, RNAs or other molecules, (DNA is not enzymatically methylated [14]) [11]. As a training set, we used 61 S. cerevisiae MTases with experimentally confirmed substrate specificity (known MTases) (Table S2) and predicted substrate specificities for 25 putative S. cerevisiae MTases with unknown substrate specificity (putative MTases). After our predictions were made, the substrate specificities of 9 MTases were confirmed experimentally, with results consistent with predictions in 89% (8 out of 9) of the cases.
We propose a mathematical framework for inferring substrate specificities from the physico-chemical and biological properties of MTases. The advantage of our method is that it yields very accurate substrate specificity predictions and explicitly provides the probabilities that a given MTase methylates a substrate from each class (RNA, protein or other molecule). The method consists of three stages. First, we estimate conditional probability for each substrate specificity based on a single property. Second, the final probability is computed based on several selected properties. The single-property probabilities are combined as described in Materials And Methods. The high number of available enzyme properties leads to a very large combinatorial space of probabilistic models for predicting the substrate specificity. To limit the search for the best model, we selected the 22 most informative properties as defined by the likelihood of the respective single property models on the training set (Table S1). For numerical variables, we chose either continuous or binned representation, as well as optimal number of bins.
The final model is selected based on optimization of up to 14 parameters (Table S2) and evaluation of 86,000 models. Since the number of properties and range of parameters considered did not allow for an exhaustive search in the model space, we optimized continuous properties using the Powell method (Text S1. Supplementary text) [15]. Because we were comparing models with different numbers of parameters, the likelihood criterion would not be appropriate. Likelihood, which describes the goodness of fit, is always increased if more variables are added to the best performing model with a given number of parameters. Therefore, to compare models with differing numbers of parameters, we instead used the Akaike Information Criterion (AIC) [16], which balances the goodness of the fit (likelihood) with informativity of the parameters. The AIC naturally selects models using the most informative sets of parameters and rejects those with highly correlated parameters. This is important in our case, as we prefer to use parameters with clear biological or physico-chemical interpretation, which in general are not mutually independent.
The probability of substrate specificity for an MTase with a certain property is given by the Bayes Theorem:(1)where P(substratei) is the probability of an MTase to methylate substrate type i (i.e. protein, RNA or other molecule) and P(property) is its probability to have a certain property (e.g. structural fold, isoelectric point (pI), expression pattern and cellular localization). P(substratei|property) is the probability that an MTase will methylate substrate type i if this MTase has the given property. P(property|substratei) is the probability of an MTase having a certain property if it methylates substrate type i and P(substratei) is the a priori probability of substrate type i. A property can be either categorical (e.g. fold) or numerical (e.g. pI, expression onset), and in either case, a range of different predictive models can be constructed. To select the best single-property model, we apply the Maximum Likelihood (ML) method to optimize P(substratei|property) on the training set (i.e. substrate specificities of the known MTases). The training set consists of 61 S. cerevisiae MTases with experimentally confirmed substrate specificity (known MTases) (Table S2). Among them 26 methylate RNAs, 24 methylate proteins and 11 methylate other molecules.
Preliminary selection of biophysical, cellular and functional properties of MTases to use in our model was based on our previous research [11], which indicated that protein isoelectric point (pI), structural fold, expression pattern, expression onset and cellular localization are all correlated with MTase substrate specificity (Fig. 1). We performed preliminary studies to determine which specific properties have the highest predictive power. Specifically, we interrogated similar properties to find out which among them are most significantly enriched in MTases sharing the same general substrate specificity. For example, we examined different data on protein cellular localization including both predicted localizations [17], which are available for all proteins, and experimentally derived data on protein localizations. We concluded that for yeast MTase substrate specificity predictions the most useful data are Gene Ontology protein localizations limited to IDA and IEA evidence codes and additionally grouped into superclusters of localizations. Similarly, we grouped structural folds into superclusters (Fig. 2A and B). The predictors selected for use in the models are discussed in detail below.
As shown in our previous study [11], yeast MTases may adopt up to nine different folds (predicted with high confidence from sequence similarity) within their catalytic domains: Rossmann-like, SPOUT, SET domain, TIM beta/alpha-barrel, transmembrane, tetrapyrrole methylase, DNA/RNA-binding 3-helical bundle, SSo0622-like and thymidylate synthetase. For predictions, we divided them into four groups based on the frequency of a particular fold being assumed by MTases and correlation with their substrate specificity preference: Rossmann-like, SPOUT, SET domain and “other”. The “other folds” category was motivated by few yeast MTases assuming them and their shared preference for “other” substrate specificity (Fig. 2A). In contrast, all eight known MTases with a SET fold methylate proteins, and all four known MTases with a SPOUT fold methylate RNA (Fig. 2A). The Rossmann-like fold MTase group has more diverse substrate specificities and comprises 62% of known MTases. About two-thirds of MTases in the “other folds” category methylate other substrates.
We observed that for known MTases, substrate specificity correlates with cellular GO localizations [18], especially for the nucleolus, nucleus and mitochondrion localizations (Fig. 2B). Moreover, the original number of GO localization terms were clearly too big in comparison with the number of known MTases. Therefore, we decided to describe MTase cellular localization by four mutually exclusive terms: (i) nucleolus, (ii) nucleus and not nucleolus, (iii) mitochondrion and not nucleus, and (iv) other. All known yeast MTases localized in the nucleolus have RNA as a substrate. MTases with ‘nucleus and not in nucleolus’ localization most often methylate proteins (50%) or RNA (41%); only two methylate other substrates. Among known MTases within the ‘mitochondrion and not nucleus’ category there is only one example of a protein MTase. The remaining twenty three known protein MTases are not localized in the mitochondria. Moreover, MTases that methylate other substrates constitute 50% of those in the ‘mitochondrion and not nucleus’ group.
As we pointed out in [11], for known MTases, global pI values correlate with their substrate specificity. Since the isoelectric point is a proxy for protein charge, we can expect proteins with a high pI to bind negatively charged molecules like RNA. Indeed, 67% of known MTases with pI≥6.5 methylate RNA. On the other hand, 65% of known MTases with a low pI<6.5 methylate proteins. MTases that methylate other substrates have a medium-range pI (Fig. 2D).
We also searched for regions with very high or low pI values, expecting that such regions of a protein might correspond to substrate binding regions or domains. For automatic identification of such regions, we computed the maximum and minimum local pI values for each sliding window size (from 15 to 185 a.a.) and for each MTase, and referred to them as pI max and pI min, respectively.
The YMC is a redox cycle lasting 300 minutes, in which genes with similar functions tend to be expressed within a specific temporal window [19]. Expression profiles of genes periodically expressed in the YMC can be grouped into three main clusters: Ox (oxidative), R/C (reductive/charging) and R/B (reductive/building) [19]. Nineteen known MTases belong to the Ox cluster, among them ten methylate RNAs and seven methylate proteins. Two-thirds of known MTases from the R/C cluster methylate other substrates, and most of those from the R/B cluster (5 out of 8) methylate proteins (Fig. 2C). To describe expression patterns, in addition to YMC expression clusters, we also used the onset of individual YMC gene expression [20]. More than half of known MTases (30 of 52) have similar YMC expression onsets around the beginning of the YMC cycle (between 280 min and 16 min). However, all but one known MTase that methylate other substrates and have assigned the onset of expression [20], have expression onsets after 16 min and before 280 min (Fig. 2E) (only genes periodically regulated during YMC, as determined by [21] have their onsets of expression assigned).
The ML method was used to select a model most likely to reproduce the observed data (i.e. general substrate specificities of the known MTases) and then AIC penalty for number was parameters was applied. The model best scoring after the AIC correction is hereafter referred to as the “best” model. The properties, along with their parameterization, log likelihood and AIC values are listed in Table S1. Surprisingly, the best scoring single property is the isoelectric point (pI). The best model with a single pI threshold had a pI threshold of 6.97. The model using this single property can correctly predict substrate specificity for 67% of the known MTases, giving even better results than the single property model based on structural fold. The second best scoring property is pI max (calculated using 125 a.a. sliding window) with a threshold of 9.85. However, we do not expect the 125 a.a. to be a biophysically important fragment size, because when two thresholds for pI max are allowed, the best fragment size is much bigger (170 a.a.). The pI max model correctly predicts 42 out of 61 proteins (69%). The third best scoring property is the protein fold, single-property model using fold correctly predicts 66% of known MTases. (The models are ranked not according to the number of correct prediction, which is not a smooth measure and is subject to Poissonian noise, but according to their AIC value, which is log likelihood with the penalty for the number of parameters).
The prior probability of having a given substrate specificity, P(substratei), that we used in our models was the fraction of known MTases with that specific substrate type. When we made predictions using prior probabilities alone, they were correct in only 43% of cases, while for the best single property model, they were correct in 67% of cases. We also verified that allowing a different P(substratei) than that observed among known MTases does not improve the outcome: optimization over different prior probabilities converges to values observed among known MTases.
We compared our approach with a simple homology method of substrate specificity inference from a well annotated protein sharing the highest sequence similarity. Such prediction from the most similar known MTase of the same catalytic fold (the closest paralog) in S. cerevisiae gave 61% correct predictions for known MTases. This shows that in our case sequence similarity, contrary to popular belief, is not the most informative property for predicting MTase substrate specificity within a single organism, as even close homologs can have different general substrate specificities. For example, MTases PPM1 and PPM2 display ∼30% sequence identity, but methylate different types of substrates: PPM1 methylates a protein while PPM2 methylates an RNA.
We studied predictive models using several properties at a time, assuming their independence (Eq. 2 Methods). In practice, the properties included are typically correlated. To address this issue, we first used the ML method to optimize parameter values for every family of the models considered (i.e. for any different combination of properties) to maximize the accuracy of predictions in known cases. Naturally, models with a higher number of parameters will produce more accurate predictions. Therefore, we used AIC for model selection to ensure that the model with the most informative properties, as opposed to the model using the most properties, would be chosen as our best model. Finally, such a chosen model (best model) was used to predict substrate specificities for 25 putative S. cerevisiae MTases with unknown substrate specificity (putative MTases).
We evaluated 86000 multi-property models dependent on up to 14 properties (Table S3 and S4). The best-scoring model uses the following properties: pI, SET fold, other folds and R/C expression cluster (Table S3). The pI property employs a single threshold of 6.95. Other properties, SET fold, other folds and R/C expression cluster, are binary properties; an MTase can either have this property or not. The pI property distinguishes known MTases that methylate RNA from those that methylate proteins, while the SET fold property indicates known MTases with protein substrate specificity. Analogously, the “other folds” property correlates with “other” substrate specificity. Detection of known MTases with other substrate specificity is additionally supported by including an R/C expression cluster category, which is employed by the top five models (Table S3). The sixth best model does not use any property derived from the expression data, but it does use localization (mitochondrion) and pI (with single threshold of 6.96), SET fold and “other folds” properties. The best model using neither localization or expression data utilizes pI (with single threshold of 6.97), SET fold and “other folds” properties. This model scores 35th in terms of best AIC and correctly predicts substrate specificities of 79% of known MTases (48 out of 61).
The best model correctly predicts substrate specificity for 83.6% of known MTases (in 51 out of 61 MTases the highest scoring substrate class coincided with the actual substrate class) (Table S5). We computed the statistical significance of obtaining 51 out of 61 correct MTase substrate specificity predictions with the null hypothesis that predictions are random. We then applied very conservative Bonferroni correction considering 86,000 alternative models for multiple hypothesis testing and obtained a very statistically significant p-value, p = 7.2×10−9, even though our search space for the best model was not restricted to the most promising candidates. This result shows that our method is capable of yielding final models with very high predictive power.
Moreover, the probabilities associated with the best scoring substrate specificity are significantly higher when the prediction is right than when it is not (p = 0.01, t-test, Fig. S1). Taken together, the overall very high-accuracy of our predictions (>83%) combined with the statistically significant correlation between correctness of our prediction and the likelihood we assign to predicted substrate specificities validates our approach and justifies the selection of classes of input parameters for our models (Fig. 1).
We succeeded in predicting substrate specificity for 88.5% (23 of 26) of RNA MTases, 70.8% (17 of 24) of protein MTases and 100% of 11 MTases that methylate other substrates. Among the MTases whose substrate specificity was not predicted correctly, four (YDL200C, YDR410C, YDR440W, YNL063W) were predicted to methylate RNA and three (YDR435C, YLR137W, YLR172C) to have other substrate specificity while they actually methylate proteins. For the last five of those MTases correct substrate specificity predictions have the second-highest probabilities. Namely, they are predicted to be protein MTases with the following probabilities: YLR172C (37%), YLR137W (36%), YDR435C (33%) and YDR440W and YNL063W (14%). Thus, known MTases methylating proteins appear to be the most difficult to predict, likely due to vast functional differences within the ‘protein’ class of substrates. On the other hand, we predicted three MTases (YDL112W, YOL141W, YOR239W) to have protein substrate specificity when in fact they are RNA MTases. Below we discuss in detail the reasons for incorrect predictions in these difficult cases: (i) ABP140 (YOR239W) has extraordinary low pI compared with other known MTases that methylate RNA; (ii) PPM1 (YDR435C) and PPM2 (YOL141W) are close homologs that methylate protein and RNA, respectively. However, they both modify the same chemical group: oxygen from a carbonyl group. Specifically, PPM1 methylates the C-terminal of protein phosphatase 2A [22], in turn PPM2 is involved in the methoxycarbonylation required for synthesis of wybutosine, an atypical nucleoside of tRNAPhe [23]. They have very similar pIs that are below our 6.95 threshold. Low pI is more typical for the known protein MTases, therefore PPM2 is predicted to methylate protein. Additionally, PPM1 is in the R/C expression cluster of the YMC, which outweighs its prediction towards methylating another substrate; (iii) MTQ1 (YNL063W), MGT1 (YDL200C), DOT1 (YDR440W), STE14 (YDR410C) are MTases that methylate proteins and are predicted to have RNA substrate specificity as they all have high pI (above 6.95 threshold). MGT1 is not a typical protein MTase because it transfers a methyl group from DNA to itself (DNA demethylation). The nucleic acid is not methylated, as predicted, but is actually a substrate in the reaction and the high positive charge of the MTase supports its binding. DOT1 is a Rossmann-like fold MTase specific for histones. We noticed a tendency for histone MTases to have relatively high pI (although it was not incorporated into our models due to there being only four histone MTases present in yeast). Specifically, SET1 and SET2 both methylate histones and also have a high pI, like DOT1 MTase. However, the model predicts them correctly as protein MTases because they have a SET fold. (iv) DPH5 (YLR172C) and YLR137W are protein MTases incorrectly predicted to methylate other substrate types. DPH5 has a tetrapyrrole methylase fold that is in the “other” folds category and YLR137W is in the R/C expression cluster. These properties overweigh prediction for those MTases to have other substrate specificity; (v) TRM3 (YDL112W) is an RNA MTase that is incorrectly predicted to methylate protein because of its low pI.
According to our best model, 13 out of 25 putative MTases methylate RNAs, ten methylate proteins and two methylate other substrates (Fig. 3). Among 18 putative MTases with a Rossmann-like fold, five are predicted to methylate proteins, two to methylate other substrates and eleven to methylate RNA. As expected, all four putative MTases with a SET fold (YHR207C, YPL165C, YJL105W and YKR029C) are predicted to methylate proteins. Our model predicts two out the three putative MTases with a SPOUT fold (YGR283C, YMR310C) to methylate RNA. Surprisingly, our model also predicts that a third putative MTase with a SPOUT fold, YOR021C, is the first known example of a SPOUT methylase in any organism to methylate a substrate other than RNA [24].
To validate our approach for general substrate specificity prediction we performed protein methylation assays for selected putative yeast MTases. We used this approach successfully in the past to identify two yeast protein MTases: YBR271W and YLR285W (NNT1) [11]. Briefly, we incubated purified recombinant proteins with total cell extracts from the wild-type yeast and respective knockout strains in the presence of tritium-labeled AdoMet ([3H] AdoMet). The reaction products were analyzed by SDS-PAGE followed by autoradiography. HMT1 (a protein MTase) and TRM4 (an RNA MTase) were used as positive and negative controls, respectively. As expected, for control reactions we observed protein methylation patterns matching known substrates for HMT1, but not for RNA MTase TRM4 (the smear at the bottom of the gels in TRM4 lane corresponds to tRNA substrate).
First, we focused on our most unexpected prediction that YOR021C is the first ever known SPOUT MTase to methylate protein (Table S2). Indeed, in the in vitro assay, we observed the presence of protein methylation products for YOR021C. YOR021C seems to methylate at least 2 proteins (∼20 and 30 kDa) detected only when the deletion strain was used (Fig. 4), which strongly suggests that these modifications are specific and stable. The same results were obtained when total RNA was removed from cell extracts using RNaseA. Combined, these data indicate that YOR021C is a protein MTase. Very recently another group independently confirmed our findings by showing that this MTase methylates a small ribosomal subunit protein Rps3 [25], with molecular weight 26.5 kDa, consistent with one of our observed methylation products. In contrast, for SPOUT MTases YGR283C and YMR310C, which we predict to methylate their usual substrate, RNA, no protein methylation was found (Fig. S2).
We also tested protein methylation for selected Rossmann-like fold MTases with unknown substrate specificity: YNL092W, YDR316W (OMS1), YIL096C and YKL155C (RSM22). An in vitro MTase activity assay suggests that YNL092W is a protein MTase. For this MTase we detected on tritium screen a methylated product corresponding to the molecular weight of YNL092W. Moreover, methylated product was also detected when purified recombinant protein was incubated only with [3H] AdoMet (Fig. 4), indicating that YNL092W methylates itself (since no other protein substrate was present). Interestingly, this seems to be the second yeast protein, after MGT1, capable of automethylation. For the remaining Rossmann-like MTases: YDR316W (OMS1), YIL096C and YKL155C (RSM22), predicted to methylate RNA, we did not observe any protein methylation (Fig. S2), supporting their predicted substrate specificity.
Our prediction that YHR209W (CRG1) methylates substrates from the “other substrate” category, has been recently confirmed by Lissina et al. [26], who showed it methylates canthardin. Another of our predictions, that YHR207C (SET5) methylates protein, has also been recently confirmed showing it to methylate histone H4 [27].
In the year 2012 CAFA experiment, F-measure (a harmonic mean between precision and recall) was used to compare performance of different models [3]. The best scoring CAFA model (Jones-UCL group) achieved F-measure of 0.6 for predictions of molecular function, while our classifier has an F-measure of 0.84. The fact that our focused method performs so much better than the best general predictor is very reassuring, although not surprising. Constructing narrower predictors allows for selecting features most relevant to the properties being predicted, and if executed well, should result in much better predictions than from predictors aiming to predict more general molecular function categories.
The framework presented in this paper can be readily applied to other biological systems and questions. Below is a discussion of the most promising areas of application requiring only minor adaptations of the approach.
How many substrate categories can be successfully predicted is a difficult question to answer without specific knowledge of the system to be studied. It depends not only on the number of known examples, but also on the distribution of properties of interest. In our experience, the number of reliably predictable categories approximates the square root of the size of the training set. Clearly, predicting fewer classes yields a higher accuracy of inference. Moreover, it is also important to choose prediction categories such that they have comparable number of known examples and no single predicted category includes very few members. It is also highly desirable that variance within categories should be limited. In a given case, the feasible number of categories can be determined empirically, by verifying, as we did, if predictions are statistically significant as compared with random predictions. In the case of yeast MTases, they were highly significant for predicting general substrate specificity (protein, RNA, other), but as expected not significant for predicting more detailed substrate specificity (histone protein, ribosomal protein, other protein, rRNA, tRNA, other RNA, lipid, small molecule, other), where number of categories exceeds the square root of number of known examples, our rule of thumb for maximal number of predictable categories.
In summary, our predictions proved to be very accurate, yielding an 84% correct prediction rate when tested on a set of MTases with known substrate specificity. After our predictions were made, substrate specificities of 9 MTases were fully or partially confirmed experimentally by us or others [26], [27], [29], with results consistent with predictions in 89% (8 out of 9) of the cases. Our work also aids in understanding how observed general substrate specificities are achieved at the molecular level. For instance we show that, surprisingly, a global biophysical property, pI, impacts MTase substrate specificity more than structural fold. Likely, pI, which closely correlates with protein charge, retains such an impact on substrate specificity because it often determines whether an MTase will bind negatively charged molecules such as RNA, or typically positively charged protein substrates. We also show that knowledge of a substrate binding site or corresponding motifs, traditionally thought to be crucial, is not essential for highly accurate general substrate specificity predictions for yeast MTases.
Our models combine inference from many sources to estimate the probabilities of given MTases having various substrate specificities. Unlike previously used classification schemes [8], [10], [11], this approach allows us to predict substrate specificity not yet observed for a given class of MTases. Indeed, we made one such prediction: that YOR021C, a SPOUT fold MTase, methylates a protein. That prediction was very surprising, as all SPOUT MTases known to date, both in yeast and other organisms, exclusively methylate RNAs. Strikingly, at the time of publication of this paper, this prediction has been confirmed both by us and independently by another group in a newly published paper [29].
In summary, we have shown that our general probabilistic framework based on fundamental laws of probability and information theory is a powerful tool to predict substrate specificity of yeast MTases. Biological expertise is still very important in our approach, but it is used only to select the initial properties plausibly related to the intended prediction; otherwise the proposed approach is completely objective and self-learning. Moreover, our model can be easily updated with new knowledge by repeating the same calculations on the updated data set. To ensure that our work is broadly applicable, as input to our model we prioritized organism-independent properties, especially ones that can be derived from sequence data alone. Therefore, our approach is also applicable to MTases in other organisms and with modifications can be used to predict the substrate specificities of other enzymes, as we discussed in the examples given above. As in the recent CAFA experiment, we conclude that the best predictions are obtained from integration of varied data types. Accuracy of our predictions, as measured by F-measure employed by CAFA, is much better than that of the best CAFA predictor. This underscores our belief that a successful classifier designed to predict more narrow functional categories should always outperform more general predictors. Given that accuracy of protein function prediction is crucial for its usefulness, more focused predictions, of the type we present, will always be needed. In the future, most successful general function predictors may employ predictors like ours for predicting function subcategories.
For each MTase, we calculate the probability that it has a given substrate specificity (e.g. RNA, protein or other molecule) based on its properties (Eq. 2):(2)For two different properties, for simplicity we assumed that they are independent. Specifically, the following equation was used:(3)where n is the number of substrate specificities.
P(property|substratei) was calculated in different ways depending on whether the property is of the categorical or continuous type. (i) For categorical variables (e.g. localization, expression cluster), we estimated probabilities P(property|substratei) for the whole population of MTases based on the sample of known MTases (Text S1. Supplementary text). (ii) For continuous variables (i.e. pI, expression onset), after dividing them into several intervals and estimating population values of P(property|substratei) as in (i), we modeled them as a smoothed step function with two to three steps (specified by chosen thresholds) (Text S1. Supplementary text).
We tested 86,000 different combinations of up to 14 property types (Text S1. Supplementary text) by calculating likelihood of prediction for MTases with known substrate specificity. The best model was selected based on the lowest value of AIC, with AIC = 2k−2ln(L), where k is the number of parameters in the model and L is the maximized value of the likelihood function for the estimated model [16].
Types of properties used in our model: structural fold, pI, expression pattern and cellular localization (Fig. 1 and Table S4), were selected based on our expert knowledge of which protein properties are relevant to MTase substrate specificity. Multiple properties belonging to these four broad categories were screened based on the statistical significance of their correlation with MTase substrate specificity. Supplementary table (Table S2) lists all S. cerevisiae MTases together with considered properties.
For comparison, we also predicted MTase substrate specificity using inference of substrate type from the closest paralog. Specifically, we assigned each yeast MTase a substrate specificity of an MTase with the same structural fold of catalytic domain and with the highest sequence similarity. To detect the closest yeast homolog, we used Meta-BASIC [30], a sensitive tool for recognition of distant similarity between proteins based on alignments of sequence profiles enriched with predicted secondary structure (meta profiles).
The following yeast strains (Euroscarf) were used in this study: BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0), BY4741 ΔYBL024W (ΔTRM4), BY4741 ΔYBR034C (ΔHMT1), BY4741 ΔYGR283C, BY4741 ΔYIL096C, BY4741 ΔYKL155C (ΔRSM22), BY4741 ΔYNL092W, BY4741 ΔYDR316W (ΔOMS1) BY4741 ΔYMR310C, BY4741 ΔYOR021C and BY4741 ΔYMR310C. The standard yeast genetic methods and selective growth media were used, as described in Rose et al. [31].
The following proteins: YBL024W (TRM4), YBR034C (HMT1), YGR283C, YIL096C, YKL155C (RSM22), YNL092W, YDR316W (OMS1), YOR021C and YMR310C were produced in E. coli (BL21-CodonPlus-RIL strain) as N-terminal HIStagSUMO tag fusions using LB medium and overnight IPTG inductions at 23°C. The bacterial pellets were lysed by sonication in buffer A (20 mM Tris-HCl pH 8.0, 200 mM NaCl, 10 mM imidazole, 10 mM 2-mercaptoethanol) and purified on His-Trap FF Crude columns (GE Healthcare). The proteins were further purified by size-exclusion chromatography on a Superdex 75 10/300 GL column (GE Healthcare) in buffer containing 10 mM Tris-HCl pH 8.0 and 150 mM NaCl. Finally, glycerol was added to the protein aliquotes (10% final concentration), which were then stored at −80°C. The purity and quantity of the proteins were assessed by SDS-PAGE.
Yeast whole-cell extracts were prepared as previously described [32]. Recombinant proteins (5–15 µg) were incubated with 30 µg of native yeast extract (from wild-type and respective knockout strains) in the presence of [3H] AdoMet (0.5 µCi/reaction) in 20 µl of reaction buffer (10 mM HEPES pH 8.0, 2 mM EDTA, 50 mM KCl, 1 mM DTT). Reactions were incubated at room temperature for 1 hr, diluted 2-fold in Laemmli buffer and resolved on a 12% SDS-PAGE gel. The gel was stained with Coomassie blue, dried and exposed overnight to tritium screen.
|
10.1371/journal.ppat.1001293 | Genital Tract Sequestration of SIV following Acute Infection | We characterized the evolution of simian immunodeficiency virus (SIV) in the male genital tract by examining blood- and semen-associated virus from experimentally and sham vaccinated rhesus monkeys during primary infection. At the time of peak virus replication, SIV sequences were intermixed between the blood and semen supporting a scenario of high-level virus “spillover” into the male genital tract. However, at the time of virus set point, compartmentalization was apparent in 4 of 7 evaluated monkeys, likely as a consequence of restricted virus gene flow between anatomic compartments after the resolution of primary viremia. These findings suggest that SIV replication in the male genital tract evolves to compartmentalization after peak viremia resolves.
| Methods to reduce the transmission of HIV-1 are hindered by a lack of information regarding early viral dynamics and evolution in the male genital tract. In the present study, we show that SIV in the blood and genital tract are homogeneous during early infection, indicating facile virus gene flow between these compartments. Importantly, the coincidence of the resolution of primary viremia with the decreased virus levels in genital secretions suggest that the dramatic fall in virus replication during early infection underlies the development of viral compartmentalization. Our demonstration of early virus compartmentalization in the male genital tract has important implications for the understanding of early events leading to infection of the male genital tract and the nature of the transmitted virus during primary retrovirus infection.
| An understanding of HIV-1 biology in the male genital tract will be central to understanding the transmission of this virus. Since HIV-1 is transmitted predominantly by sexual contact [1], [2], [3], semen represents an important source of transmitted virus. While the transmission risk of HIV-1 has been associated with virus RNA levels in the peripheral blood of infected transmitting individuals, virus RNA levels in the blood only serve as a surrogate for the level of virus in ejaculate [1], [3], [4], [5], [6]. Indeed, during primary infection when HIV-1 transmission is high, virus levels in both the blood and the seminal plasma are both elevated [1], [7]. Yet, during chronic infection blood and semen levels of HIV-1 can be discordant [1], [7].
HIV-1 exists as a diverse population of related genetic variants that segregate into subpopulations within anatomic compartments of infected individuals [8], [9]. The intra-host compartmentalization of these variants has been documented in multiple anatomic regions including the peripheral blood, lungs, central nervous system, breast milk, gut and the female genital tract, [1], [10], [11], [12], [13], [14], [15], [16].
Importantly, the male genital tract serves as a reservoir for HIV-1 that has been shown by some investigators to harbor virus that is distinct at the sequence level from virus found in the blood [12], [15], [17], [18], [19]. It has been suggested that this compartmentalization may be a consequence of differences between the male genital tract and peripheral tissues in both immunological and target cell properties [14], [15], [20], [21], [22], [23], [24], [25], [26], [27], [28]. Histopathologic studies of genital tissues from SIV-infected monkeys have confirmed that infection can occur at these sites [24], [29]. Male genital tract-derived viral variants have been reported to have unique phenotypes with regard to drug resistance [11], [30], [31], [32], [33] and cellular tropism [34], [35]. Some groups have reported that HIV-1 in the male genital tract [1], [36], [37], [38] is similar to that in the blood [39], while other groups have suggested that there is an anatomic sequestration of virus in some individuals during primary infection [10], [34], [35], [40], [41]. Overall, the mechanisms that contribute to compartmentalized virus in the male genital tract are not entirely clear. Virus compartmentalization could be the result of founder effects, differential immune selection, altered tropism or restricted gene flow between resident viral populations [42], [43].
In the present study we sought to define the dynamics of SIV compartmentalization in semen during primary infection in rhesus monkeys. We accomplished this by analyzing SIV envelope (env) sequences in the blood and semen at the peak and set point of SIV replication in rhesus monkeys. We observed a striking progression toward increasing viral compartmentalization over time following infection, raising the possibility that the abrupt reduction in virus replication that occurred in these monkeys after the resolution of peak viremia may contribute to a restriction in viral gene flow in and out of their genital tracts.
Twenty-six rhesus monkeys were vaccinated and then monitored for 16 weeks after experimental SIV infection as previously described [44]. Thirteen were vaccinated using Gag-Pol immunogens with a heterologous prime/boost regimen that included both plasmid DNA and replication-defective recombinant adenovirus serotype 5 (Ad5) immunogens. The remaining 13 control monkeys received sham vaccine constructs. Animals were challenged by the i.v. route using a SIVmac251 stock with a previously reported sequence diversity [45].
To determine whether virus in the semen and peripheral blood were compartmentalized following infection, we assessed the phylogenetic relationship between semen- and blood-derived SIV env sequences. Single genome amplification (SGA) was used to generate 789 env amplicons for sequencing and subsequent phylogenetic analysis. Sequence data were generated from 2 time points post-challenge, at week 2 (peak virus replication) and week 16 (virus set point).
Viral RNA levels were comparable for each group, and we detected no significant differences in week-2 and week-16 blood or seminal plasma viral loads between vaccinees and controls. Similarly, no significant differences were detected between animals included in this study and animals excluded as a result of insufficient sequence data (Wilcoxon P>0.01) as described [44].
SIV env sequences were first generated from both the blood and semen of 14 rhesus monkeys (7 vaccinated and 7 control) from specimens obtained 2 weeks following infection. The genetic diversity of the virus from each anatomic compartment of each monkey was calculated as the average pair-wise genetic distance of SIV env within each semen or blood population from the inoculum consensus. Data from the vaccinated and control animals were analyzed independently. Since these represented analyses of sequence from virus obtained very early after infection, little divergence was seen from the founder viruses that initiated infection in each monkey (Table 1). Moreover, we observed little genetic diversity between sequences of blood and semen amplicons at this early time point after challenge, representative examples from 4 monkeys (2 control, 2 vaccinated) are shown (Fig. 1). Among the sequences from the blood, we observed no difference in average pairwise env distance from the inoculum consensus per animal between vaccinees (0.29% versus 0.28% respectively, P = 0.9). The sequence divergence for all week 2 semen isolates were not significantly different in control and vaccinated monkeys, with a mean distance among control animals of 0.27% and 0.15% among vaccinated monkeys (P = 0.4). These differences are consistent with the lower levels of viral replication in the vaccinated group [44].
We next analyzed blood and semen sequences from specimens obtained at week 16 following infection. Post set point, we were able to obtain sufficient numbers of concurrent semen- and blood-derived SIV sequences from 7 monkeys to allow a phylogenetic analysis of these viruses. Consistent with the lower level of virus replication in the vaccinated group, we observed reduced diversification of sequences obtained from the vaccinated monkeys, particularly in the seminal plasma. Specifically, blood-derived env sequences from control animals showed a trend towards greater divergence (0.45%) than env sequences derived from the blood plasma of vaccinees (0.35%, P = 0.2). In week 16 sequences obtained from seminal plasma, the mean pair wise distance per animal was 0.31% and 0.22% for controls and vaccinees, respectively (P = 0.2). We next analyzed these sequence data using available algorithms to determine if changes in N-linked Env glycosylation sites had occurred during the sampling period [46]. No significant changes in potential Env glycosylation sites were observed.
We then analyzed the sequences from the inoculum and week 16 samples for evidence of selection on SIV env by fixed-effects likelihood (FEL) [47]. As the results of this test can be influenced by the presence of recombinant sequences, we first used G.A.R.D. [48] to test for recombination and, if significant breakpoints (p<0.01) resulted, performed FEL tests separately for sequence alignments partitioned at the breakpoints.
The presence of negative selection could not be determined as intra-host viral populations had not diversified sufficiently for such events to occur at detectable levels. The positive selection tests that reached significance (p<0.05) are indicated (Table S1). Among the sequences from inoculum and week 16 sequences, we found 12 codons in 6 monkeys (3 vaccinees and 3 from the control group) under positive selection. One of these codons was under positive selection in 2 different animals (codon 427 in AX93 and AY47).
A number of investigators have postulated that specific amino acid signatures may be associated with HIV found in the semen [17], [49], [50]. To investigate this possibility, we applied a tree-corrected contingency table method to identify compartment specific amino acid signatures [51]. This analysis yielded no evidence for specific signatures that might be associated with SIV tropism in the semen up to 16 weeks following infection.
We then conducted additional signature analyses using only compartmentalized animals, but a sample size of 4 compartmentalized animals would lack sufficient statistical power to identify robust predictive signatures of compartmentalization. Therefore, we adapted a related strategy used by Pillai et al. [49]. Pillai reported a combination of HIV ENV residues (HXB2 270, 291, 387, and 464) associated with blood-semen compartmentalization. All of these except the first are located 1-2 residues downstream of putative N-linked glycosylation sites. We evaluated these sites in the SIV env sequences generated in this study by identifying homologous sites in HIV-1 and SIV from an alignment that incorporates HIV-1, HIV-2, HXB2, and SMM239 Seqs. (http://www.hiv.lanl.gov/content/sequence/NEWALIGN/align.html).
Reviewing the homologous signature sites in the SIV env sequences, we found 3 of 4 to be located 1–2 residues downstream of glycosylation sites. However, mutations in these sites (and in the neighboring 4 residues) were equally common in sequences derived from blood and semen, regardless of whether we included only sequences from compartmentalized animals. Thus, we were unable to confirm semen signatures in these SIV data.
We then sought to determine if SIV compartmentalization was demonstrable during primary infection. To detect compartmentalization in contemporaneous blood- and semen-derived SIV env sequences, we applied both phylogenetic analyses and the Slatkin-Maddison (SM) test from the HYPHY software package (v0.99b) [52], [53]. At the peak of SIV viremia, similar env sequences were observed in the control and vaccinated animals. Importantly, we observed the mixing of env taxa from blood- and seminal plasma-derived virus, and the association of these sequences into closely related phylograms (Fig. 1). Analysis of the week 2 blood and seminal virus sequences indicated that all 14 monkeys are similar on the basis of Neighbor-Joining (NJ) tree topologies. Importantly, we observed clusters of intermixed monotypic blood and semen sequences with minimal genetic divergence from the challenge SIV env sequences, termed “clusters of identity” in all animals, although the amount varied between individual monkeys (Fig. 1).
Representative NJ phylogenetic trees generated from SIV sequences from control monkeys CK7J and CM8X (Figs. 1 A, B), and from vaccinated monkeys AY47 and AX93 (Figs. 1 C, D) are shown. In control monkey CK7J, we observed one cluster of closely related semen and blood-borne sequences in the phylogenetic tree. In control monkey CM8X, two clusters of intermixed blood/semen sequences are apparent in the tree structure. Similar results were observed in the Gag-Pol vaccinated monkeys. For example, in vaccinated monkey AY47 a single phylogenetic cluster of interspersed semen and blood env sequences are apparent in the tree. Monkey AX93, has three clusters of identical, monotypic semen and blood sequences (Fig. 1 C and D).
While we did not observe topological evidence for virus compartmentalization between the semen and blood during early infection in any of the 14 monkeys analyzed, compartmentalization effects were also assessed using the SM test. Applying this metric, we observed no evidence of virus sequestration in week 2 blood- or semen-derived sequences in 13 of the 14 monkeys (Table 2). Control monkey AX89 had compartmentalized semen env sequences (s = 14, P<0.0001) (Fig. 2). However, the presence of large numbers of identical monotypic sequences might bias statistical measures of compartmentalization [54], [55]. To ensure that the large numbers of isogenic sequences in the semen of AX89 were not confounding this statistical analysis, we conducted a secondary analysis that collapsed all equivalently rooted monotypic sequences to a single representative env taxon. We then applied the same SM test to the normalized set of env sequences from monkey AX89. This modified analysis also indicated significant compartmentalization between SIV env in the blood and the male genital tract of monkey AX89 (s = 14, P = 0.0007). Therefore, there was weak evidence for virus compartmentalization in only 1 of the 14 monkeys by 2 weeks post-infection.
We next explored the source of virus found in the male genital tract at the time of peak viremia. By day 14 following infection, significant levels of cell-free virus RNA and cell-associated provirus are present in the semen of both control and vaccinated animals (Fig. 3A) [44]. Examination of SIV env sequences from seminal plasma and semen-associated cells from monkey CK7J using NJ- tree parameters (Fig. 3B) indicated that several clusters of monotypic cell-free virus were identical in sequence to the semen provirus. Moreover, several clusters of cell-free and cell-associated virus in the genital tract of this monkey were phylogenetically indistinguishable from SIV env sequences derived cell-free and cell-associated virus from the blood. These data argue that in early SIV infection, during the period of peak viremia, there is significant virus gene flow between anatomic compartments. There are also significant levels of cell-associated and cell-free sources of virus present early in infection that may constitute a source of transmissible virus.
By week 16 following infection, we could readily detect virus compartmentalization by phylogenetic inference in 4 of the 7 monkeys that had sufficient numbers of sequences for analysis (Fig. 4). Compartmentalization is evident in these trees as independent clades consisting of sequences from one a single sample source, i.e. blood or semen. Included as a Supplement, are trees that combine sequencing data from both compartments at both acute and chronic time points (Fig. S1–S10 in Text S1). Typically, we observed that week 16 semen sequences tend to not be associated with those from week 2, rather they are evident as independent clades.
We also conducted an analysis, as described above, to ensure that isogenic sequences were not confounding the statistical analyses. This modified analysis did not significantly alter the findings as determined by tree topology. We also applied the SM test to phylogenies obtained from excluding duplicate sequences (Table 3) and found that, after correction for multiple testing, two animals, C179 and AX89, had marginally significant SM values (denoted Pn).
Our inability to demonstrate definitive compartmentalization of virus using these strategies led us to consider whether other biologic events might be obscuring compartmentalization. Therefore, we evaluated the possibility that recombination may confound the SM analysis. To assess recombination effects we applied the G.A.R.D. algorithm [48], [56] and found significant levels of recombination (P<0.001) within SIV env, (data not shown). Next, we analyzed our data using a “segmented” SM-test based on env recombination breakpoints (not shown) and found specific regions of the SIV envelope from RMs C179 and AX89 were in fact significantly compartmentalized (P<0.0002, Fig. 5).
We next determined the relative levels of virus migration between the blood and semen by computing the minimum number of SM migration events for each inferred phylogenetic tree. These migrations occur where an internal node exhibits a change of sample source (blood or semen) among immediate descendents of that node. When these migration points were mapped directly onto phylogenies, we observed a concordance of semen compartmentalization with the number of migration events between these anatomic compartments, in individual monkeys. For example, the frequent migration events between the blood and male genital tract, as shown for monkeys C171, and DA2D (Fig. 6A and B), are consistent with unimpeded gene flow resulting in homogenous co-evolving SIV env populations. Restricted migration, as shown for monkey AY89 and AX93 (Fig. 6C and D) is consistent with the independent evolution of virus replicating in the blood and male genital tract. Therefore, we have documented examples of both unimpeded and restricted gene flow with the latter manifesting itself as virus compartmentalization.
We have therefore shown there was no evidence of SIV compartmentalization between the blood and semen in 13 of 14 monkeys analyzed at the peak of virus replication during primary infection. However, SIV compartmentalization was evident in 4 of the 7 analyzed monkeys by the time of virus set point.
To explore the mechanisms responsible for this changing pattern of viral evolution in the male genital tract, we evaluated 2 virologic correlates associated with compartmentalization. First, we observed that cell-associated virus levels in the male genital tract during peak infection are diminished significantly in the animals by week 8 following infection (Fig. 7A). Second, cell-free virus levels in the semen are also significantly diminished between weeks 2 and 16 following infection (Fig. 7B). In fact, these cell-free virus data demonstrate a threshold effect whereby viral RNA is only detectable in the seminal plasma when virus levels in the blood plasma exceed 104 RNA copies/ml. The coincidence of the resolution of primary viremia with the decrease in levels of virus in genital secretions suggest that the dramatic fall in virus replication, with the accompanying restriction of gene flow, during this early period of infection underlies the development of viral compartmentalization.
While HIV-1 transmission in humans occurs most frequently through mucosal exposure, the present study was undertaken using rhesus monkeys infected by intravenous inoculation of SIV using artificial means to collect longitudinal semen samples. Although infection by this virus and route therefore does not model mucosal HIV-1 acquisition, the present findings have important ramifications for the understanding of early events leading to infection of the male genital tract, and semen as a source of transmitted virus. Low-risk sexual transmission of HIV-1 usually results in an infection initiated by a very limited number of founder viruses [45], [57], [58]. These viruses very rapidly appear in the blood after the transmission event, presumably as a consequence of the larger number of target cells in the blood than in the genital tract. In the periphery, HIV or SIV rapidly replicates to high levels and any subsequent virus diversification is likely a consequence of differences in virus replication in these compartments, independent of how infection was initiated.
During peak SIV replication we found little evidence for genetic segregation of virus between the semen and blood of infected rhesus monkeys. Instead, we observed an equilibration of SIV between these compartments manifested as closely related clusters of intermixed virus sequences derived from the semen or blood. As we applied SGA technology to generate all sequences [59], [60], the large numbers of identical sequences observed at peak virus replication are not likely to be a methodologic artifact. Moreover, since the SIV population size in the semen during peak virus replication was high, error as a result of re-sampling seems unlikely [61].
Based on the phylogenetic relationship among virus in the semen at peak replication, significant amounts of virus appear to be derived as a result of clonal expansion of infected cells. The homeostatic proliferation of SIV infected cells trafficking from the blood into genital tissues is a plausible means of producing the oligoclonal “bursts” of virus observed in ejaculate. These findings account for the magnitude and delay in peak semen viremia relative to the blood that has been reported in acutely HIV-1 infected patients [7], [36], [44]. Moreover, the influx of large numbers of SIV-infected cells into genital tissues make it unlikely that a specific virus amino acid “signature” could be a determinant of early viral tropism in the male genital tract [17], consistent with our inability to identify SIV Env amino acids that were significantly enriched in the seminal plasma. However, our analysis did not span a sufficient period of time to determine the presence of semen signatures during chronic infection [49], [50].
Mechanistically, the high-level of virus present in the blood during early infection results in a “wave” or spillover of cell-free and cell-associated virus into the male genital tract, equilibrating virus between the two compartments. Host control of SIV replication results in a receding “tide” of virus as virologic set point is achieved. When peak virus replication is reduced to levels below the threshold required for intra-compartment gene flow, free virus or infected cells may be effectively “trapped” within the genital tissues at “low tide.” Continued replication of virus in the male genital tract, in the absence (or reduced rate) of inter-compartment gene flow would then account for the localized diversification of virus variants in the genital tissues.
In fact, we observed less env diversification in the genital tract than in the blood of these monkeys. This finding is consistent with the lower level of SIV replication in the genital tract than in the blood at 16 weeks following infection. Moreover, the difference in virus diversification in these two anatomic compartments was particularly evident in the vaccinated monkeys after challenge (i.e., 3 of the 4 compartmentalized monkeys), consistent with the low-level virus replication in the male genital tract after set point [44]. Vaccine-elicited immune responses and alterations in target cells in the genital tract may have also contributed to the observed effects.
Although the male genital tract is generally considered to be an immune privileged compartment, inflammatory conditions present during acute SIV infection may facilitate the infiltration of both free virus and infected cells. Over time, differences between the blood and the genital tract, as a result of restrictions in gene flow would foster local replication in the genital tract leading to the compartmentalization observed at set point. Inflammation is a likely cause for the trafficking of cells from the peripheral circulation into the genital compartment, and may explain the observations of intermittent viral shedding in the semen of chronically HIV-infected men as a result of bacterial or viral infection of the male genital tract [34], [62], [63], [64], [65]. We did not determine if genital tract infections were present in any of the animals in this study.
These apparent differences in selective pressure between the blood and male genital tract should also be considered in light of the reduced effective SIV population size in the genital tract as virus set point is achieved. Virus populations of reduced size and altered genotype relative to the blood may contribute to the described genetic bottleneck associated with HIV-1 transmission, highlighted by several investigators [58], [66], [67]. This bottleneck is manifested by the acquisition of HIV-1 variants in newly infected individuals that are not highly represented in the peripheral blood of the already infected, transmitting partner. The findings in the present study, despite limitations in experimental infection route and virus dose, suggest an alternative explanation for these observations. Thus, the early compartmentalization of virus in the male genital tract, vis-à-vis restrictions in virus gene flow, may account for the transmission of distinct viral quasispecies that are not represented in the blood of the already-infected individual at the time of transmission.
All Indian-origin rhesus monkeys used in this study and analysis were maintained according to the guidelines of the NIH Guide to the Care and Use of Laboratory Animals and the approval of the Institutional Animal Care and Use Committee of Harvard Medical School and the National Institute of Health. All monkeys were housed in a facility fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC). All procedures were conducted in strict accordance with the Guide for Care and Use of Laboratory Animals, and approved by the Institutional Animal Care and Use Committee (IACUC) of Harvard University.
This study monitored 26 adult Mamu-A*01 negative male rhesus monkeys (Macaca mullata) that were 4–6 years of age. All animals were confirmed as Mamu-A*01 negative by both PCR and MHC-amplicon sequencing. All animals were screened and found to be negative for simian immunodeficiency virus (SIV) and simian retrovirus type D (SRV-1) prior to the initiation of this study. All experiments were conducted in accordance with IACUC standards. Prior to, and routinely after SIV infection, all animals were monitored by physical exam, CBC with differential, and T cell subset analysis. The monkeys were housed under Biosafety level-2+ conditions.
Rhesus monkeys were immunized using a heterologous prime/boost regimen that included both plasmid DNA and replication-defective recombinant adenovirus serotype 5 immunogens. One group of 13 monkeys received these vaccine vectors expressing SIVmac239 gag, and pol, and a second group of 13 monkeys received sham vaccine constructs. The monkeys were challenged by the intravenous route with a pathogenic SIVmac251 quasispecies, 16 weeks after the last immunization. The experimental challenge consisted of injection with a 1 ml bolus of SIVmac251 (2.11×105 viral RNA copies), directly into the saphenous vein.
Viral levels were monitored in the semen and blood for 16 weeks after infection. Specimens were collected weekly for the first 6 weeks; then bi-weekly for the study duration. Semen was collected by eletroejaculation via rectal probe electrostimulation as, described [68]. Briefly, stimulation is provided by rhythmic pulsation from the stimulator, initially to 5 volts, with the voltage increasing over a 2 second period, with the stimulus maintained for 5–15 seconds and being reduced over 2 seconds. This rhythm is repeated with the maximum voltage increasing in 3-volt steps to a maximum of 20 volts. Stimulation can be repeated. Seminal emission is collected into a sterile container. Semen was centrifuged immediately after collection at 4°C and immediately frozen at −80°C. Semen cell pellets were washed 3× in cold PBS followed by the addition of 250 µl of RNAlater (Ambion, Austin, TX). Samples were immediately frozen and maintained at −80°C until use.
Viral RNA levels were monitored in the seminal plasma and semen-associated cell fraction for 16 weeks after infection. Viral RNA was routinely isolated from 200 µL of cell-free, clarified peripheral blood or seminal plasma using the NucliSENS Isolation Kit (Biomerieux, Lyon, France) following the manufacturer's protocol.
The SIV RNA standard was transcribed from the pSP72 vector containing the first 731 bp of the SIVmac239-Gag gene using the Megascript T7 kit (Ambion Inc.). RNA was isolated by phenol-chloroform purification followed by ethanol precipitation. All purified RNA preparations were quantified by optical density. RNA quality was determined by Agilent bioanalyzer RNA chip (Agilent Inc., Santa Clara CA).
QRT-PCR was conducted in a 2-step process. First, RNA was reverse transcribed in parallel with an SIV-gag RNA standard using the gene-specific primer sGag-R 5′CACTAGGTGTCTCTGCACTATCTGTTTTG-3′under the following conditions: the 50 µL reactions containing 1× buffer (250 mM Tris-HCL pH 8.3, 375 mM KCl, 15 mM MgCl2), 0.25 µM primer, 0.5 mM dNTPs (Roche), 5 mM dTT, 500 U Superscript III RT (Invitrogen, Carlsbad, CA), 100 U RnaseOUT (Invitrogen), and 10 µL of sample. RT conditions were 1 hour at 50°C, 1 hour at 55°C and 15 minutes at 70°C. All samples were then treated with RNAse H (Stratagene) for 20 minutes at 37°C. All real-time PCR reactions used EZ RT PCR Core Reagents (Applied Biosystems, Foster City, CA) following the manufacturer's suggested instructions under the following conditions: the 50µL reactions containing 1× buffer (250 mM Bicine, 575 mM potassium acetate, 0.05 mM EDTA, 300 nM Passive Reference 1, 40% (w/v) glycerol, pH 8.2, .3 mM each of dATP, dCTP, dGTP, .6 mM dUTP, 3 mM Mn (OAc)2, .5 U uracil N-glycosylase, 5 U rTth DNA Polymerase, .4 uM of each primer, and 10µL of sample template. Following 2 minutes at 50°C, the polymerase was activated for 10 minutes at 95°C, and then cycling proceeded at 15 seconds at 95°C and 1 minute at 60°C for fifty cycles. Primer sequences were adapted from those described by Cline et al [69], forward primer s-Gag-F: 5′-GTCTGCGTCATCTGGTGCATTC-3′, reverse primer s-Gag-R: 5′-CACTAGGTGTCTCTGCACTATCTGTTTTG-3′, and the probe s-Gag-P: 5′-CTTCCTCAGTGTGTTTCACTTTCTCTTCTGCG-3′, linked to Fam and BHQ (Invitrogen, Carlsbad, CA). All reactions were carried out on a 7300 ABI Real-Time PCR system (Applied Biosystems) in triplicate according to the manufacturer's protocols.
Preliminary experiments were done to evaluate the quantitation of blood, and seminal plasma viral RNA levels and their correlation with known viral RNA quantities spiked into blood or semen sampled from SIV-uninfected monkeys (data not shown). To determine the linear range of the assay, RNA extraction efficiency and potential inhibition of reverse transcriptase activity, known quantities of serially diluted SIVmac251 viral stock were spiked into samples of peripheral blood, or semen that were recovered from three SIV-naïve monkeys. No significant differences in extraction efficiency were observed between the blood and samples derived from other compartments (data not shown). Furthermore, using the extraction methods described above, we observed no inhibition of RT activity or diminution of input RNA as a consequence of RNAse activity present in any of the virus-spiked specimens (data not shown). These findings indicated, therefore, that it was appropriate to employ this technique to isolate viral RNA from study samples of peripheral blood and semen from infected rhesus monkeys. The analytical sensitivity of our quantitative RT-PCR assay was determined as described [44], [70].
Cellular DNA was isolated using a QIAamp DNA Mini kit (QIAGEN). The number of SIV gag DNA copies was calculated as described previously [71], [72] and is shown as log-transformed copies per 1×105 semen-associated cells. All target sequences were normalized to Albumen levels in genomic DNA in multiplex reaction performed in duplicate (Vic-labelled Taqman reagents, Applied Biosystems).
Viral cDNA was diluted in 96-well plates to yield fewer than 30% wells positive for amplification to ensure that positive amplifications were a result of a single cDNA (Palmer et al., 2005). First-round PCR was carried out in a reaction mixture containing: 1× buffer (Platinum Taq HF Kit, Invitrogen), 0.2 mM dNTP mix, 2mM Mg2SO4, 0.2 µM primer OF6207 (5′ – GGGTAGTGGAGGTTCTGGAAG – 3′), 0.2 µM primer OR9608 (5′ – CTCATCTGATACATTTACGGGG – 3′), 0.025 units of Platinum Taq High Fidelity polymerase in a total volume of 20 µL. PCR mixtures were loaded into MicroAmp Optical 96-Well Reaction Plates (Applied Biosciences). PCR conditions were programmed as follows: 5 minutes at 94°C, 35 cycles of 15 seconds at 94°C, 30 seconds at 52°C, 4 minutes and 15 seconds at 68°C, followed by a final extension time of 10 minutes at 68°C [59]. For the second-round PCR, 2 µL of first-round PCR product was mixed with 1× buffer (Platinum Taq HF Kit, Invitrogen), 0.2 mM dNTP mix, 2mM Mg2SO4, 0.3 µM primer IF6428 (5′ – CGTGCTATAACACATGCTATTG – 3′), 0.3 µM primer IR9351 (5′ – CCCTACCAAGTCATCATCTTC – 3′), 0.025 units of Platinum Taq High Fidelity polymerase in a total volume of 20 µL. PCR conditions were: 5 minutes at 94°C, 45 cycles of 15 seconds at 94°C, 30 seconds at 51°C, 3 minutes and 30 seconds at 68°C, followed by a final extension time of 10 minutes at 68°C. Amplicons from cDNA dilutions resulting in less than 30% positive wells were sequenced at the Dana-Farber/Harvard Cancer Center DNA Resource Core.
Raw cDNA sequence data was assembled using GeneCodes Sequencher 4.8 DNA sequencing software. All assembled sequence contigs were manually corrected for individual ambiguous nucleotide errors and further quality controlled to exclude any amplicons derived from multiple templates according to that described by Learn et al, [73]. Nucleotide alignments were made using the GeneCutter algorithm as described below (http://www.hiv.lanl.gov/content/sequence/GENE_CUTTER/cutter.html).
The median and range of diversity was determined for each monkey, for both blood and semen, using a strategy suggested by Gilbert et al., that controls for the correlation of inter-sequence distances from the same animal and compared env divergence between animals receiving vaccine or sham-vaccinated controls. We modified the method described by Gilbert et al., to include a correction for small sample size bias (Giorgi and Bhattacharya, unpublished data), and restricted our analysis to animals those that had more than 3 sequences [74]. A nucleotide alignment of all sequences was generated with using GeneCutter, and then manually refined to maintain intact codons. The aligned sequences were trimmed to 2499 nucleotides, a region spanning the start codon of env through the first 17 nt of nef, to yield consistent phylogenetic signal among sequences, rather than variable signal due to 3′ gaps in some sequences but not others.
We used both Neighbor-Joining (NJ) and maximum likelihood (ML) for phylogenetic inference. The trees in figures 1,3,4,5, and 6 are from NJ via APE, version 2.5–3 [75] and BIONJ [76] with uncorrected pairwise distances and nodes with over 70% support from 1000 bootstrap replicates are labeled. The trees in the Supplementary data section were also obtained using Neighbor-Joining, APE and BIONJ with uncorrected pairwise distances and nodes with over 70% support from 100 bootstrap replicates are labeled. Use of an uncorrected substitution model is justified for representing within-host evolution because multiple mutations at the same site are unlikely to have occurred within 25% of the genome by 16 weeks post-infection. We inferred trees for the SM compartmentalization tests by PhyML [77] using the HKY85 substitution model, a discrete 4-parameter approximation to a gamma distribution of rate variation, and a term for invariant sites. Model parameters were also obtained via ML. We independently recomputed all trees in PAUP*, version 4.0b10 using ML substitution model parameters and NJ trees. Aside from minor differences in branch lengths and a tendency for NJ to yield negative lengths on some short branches, we did not find that results differed whether likelihood or distance-based tree optimality criteria were employed, regardless of the software implementation.
Signature detection utilized a phylogenetic correction to avoid over-counting amino acid substitutions among lineages of related sequences. To do this, we used methods described previously [78], and constructed a two-by-two contingency table for each amino-acid site in the alignment, populating the table with counts based on the immediate ancestor for each leaf in the tree. Columns in the contingency table correspond to whether or not the amino acid matches the ancestral state and rows correspond to whether the sequence is from blood or semen. Fisher's exact test identified significant deviations from equal proportions of amino-acid substitutions in blood versus semen-derived env sequences, with a false-discovery rate correction applied to the resulting p-values to accommodate multiple tests.
To test for compartmentalization in contemporaneous blood- and semen-derived SIV env sequences, we used the Slatkin-Maddison test [52] as implemented in the HYPHY software package (v0.99b) as described by Pond et al. [53]. The SM test computes s, the observed number of migration events required to yield the observed distribution of blood and semen sequences on a tree. It then assesses significance by comparing s with a re-sampled null distribution, which is obtained by holding the tree topology constant and randomly permuting the order of blood and semen labels across all leaves on the tree, computing s for each permutation. Statistical support for compartmentalization is attained when a large proportion of permutations yield more migration events than for the empirical s. That is, if fewer than 5% of 10,000 replicate permutations have at most s migration events, then P<0.05.
Analysis for recombination was performed by the visual inspection of Highlighter plots and confirmed by using the G.A.R.D. analysis, where significance was assessed at (P<0.001) (13, 14).
Protein-coding regions for all Env were tested for specific codons under positive selection using FEL. A Neighbor-Joining tree and nucleotide substitution model were computed for each aligned gene before FEL analysis [47]. Sites with greater nonsynonymous than synonymous substitution rates (dN>dS) and P<0.05 were taken as significant for positive selection. FEL results were not corrected for multiple testing, as according to Pond and Frost.
Statistical analysis was conducted using commercially available software GraphPad Prism version 4.0b (GraphPad Software, Inc. San Diego, CA 92130 USA). Linear regressions were conducted to determine the significance of correlations. Significance of differences between groups was determined using Kruskall–Wallis ANOVA as appropriate. Probability values of P<0.05 were interpreted as significant.
|
10.1371/journal.pbio.2005056 | Death and population dynamics affect mutation rate estimates and evolvability under stress in bacteria | The stress-induced mutagenesis hypothesis postulates that in response to stress, bacteria increase their genome-wide mutation rate, in turn increasing the chances that a descendant is able to better withstand the stress. This has implications for antibiotic treatment: exposure to subinhibitory doses of antibiotics has been reported to increase bacterial mutation rates and thus probably the rate at which resistance mutations appear and lead to treatment failure. More generally, the hypothesis posits that stress increases evolvability (the ability of a population to generate adaptive genetic diversity) and thus accelerates evolution. Measuring mutation rates under stress, however, is problematic, because existing methods assume there is no death. Yet subinhibitory stress levels may induce a substantial death rate. Death events need to be compensated by extra replication to reach a given population size, thus providing more opportunities to acquire mutations. We show that ignoring death leads to a systematic overestimation of mutation rates under stress. We developed a system based on plasmid segregation that allows us to measure death and division rates simultaneously in bacterial populations. Using this system, we found that a substantial death rate occurs at the tested subinhibitory concentrations previously reported to increase mutation rate. Taking this death rate into account lowers and sometimes removes the signal for stress-induced mutagenesis. Moreover, even when antibiotics increase mutation rate, we show that subinhibitory treatments do not increase genetic diversity and evolvability, again because of effects of the antibiotics on population dynamics. We conclude that antibiotic-induced mutagenesis is overestimated because of death and that understanding evolvability under stress requires accounting for the effects of stress on population dynamics as much as on mutation rate. Our goal here is dual: we show that population dynamics and, in particular, the numbers of cell divisions are crucial but neglected parameters in the evolvability of a population, and we provide experimental and computational tools and methods to study evolvability under stress, leading to a reassessment of the magnitude and significance of the stress-induced mutagenesis paradigm.
| The effect of environmental stress on bacterial mutagenesis has been a paradigm-shift discovery. Recent developments include evidence that various antibiotics increase mutation rates in bacteria when used at subinhibitory concentrations. It is therefore suggested that such treatments promote resistance evolution because they increase the generation of genetic variation on which natural selection can act. However, existing methods to compute mutation rate neglect the effect of stress on death and population dynamics. Developing new experimental and computational tools, we find that taking death into account significantly lowers the signal for stress-induced mutagenesis. Moreover, we show that treatments that increase mutation rate do not always lead to increased genetic diversity, which questions the standard paradigm of increased evolvability under stress.
| One of the most puzzling and controversial microbial evolution experiments of the 20th century may be the one performed by Cairns and colleagues [1,2] in which lac− cells are plated on lactose as the sole carbon source and therefore cannot grow. Revertants toward the lac+ genotype continuously appear after plating at a rate and timing seemingly incompatible with the Darwinian hypothesis of selection of preexisting mutants. In the lac− construct, the lacZ coding sequence is present but nonfunctional, because it is out of frame with the start codon. The lac+ revertants are thus frameshift mutants in which this coding sequence is back in frame with the start codon. Most of the controversy initially came from the question of whether these reversion mutations were Lamarckian, in the sense that they would arise at a higher rate when the cells would “sense” that these mutations would be beneficial [3]. However, many additional experiments quickly suggested that this phenomenon can be explained by more standard Darwinian mechanisms, in which genetic changes are not targeted but occur randomly and are then selected or not. While two seemingly conflicting molecular explanations—the stress-induced mutagenesis model and the gene amplification model—emerged, both are conceptually very similar.
In both explanations, mutations occur randomly and independently of their effect on fitness, but the specific conditions of carbon starvation increase the rate at which genetic diversity is generated at the relevant locus (lacI-lacZ sequence). In the stress-induced mutagenesis model [4], the genome-wide mutation rate is increased as an effect of the stress response triggered by starvation. In the gene amplification model [5,6], random duplications of the lacI-lacZ system happen before plating on lactose and are then selected in presence of lactose because the frameshift mutation is leaky. A small amount of Beta-galactosidase is still synthesized, permitting cryptic growth due to rare expression errors, which compensate the frameshift. This residual expression becomes higher with more copies of the leaky system. As the copy number of the system increases, a reversion mutation in lacI-lacZ becomes more likely because of increased target number.
While it is still not clear whether stress-induced mutagenesis is the sole explanation of the phenomenon, the attempts to explain the data presented by Cairns and colleagues have led to a much better understanding of control over mutation rate in response to the environment. An increase in mutation rate under starvation has also been reported in other systems, such as nutrient-limited liquid cultures [7,8] and “aging” colonies on agar plates [9,10]. However, Wrande and colleagues [11] reported that the accumulation of mutants in aging colonies observed by Bjedov and colleagues [10] can be explained by growth under selection without elevated mutagenesis, because the mutation used by the original authors to infer mutation rate is beneficial in these specific environmental conditions.
An emblematic molecular mechanism permitting this regulation of mutation rate is the SOS response (suggested in 1970 [12,13]), in which DNA damage is sensed by bacterial cells and leads to the up-regulation of many genes, permitting mutagenic repair and replication of damaged DNA. While the responsible enzymes were unknown at the time, it has indeed been found subsequently that the SOS response increases the dosage of polIV and polV [14,15]. These error-prone polymerases are able to replicate damaged DNA that the classical DNA polymerase polIII could not replicate, albeit at the price of a higher error rate [16,17]. This strategy, favoring “survival at the price of the mutation,” is only one side of the story. There is a line of evidence suggesting that this higher error rate is not only an unavoidable trade-off with survival. It is also supposed to be a selected property to increase mutation rate under stressful conditions, increasing the chances that one of the descendants obtains a beneficial mutation that makes it able to better withstand the stress [18,19].
The evolution of traits that increase mutation rate under stress needs be considered in the context of second-order selection [20]. Second-order selection relies on the idea that natural selection does not only act on the individual's phenotype and instant fitness but also on its ability to generate fit descendants, leading to selection of properties such as evolvability and mutational robustness [21]. In parallel to the study of environmental control over the mutation rate, genetic determinants of mutation rate have also been studied. It has been shown and is widely accepted that alleles increasing mutation rate (for example, defective mismatch repair or DNA proofreading) can be selected when hitchhiking with the beneficial mutations they permit to generate [22,23]. On the other hand, the possibility of selection of mechanisms increasing mutation rate under stress but not constitutively has been subject to a more philosophical debate [24,25]. While modeling shows such selection is possible [19], it is hard to distinguish whether an observed increase in mutation rate under a specific stress is (i) an evolvability strategy; (ii) an unavoidable trade-off of selection for survival, such as replicating damaged DNA to avoid death at the price of making mutations; or (iii) a direct effect of the stress and not of the stress-response system [26].
This debate is important for a full understanding of the evolutionary relevance of the phenomenon but does not affect the medical implications concerning the risk of de novo evolution of resistance during antimicrobial treatment [27]. Here, we are interested in the general case of mutation rate in growing stressed populations, and we especially focus on antibiotic stress, although our findings may be valid for other biotic and abiotic stresses. It has been suggested that treatment with subinhibitory doses of antibiotics increases bacterial mutation rate, due to induction of various stress-response pathways [28–32]. Many molecular mechanisms underlying this stress response have been elucidated, including the SOS response [29] or the RpoS regulon [30]. Oxidative damage has also been suggested to play a role in antibiotic-induced mutagenesis [28] and death [33–35]. Although still controversial [36], these findings link antibiotic stress to the older question of how bacteria deal with oxidative stress and how oxidative damage impacts mutation rates [37].
However, all the evidence for stress-induced mutagenesis relies on accurately measuring mutation rates of bacteria growing in stressful conditions, and comparing them to those of the same strains growing without stress. Computing such mutation rates under stress is harder than it may seem, because stress may change population dynamics and may thus invalidate the assumptions made by the mathematical models used to compute mutation rate. For example, in the case of subinhibitory concentrations of antibiotics in which net population growth is positive, death may nevertheless happen at a considerable rate. Death events, however, are not detected by standard microbiology methods and are not taken into account by the mathematical tools used to compute mutation rate [38–40].
Indeed, such tools take as inputs only the number of observed mutants at a chosen locus and the final population size, making the underlying assumption that there is no death and that population size is thus sufficient information to summarize growth dynamics. The final population size is used to infer the number of DNA divisions leading to the final observed population from a small initial inoculum. If there is death, more divisions are needed to reach this population size, thus giving more opportunities to acquire mutations. The mutation rate will then be overestimated, because the number of DNA replications will be underestimated.
In this work, we developed an experimental system to compute death rates in populations growing under stress and a computational method to compute mutation rates from fluctuation assays under stress using the computed death rates. We applied this framework to re-estimate mutation rates of Escherichia coli MG1655 growing under sub- minimal inhibitory concentration (MIC) doses of kanamycin (an aminoglycoside acting on protein synthesis [41]), norfloxacin (a fluoroquinolone acting on the DNA-gyrase complex and potentially leading to the creation of DNA breaks through the cell machinery [42]), and hydrogen peroxide (an oxidizing agent producing reactive oxygen species that directly affect DNA independently of the cell machinery [43,44]). All these antimicrobials have previously been reported to significantly elevate mutation rate [28,31]. We find the same pattern when computing mutation rate without taking death into account. However, for norfloxacin and kanamycin, the estimated increase of mutation rate due to treatment is strongly reduced after conservatively correcting for death. There remains no signal of stress-induced mutagenesis in the case of kanamycin. These findings confirm our suspicion that neglecting death leads to substantial overestimation of mutation rate under stress.
We also show that mutation rate estimation does not only present experimental and mathematical challenges; it is also not the most relevant measure of evolvability, meaning the capacity of a population to generate adaptive genetic diversity. Indeed, some of the studied subinhibitory treatments cause a significant drop in population size due to both bactericidal and bacteriostatic effects and thus lead to a smaller genetic diversity despite a higher mutation rate. Ironically, evolvability, approximated as the generation of genetic diversity, can be much more easily estimated from experimental data than mutation rate. In our experiments, antibiotics and hydrogen peroxide have very different effects on evolvability: both subinhibitory norfloxacin and kanamycin treatments significantly reduce it, while hydrogen peroxide treatment strongly increases it.
Subinhibitory treatments are not necessarily sublethal, because minimal inhibitory concentration is defined at population scale. An antimicrobial treatment is subinhibitory if the population grows (i.e., colony-forming unit [CFU]/mL increases or, more crudely, culture tubes inoculated at low density are turbid after 24 h). However, the death rate can be high, as long as the division rate is higher. Such death events will not be visible to the observer if only population size (CFU/mL) is tracked over time (Fig 1). To reach a given observed final population size, the number of cell divisions has to be higher if there is death. This means that when computing mutation rate using the classical approach (described in the Materials and methods), the number of cell divisions will be underestimated. This is because it is implicitly assumed that there is no death and thus that the final population size is a good approximation for the number of cell divisions. The mutation rate, computed as the number of mutational events divided by the number of cell divisions, will then be systematically overestimated.
The above statement—that mutation rates are systematically overestimated when there is death—is the first intuition motivating our work. We explore this intuition more rigorously (Fig 2) using a simulation approach. For an arbitrary chosen value of mutation rate toward a neutral arbitrary phenotype, we simulate the growth of a population of bacteria inoculated from a small number of nonmutant cells with a chosen constant death rate and track the number of mutant and nonmutant cells. We then compute the mutation rate based on the final state of these simulations, using the standard approach (i.e., the fluctuation test as described in the Materials and methods) to test whether we recover the true value of the mutation rate. As shown in Fig 2, the mutation rate is systematically overestimated when there is death, and the higher the death rate, the higher the overestimation. This result is unchanged when varying other population growth parameters, such as the initial and final population sizes, the mutation rate, and the plating fraction (underlying data have been uploaded to Zenodo 10.5281/zenodo.1211765).
In the previous section, we show that it is necessary to take death into account when computing mutation rate. For this, tracking population size (and thus net growth rate) during antibiotic treatment, as classically done by plating and counting colony forming units, is not sufficient. It is not possible to know whether a decreased net growth rate in the treatment compared to the untreated control is due to a purely bacteriostatic effect (i.e., the population grows more slowly, but without death) or to a bactericidal effect (i.e., the bacteria keep dividing, potentially at the same rate as without antibiotic, but also die). The first scenario will have no effect on the accumulation of mutants as a function of population size, while in the second scenario, turnover implies a higher number of DNA replications and thus more mutants for a given population size, as explained above.
To disentangle these two effects, we designed a method allowing us to compute growth rate and death rate simultaneously using a segregative plasmid. The segregation dynamic allows us to estimate the number of bacterial cell divisions. Combining this information with the change in population size allows us to estimate growth rate and death rate, as explained in the Materials and methods.
Our ultimate goal is to reliably estimate mutation rates of bacteria treated with subinhibitory doses of antimicrobials. To this end, we quantify population dynamics and compute mutation rates toward a chosen neutral phenotype (resistance to rifampicin, conferred by substitutions in the gene rpoB) in populations exposed to subinhibitory doses of other antimicrobials. Our mutagenesis protocol is inspired by the standard fluctuation test with additional measurements of plasmid segregation to compute death rate, as detailed in the Materials and methods. The population dynamics are quantified as a combination of two variables: CFU at various time points (e.g., 0 h, 3 h, 6 h, and 24 h after treatment starts) and relative death rate (compared to birth rate) between pairs of two successive time points. We represent these population dynamics in Fig 3 for the chosen subinhibitory antimicrobial treatments. We use kanamycin at 3 ug/mL, norfloxacin at 50 ng/mL, and hydrogen peroxide (H2O2) at 1 mM, allowing direct quantitative comparisons with the data from Kohanski and colleagues [28]. We also include untreated populations, in which we assume there is no death, to fit the plasmid segregation parameters (see Materials and methods).
We find that for norfloxacin, there is a strong death rate in all phases of growth and a strong impact of the treatment on final population size. For H2O2, death is only detectable in stationary phase and the treatment is mostly bacteriostatic during growth. For kanamycin, the dynamics are more complex, because an initially high death rate leads to a strong decline of population size during the first 6 h of growth, followed by a recovery leading to a final population size close to the one reached in untreated controls. During this second phase of growth following the bottleneck at 6 h, death rate is still substantial. This clearly shows that none of the three studied treatments are fully sublethal and thus that the implicit assumption of no death made when using the standard methods of computation of mutation rate (as done by Kohanski and colleagues [28]) does not apply.
We developed computational tools to quantify mutation rate, taking into account the measured population dynamics and accounting for death. Our software, ATREYU (Approximate bayesian computing Tentative mutation Rate Estimator that You could Use), is described in the Materials and methods. It takes as input any arbitrary population dynamics, described as a list of population sizes (i.e., CFU/mL for several time points) and an associated list of death rates between pairs of consecutive time points. This input is thus exactly what is shown in Fig 3. We apply this method to analyze the results of our mutagenesis protocol, to quantify whether and by how much subinhibitory treatments with kanamycin, norfloxacin, or H2O2 increase mutation rate. We show the effect of treatment on mutation rate in Fig 4. We also plot the uncorrected mutation rate estimate, assuming no death as would be obtained by methods such as FALCOR (Fluctuation AnaLysis CalculatOR) [39], bzRates [40], or rSalvador [38]. Clearly, not taking death into account leads to a strong overestimation of the mutation rate for both kanamycin and norfloxacin. In the case of kanamycin, correctly computing the mutation rate removes all signal for stress-induced mutagenesis. In the case of norfloxacin, this signal is strongly lowered, from a 14-fold to a 6-fold increase. For H2O2, the signal is less affected, which can be attributed to death rate only being significant in stationary phase. This confirms that neglecting death leads to a systematic overestimation of mutation rates and that taking into account the full population dynamics is necessary and leads to significantly different patterns depending on the antimicrobial and its effect on growth and death.
The quantification of mutation rate in different conditions is not sufficient to answer the question of whether subinhibitory antibiotic treatments increase evolvability in general and, in particular, increase the likelihood of emergence of a resistant mutant and thus the probability of treatment failure. Indeed, mutation rate is expressed per DNA division, but, as we have shown in the previous section, antibiotic treatment may significantly change the number of susceptible cells and the number of replications that these cells have undergone. Intuitively, if a treatment multiplies mutation rate by 10 but divides population size by 100, it is not likely to lead to an increased genetic diversity. This intuition has also been given by Couce and Blázquez (Fig 2 of [45]) but has been largely ignored in the literature as it was not the main message of this review. Conversely, a treatment that does not affect mutation rate and only slightly affects carrying capacity but causes death and turnover may result in a significantly increased genetic diversity.
We first show the effect of subinhibitory treatment on final population size in Fig 5. While H2O2 does not affect final population size, there is a strong effect of 1–2 orders of magnitude for norfloxacin and a significant but smaller effect of around 50% reduction for kanamycin. This supports our intuition that at least for norfloxacin, the few-fold increase in mutation rate we report (Fig 4) is probably uncorrelated to any increase in genetic diversity.
We expect the generation of genetic diversity to depend on (i) the number of cells alive, (ii) the population dynamics of these cells, and (iii) their mutation rate. Addressing the effect of stress on mutation rate as done in the previous section is necessary for a proper understanding of the bacterial stress response and of DNA repair mechanisms. Nevertheless, mutation rate is not the relevant measure to understand the effect of stress on the generation of genetic diversity and thus on evolvability.
As a simple quantification of the generation of genetic diversity and thus an approximation of evolvability, we measure the number of mutants at a neutral locus, here the base-pair substitutions conferring resistance to rifampicin in the gene rpoB.
We plot in Fig 6 the absolute number of rifampicin-resistant mutants in the final population for all treatments and for untreated control. Evolvability is reduced by a few-fold by kanamycin treatment (as expected, since this treatment decreases population size without increasing mutation rate). While norfloxacin and H2O2 both induce a small increase in mutation rate, they interestingly have strongly opposite effects on evolvability. Treatment with H2O2 increases evolvability by more than one order of magnitude, while treatment with norfloxacin reduces it by a similar amount. This is due to the very different effects these antimicrobials have on population dynamics: while H2O2 does not affect final population size, norfloxacin causes a strong decrease in population size due to both bactericidal and bacteriostatic effects.
So independently of the question whether antibiotics increase mutation rate, we show that the sub-MIC treatments we studied do not in any way increase evolvability. Thus, the standard rationale, that these subinhibitory treatments would increase the risk of emergence of resistance and treatment failure because of a higher generation of genetic diversity [46], does not hold.
This effect is largely due to a strong reduction in population size, which implies a loss in genetic diversity. Population size and mutation rate are not, however, the only factors affecting evolvability. We may also ask how much the measured turnover in our experiments contributes to evolvability. To answer this question, we simulate the same population dynamics as observed for each treatment but without death: Each population reaches the same final population size as measured in our experiments, with the same mutation rate as computed, but with no death. This is similar to what would happen if the antibiotics only had a bacteriostatic effect. For each simulation, we quantify evolvability using the same measure as previously, i.e., the absolute number of mutants for our phenotype of interest in the final population. We compare this simulated evolvability without turnover with the actual measured evolvability in Fig 7. For kanamycin and norfloxacin, turnover significantly increases evolvability by a few-fold.
In summary, our results show that (1) mutation rate is systematically overestimated in subinhibitory treatments because of death, (2) mutation rate is not the only parameter that controls the generation of genetic diversity or evolvability, (3) population size and turnover play a key role in evolvability, and (4) treatment with subinhibitory doses of norfloxacin or kanamycin significantly decreases evolvability, measured as the generation of genetic diversity at population scale. These results are in apparent disagreement with the conclusions of previous studies on antibiotic-induced mutagenesis. This discrepancy is due to both miscalculation of mutation rates (which occurs when one neglects the effect of population dynamics) and misconceptions about the link between mutation rate and evolvability in these classical papers.
Understanding genetic and environmental control of evolvability is central for the understanding of microbial adaptation to constantly changing environments. Evolvability is defined as the capacity of a population to generate adaptive genetic diversity. This can be decomposed in two variables: the amount of genetic diversity generated by a population (often inaccurately attributed to the mutation or recombination rate only) and the fraction of this diversity that is adaptive. We are here interested in the former. Genetic control over the amount of generated genetic diversity has been studied for a long time in the field of mutation rate evolution [20,22]. The existence of constitutive mutator alleles in bacteria has been discovered before the mechanisms of DNA replication [47], and the selection pressures leading to their transient increase in frequency have been elucidated through both theoretical and experimental studies [23,48]. Observing the evolution and fixation of such mutator alleles from nonmutator lineages in a long-term evolution experiment [49] plausibly facilitated the acceptance of these theories. On the other hand, plastic, environment-dependent control over the generation of genetic diversity has been a controversial paradigm shift in bacterial evolution [18].
It has been proposed for a long time that various stresses can increase mutation rates in bacteria [9,10], including those triggered by antimicrobial treatments [28–32]. Several molecular pathways have been shown to be implicated in this phenomenon, the emblematic one being the SOS response [12]. In this work, we have shown that the effect of stress on mutation rate can not be computed properly with the existing tools, because the underlying mathematical models make the assumption that there is no stress or, more precisely, that the stress does not affect population dynamics. We develop experimental and computational tools to measure population dynamics and compute mutation rates under stress and apply them to the question of mutagenesis due to antibiotic treatment. We have shown that the intuition that low doses of antibiotics are dangerous because they lead to a higher generation of diversity is based on a misinterpretation of valid experimental data for two reasons: (1) the increase in mutation rate is overestimated due to overly simplistic assumptions, and (2) a higher mutation rate does not lead to a higher genetic diversity if population dynamics are affected (e.g., if population size is reduced).
The question of emergence of resistance alleles due to low doses of antibiotics (reviewed by Andersson and Hughes [50]) cannot, however, be entirely addressed by measuring the generation of genetic diversity. The study of adaptive evolution can be decomposed in two parts: generation of diversity and natural selection acting on this diversity. While we have shown that treatment with a subinhibitory dose of norfloxacin does not increase but rather strongly decreases the amount of generated genetic diversity, it has also been reported that resistance alleles can be maintained and enriched by selection, even at very low antibiotic concentration [51]. Such selection of preexistent alleles may be a much more valid reason for concern about subinhibitory treatments. However, the literature is not as unanimous regarding bacteria residing within a patient with an immune system, rather than in a test tube [52]. It has, for example, been suggested that treating with a lower dose of antibiotics could slow down the selection of existing resistance alleles by decreasing their fitness advantage compared to the sensitive, wild-type strain, without compromising the success of the treatment [53, 54]. Combining our results with these papers calls for a reevaluation of the evolution of antibiotic resistance at low doses of antibiotics.
The question of the potentially adverse effects of low doses of antibiotics has been of longstanding interest in the medical community, as is evidenced by the famous quote from Alexander Fleming's Nobel lecture [55], “If you use penicillin, use enough.” However, given the time of this research (penicillin was discovered in 1928 and thus 15 years before Luria and Delbrück), one should not be surprised that this often cited out-of-context advice relies on a rather Lamarckian reasoning in terms of educating rather than selecting for resistance:
Our findings are also relevant outside of the context of evolution during antibiotic treatment. As we mentioned, mutagenesis in bacteria under nutritional stress was a key development in the understanding of the bacterial stress response and DNA repair, with a recent regain of interest [7,8]. Our experimental system can a priori not be applied to study starving bacteria, for two reasons: (1) our plasmid segregation method only gives sufficient signal in nonstationary populations, and (2) many of the observations on starvation-induced mutagenesis are dependent on the presence of some spatial structure (for example, bacterial colonies on agar plates [2,9,10]). In this second case, the population dynamics become much more complex and are unlikely to be realistically approximated by a single relative death rate parameter. But the exact same questions remain to be elucidated in this field: How many cell divisions happen in these starving colonies? In batch cultures, is stationary phase really stationary, or is there some turnover and recycling as recently suggested [56]? And more importantly, where does death come from: is it an unavoidable, externally caused phenomenon; or is there an internal component, such as an altruistic programmed cell death [57], or just traits selected in other environments that give a maladaptation to certain stresses [58]?
Stress-induced mutagenesis is of interest for several research fields, with different questions. We showed that the relevant question in a clinical setting is not directly about mutation rate but about evolvability and that the link between both is confunded by the effect of treatment on population dynamics.
One the evolutionary side, the central question about stress-induced mutagenesis is “Is it adaptive?” Studying the molecular mechanisms of stress response will shed light on one part of the answer: is the increase in mutation rate controlled by the cell, or is it an unavoidable consequence of the stress? In this regard, the three antimicrobials we study seemingly have very different properties. H2O2 is creating reactive oxygen species that directly damage DNA independently of the cell machinery, iron being the only necessary catalyst [43,44]. The way DNA damage leads to mutations is controlled by the cell but is more likely to be a consequence of selection for survival (“survival at the price of the mutation”), rather than selection for evolvability. On the other hand, kanamycin acts on protein synthesis [41], and any hypothetical mutagenic effect would thus go through the cell machinery. Norfloxacin is in between, because it acts on the DNA-gyrase complex, leading to an arrest of DNA synthesis and, in some conditions, to double strand breaks [42].
Recent findings from J. Collins and colleagues, however, suggest that these different scenarios are not as distant as they may seem, because they suggest that the production of reactive oxygen species is a feature of all bactericidal antibiotics [33–35]. While supporting the idea that antibiotic treatment increases mutation rate and does so in correlation with bactericidal activity, these debated findings would also suggest that such increase in mutation rate does not stem from selection for evolvability.
In a nutritional stress scenario, Maharjan and colleagues [8] show that at equal effect on growth rate, limitation of different nutrients has very different effects not only on mutation rate but also on mutational spectrum, again showing the need for a mechanistic understanding of the molecular details and suggesting that the evolutionary outcome is much more complex than a linear increase in mutation rate in response to starvation.
We provide tools that may help further developments of these questions. Our software, ATREYU, can be used to compute mutation rates from mutant counts in populations with arbitrary but known birth and death dynamics. The mutant counts are obtained by a protocol similar to the classical fluctuation test. The birth and death dynamics can be obtained by several methods. We used plasmid segregation, but other methods may be possible, such as segregation of engineered self-assembling fluorescent particles [59], isogenic strain tagging [60], Carboxyfluorescein succinimidyl ester (CFSE) membrane staining [61], or direct microscopic observations at single-cell resolution. Microscopic observations with cell tracking may give much more precise and less noisy information than other methods but are only suitable when the death rate is sufficiently low, because only a limited number of cells can be tracked. We believe that death and cell turnover are crucial factors in evolutionary microbiology but are often neglected, in part due to the lack of standard methods to measure them. In immunology, in which the population dynamic of lymphocytes has been recognized as a central question [62], many methods have been developed [63], including the aforementioned CFSE membrane staining.
While our method could be adapted for many nonstandard assumptions other than death (e.g., fitness cost of the mutation or partial plating), it shares some of the limitations of the more classical systems. Firstly, we suppose that each cell is fully monoploid and has only one chromosome and thus that the number of DNA replications is the number of cell divisions. However, some quinolone antibiotics are known to cause filamentation [64], increasing the number of chromosomes per cell [42] and potentially changing the evolutionary dynamics [65]. Further complicating the picture, recent work [66] shows that even within a single chromosome, multifork replication may cause different ploidy levels on different loci, affecting mutation rate estimates and evolutionary dynamics. Secondly, we also consider that time does not matter, in the sense that the probability of mutation per division is independent of the growth rate, and that nondividing bacteria do not accumulate mutations, justifying the expression of mutation rate as a quantity of mutations per division event and not per unit of time. Since Luria’s and Delbrück’s experiment, this has been the standard assumption both on the microbiological and mathematical side [67]. However, recent data on fission yeasts suggest that nonreplicating cells may accumulate mutations at a different rate and spectrum than diving cells [68]. Finally, we make the assumption of homogenous behavior in the population, excluding the possibility that different subpopulations have different death and mutation rates. The question of whether a small subpopulation in a different physiological state may contribute most of the mutational supply is still unresolved. Theoretical work [69] shows that such situation could have a large impact on the evolutionary dynamics.
Zooming out from evolutionary microbiology, mutagenesis research in bacteria shows an interesting parallel with recent advances in cancer research. For a given cell growth dynamic (organogenesis, from stem cells to an organized population of differentiated cells), a higher mutation rate (expressed per cell division) will boost the accumulation of mutations and thus the risks of cancer. This increase in mutation rate can be genetic, such as in the case of hereditary nonpolyposis colon cancer caused by a deficiency of mismatch repair [70], or environmental, such as exposure to carcinogenic compounds [71–73]. All of this is now part of textbook science on cancer and is similar to an increase in mutation rate in a bacterial population due to genetic (mutator alleles [47]) or environmental (stress-induced [18] or stress-associated [26] mutagenesis) factors.
Tomasetti and Vogelstein [74] recently reported that the number of stem cell divisions is a strong predictor of cancer risk per organ. This is in parallel with our findings, which show that the number of cell divisions is central to predict the generated genetic diversity in a population of cells. Tomasetti and Vogelstein caused a major controversy by concluding that cancers would thus mostly be due to “bad luck” (i.e., unavoidable consequences of the large number of cell divisions) rather than to environmental factors (e.g., exposure to mutagenic chemicals). We show here that the generation of genetic diversity depends on both mutation rate and cell population dynamic, which is in line with many studies that have criticized the interpretation of the data made by Tomasetti and Vogelstein.
The challenge of understanding evolvability in bacterial population is thus strikingly similar to the one of understanding cancer, in the sense that the outcome depends on a complex interplay of extrinsic and intrinsic factors acting at different scales. In the case of bacteria, additional complexity stems from the fact that the same treatments may both impact the number of cell divisions (death and turnover) and the mutagenicity of each division. The picture is further complicated by the difficulty of disentangling the direct effects of the drug from the effects of the stress response triggered by the drug. But fortunately, while separating and measuring each factor requires complex experimental methods and mathematical tools, measuring evolvability on neutral loci is simpler, at least in bacteria. We hope that our study will encourage researchers in the field to question more not only the appropriateness of the tools they use for mutation rate estimation and the assumptions implicitely made by using these tools but also the pertinence of the variable they choose to report.
Our mutagenesis protocol is directly inspired by the one used by Kohanski and colleagues [28] (which is in turn similar to that of Luria and Delbrück [75]) with the inclusion of a segregative plasmid to compute death rate, as explained further below and graphically represented on Fig 8.
A culture of E. coli MG1655 (with plasmid pAM34) is inoculated from a freezer stock and grown overnight in LB supplemented with 0.1 mM IPTG and 100 ug/mL of ampicillin (to ensure maintenance of pAM34). After the culture reaches stationary phase (at least 15 h of growth), it is washed 3 times in normal saline (9 g/L NaCl) to remove traces of IPTG and then diluted 10,000 times in a 500 mL baffled flask containing 50 mL of LB (to maximize oxygenation). After 3.5 h of growth, the culture is inoculated at a ratio of 1:3 in 24 culture tubes containing a total volume of 1 mL of LB supplemented with one of the studied antimicrobials at subinhibitory concentration (3 ug/mL kanamycin, 50 ng/mL norfloxacin, 1 mM hydrogen peroxide, or untreated control). After 24 h of growth at 37°C, the cultures are plated at appropriate dilutions on 3 different LB agar medium: LB only to count the total number of bacteria (CFU), LB supplemented with 100 ug/mL ampicillin + 0.1 mM IPTG to count the number of bacteria bearing a copy of the segregative plasmid, and LB supplemented with 100 ug/mL rifampicin (plated volume: 200 uL) to count the number of mutants toward the phenotype of interest. Additionally to this 24 h time point, cultures are also plated on LB and LB + ampicillin + IPTG at intermediate time points (3 h and 6 h) to have a more accurate quantification of plasmid segregation dynamics and thus a better time resolution for the estimation of death rate. The plates are incubated between 15 h and 24 h for LB and LB ampicillin IPTG and exactly 48 h for LB rifampicin before counting colonies. Further experimental details are given in S5 Supporting information.
pAM34 is a colE1 derivative whose replication depends on a primer RNA put under the control of the inducible promoter pLac [76]. Under the presence of 0.1−1mM IPTG (nonmetabolizable inducer of the lactose operon), the plasmid is stably maintained in every cell. When IPTG is removed from the growth medium, the plasmid is not replicating anymore, or not as fast as the cells divide, and thus is stochastically segregated at cell division. The decrease in plasmid frequency between two time points then allows us to compute the number of bacterial cell divisions that occurred between these two time points. Combined with the change in population size, this allows us to compute average death rate and growth rate between these two time points (see Fig 9 and mathematical explanations below). Such segregation measures have been used in a less quantitative way by other researchers [77,78] to crudely infer overall population turnover in vivo.
pAM34 also carries a betalactamase. The number of plasmid-bearing bacteria can thus be counted by plating an appropriate dilution of the culture on LB supplemented with 0.1 mM IPTG (to ensure maintenance of the plasmid within colonies founded by a plasmid-bearing cell) and 100 ug/mL ampicillin (to only permit growth of colonies founded by a plasmid-bearing cell). The total number of bacteria is determined by plating an appropriate dilution of the culture on LB.
Because mutational dynamic does not depend on time, we chose to compute relative death rate (ratio of death rate and growth rate as functions of time), which is the average number of death events per division event.
The link between plasmid segregation, death, and number of divisions between two time points can be expressed mathematically as follows.
If we have the following:
then the plasmid is diluted/segregated at each division following the equation
Ffinal=Finitial×(1+res2)g
So we can estimate
g=log2(FfinalFinitial)/log2(1+res2)
Without any death, we would have
gno−death=log2(Nfinal/Ninitial)
The difference between the true number of generations g computed from plasmid frequency and this number of generations gno−death computed based on the assumption that there is no death, allows us to estimate relative death rate as follows:
Nfinal=Ninitial×2(1−d)*g
This yields
d=1−log2(Nfinal/Ninitial)g
and thus
d=1−log2(Nfinal/Ninitial)log2(Ffinal/Finitial)×log2(1+res2)
The only remaining free parameter to estimate is res, which is estimated by performing growth kinetics without antibiotic treatment (in LB medium) and thus without (or with negligible) death. We then have
g=gno−death
and thus
log2(1+res2)=log2(Ffinal/Finitial)log2(Nfinal/Ninitial)
from which we can fit the value of the segregation parameter log2(1+res2) based on the values of F and N estimated by plating. Further experimental and mathematical details on the plasmid segregation system are given in S2–S5 Supporting Informations.
Most modern measures of mutation rate rely on the same standard protocol, the fluctuation test [79], directly inspired by the Luria and Delbrück experiment [75]: several cultures are inoculated with a small population of nonmutant bacteria, are grown overnight and are then plated on selective media (to count the number of mutants in the final population) and on nonselective media (to count the total number of bacteria in the final population). The number of mutants r in the final population (or, rather, its distribution over several replicate populations) is used to estimate the number of mutational events m happening during growth. One should note that these two numbers are not equivalent, because one mutational event can lead to several mutants in the final population if it happens early during growth, making this part of the computation complicated for intuition, although good mathematical tools are available. The total number of bacteria N is assumed to be very close to the number of cell divisions (and thus the number of genome replications) because the initial number of bacteria is much smaller. Mutation rate can thus be estimated as μ = m/N.
The many existing software packages used to compute m from the observed distribution of r use an analytical expression of the probability-generating function (pgf) of the number of mutants in the final population [80]. The only free parameter is the number of mutational events (equivalent to the value of the mutation rate per division when scaled with population size). This parameter is estimated from plating data using the maximum likelihood principle. The most used implementation of this idea is FALCOR [39], available on a webpage: http://www.keshavsingh.org/protocols/FALCOR.html.
Other software packages implementing the same ideas have been developed more recently, including, for example, rSalvador [38] and bzRates [40], which also implement a few alternative assumptions such as fitness impact (cost or benefit) of the focal mutation or a more accurate correction for plating efficiency than the one suggested by FALCOR [81].
However, to this day, no available software allows users to compute mutation rate when there is death. Some papers derived analytical expression of the pgf of the number of mutants in the final population in conditions in which there is death [82], but this has to our knowledge never been applied to real data nor implemented in a software package. In theory, such computations could easily be implemented in a tool similar to FALCOR (web server) or rSalvador (software package). However, the basic assumption of the derived formula is that death rate is constant. This assumption is the price to pay for an analytical expression for the pgf and is unfortunately not appropriate in our case, given the observed death kinetics (see Fig 3). On the other hand, given the computational power available today, we believe that analytical computations are not always necessary. In our case, while the measured population dynamics do not allow us to derive an analytical expression of the pgf, it is straightforward to simulate many times such population dynamics with an arbitrary mutation rate and to obtain an empirical distribution of the number of mutants. Running these simulations for any possible value of the mutation rate parameter then allows Bayesian inference: we look for the simulated mutation rate that gives the closest distribution to the one experimentally observed, as graphically represented in Fig 10. Such methods are classically referred to as Approximate Bayesian Computing. We implemented such simulations and inference in a Python software package, ATREYU, and use this software as the heart of our data analysis.
|
10.1371/journal.pntd.0001380 | The Field-Testing of a Novel Integrated Mapping Protocol for Neglected Tropical Diseases | Vertical control and elimination programs focused on specific neglected tropical diseases (NTDs) can achieve notable success by reducing the prevalence and intensity of infection. However, many NTD-endemic countries have not been able to launch or scale-up programs because they lack the necessary baseline data for planning and advocacy. Each NTD program has its own mapping guidelines to collect missing data. Where geographic overlap among NTDs exists, an integrated mapping approach could result in significant resource savings. We developed and field-tested an innovative integrated NTD mapping protocol (Integrated Threshold Mapping (ITM) Methodology) for lymphatic filariasis (LF), trachoma, schistosomiasis and soil-transmitted helminths (STH).
The protocol is designed to be resource-efficient, and its specific purpose is to determine whether a threshold to trigger public health interventions in an implementation unit has been attained. The protocol relies on World Health Organization (WHO) recommended indicators in the disease-specific age groups. For each disease, the sampling frame was the district, but for schistosomiasis, the sub-district rather than the ecological zone was used. We tested the protocol by comparing it to current WHO mapping methodologies for each of the targeted diseases in one district each in Mali and Senegal. Results were compared in terms of public health intervention, and feasibility, including cost. In this study, the ITM methodology reached the same conclusions as the WHO methodologies regarding the initiation of public health interventions for trachoma, LF and STH, but resulted in more targeted intervention recommendations for schistosomiasis. ITM was practical, feasible and demonstrated an overall cost saving compared with the standard, non-integrated, WHO methodologies.
This integrated mapping tool could facilitate the implementation of much-needed programs in endemic countries.
| Neglected tropical diseases (NTDs) cause significant physical debilitation, lowered economic productivity, and social ostracism for afflicted individuals. Five NTDs with available preventive chemotherapy: lymphatic filariasis (LF), trachoma, schistosomiasis, onchocerciasis and the three soil-transmitted helminths (STH); have been targeted for control or elimination, but resource constraints in endemic countries have impeded progress toward these goals. We have developed an integrated mapping protocol, Integrated Threshold Mapping (ITM) for use by Ministries of Health to decide where public health interventions for NTDs are needed. We compared this protocol to the World Health Organizations disease-specific mapping protocols in Mali and Senegal. Results from both methodologies indicated the same public health interventions for trachoma, LF and STH, while the ITM methodology resulted in a more targeted intervention for schistosomiasis. Our study suggests that the integrated methodology, which is also less expensive and logistically more feasible to implement, could replace disease-specific mapping protocols in resource-poor NTD-endemic countries.
| Neglected tropical diseases (NTDs) are parasitic and bacterial diseases that affect an estimated 2.7 billion of the world's poorest people, causing significant physical debilitation, lowered economic productivity, and social ostracism for afflicted individuals [1]. Five NTDs with available preventive chemotherapy: lymphatic filariasis (LF), trachoma, schistosomiasis, onchocerciasis and the three soil-transmitted helminths (STH); have been targeted for control or elimination, but resource constraints in endemic countries have impeded progress toward these goals [2]. In order to achieve the rapid scaling-up of programs necessary to reach elimination and control targets, some by as early as 2020, the global health community is focusing on developing strategies that capitalize on synergies between previously independent elimination and control programs for these diseases. Traditional efforts to treat and prevent NTDs through vertical programs are often costly, and the integration of program components has the potential to cut the costs of NTD programs [3], [4].
At the core of public health efforts to fight the five NTDs mentioned above is a strategy of mass drug administration (MDA) of at-risk populations with safe and effective drugs often donated by pharmaceutical companies [5]. Before an MDA can be launched, a country must demonstrate that the disease threshold for public health intervention, as established by the World Health Organization (WHO), has been surpassed [6], [7]. The first step in this process is to review the available data, followed by collecting missing data by conducting prevalence surveys. Currently, each NTD program has its own methodology [8]. Conducting multiple surveys in the same country can be costly and burdensome to national disease programs. As a result, prevalence surveys are only conducted when funding has been secured, and data to help international partners and program managers with planning and advocacy, and drug donation programs with drug projections are unavailable in many parts of sub-Saharan Africa [8]. For this reason, there is a need for feasible and practical protocols that can be implemented by Ministries of Health and whose results are accepted by WHO and drug donation programs. Because NTDs tend to overlap in geographic areas, it is logical that an integrated approach to mapping NTDs might result in more efficient identification of populations needing treatment [9].
This paper describes a novel methodology, the integrated threshold mapping (ITM) methodology, for an integrated mapping survey for LF, trachoma, schistosomiasis, and STH. Our methodology does not provide prevalence figures, its intended purpose is to determine whether a disease has attained the threshold for public health intervention by attempting to balance epidemiologic rigor and field practicality. The ITM protocol has been field tested in one district in each of two countries by comparing the ITM methodology to the standard disease-specific WHO mapping methodologies. Results from both methodologies were compared in terms of public health intervention based on the disease specific data and feasibility, including cost.
The ITM methodology is derived from rapid mapping methodologies used by LF and schistosomiasis programs. It is designed to provide programmatically useful mapping data in a logistically practical and cost-efficient way.
In Mali, the research site included nine sub-districts of the Banamba district (Koulikoro region), an area with an estimated population of 88,232 persons (2008). In Senegal, research was conducted in all 11 non-urban sub-districts of Diourbel district (Diourbel region), an area with an estimated population of 150,889 persons (2010). In both countries, the majority ethnic groups, Bambara and Wolof, respectively, are subsistence farmers living in established settlements; a minority Peuhl group in Mali is semi-nomadic and engages in cattle herding. The entirety of both the Banamba and Diourbel districts is located within the Soudanian ecological zone [10]. NTD program activities including the provision of preventive chemotherapy for the five NTDs have been ongoing in Banamba district, but not in Diourbel district.
The standard indicators and target age groups currently recommended by the WHO for the targeted NTDs were employed for both the ITM and WHO methodologies, although some minor adaptations were made for trachoma Table 1. For LF, the immunochromatography card test (ICT, BinaxNOW Filariasis, Alere Inc.), which tests blood drawn by finger stick, was used to measure the presence of Wuchereria bancrofti antigen in persons aged ≥15 years and resident in the village for at least ten years [11]. Results were read after ten minutes and recorded by a trained laboratory technician. For trachoma, an ophthalmic technician examined both eyes of children aged 1–9 years for clinical signs of active trachoma (trachomatous inflammation follicular [TF]) and of females aged ≥15 years for clinical signs of trichiasis (trachomatous trichiasis [TT]) using a binocular loupe and natural light according to the WHO Simplified Grading System [12]. For Schistosoma haematobium, the presence of haematuria in urine samples from children aged 6–9 years was detected by urine reagent strips. Reagent strips were dipped in the urine samples and read after one minute by comparing them to a colorimetric scale by a trained laboratory technician. At the request of the Ministry of Health (MoH), urine samples were also tested by filtration in Senegal as part of the WHO methodology [13]. For Schistosoma mansoni and STH, the prevalence and egg load (eggs per gram of feces) were calculated using the Kato-Katz method [14].
The study protocol received ethical approval from the Ethics Committee of the Faculty of Medicine, Pharmacy and Odontology-Stomatology, University of Bamako, Bamako, Mali; from the National Ethics Committee for Health Research (CNERS), Dakar, Senegal; and from CDC's Internal Review Board, Atlanta, GA, USA.
All persons over the age of eighteen years were asked to provide written consent to participate in the study. The school director or a member of the team explained the study to the school-aged children (5–14 years) using a verbal assent script describing the study, and the children who participated in the study were asked to give their verbal assent. Written consent to participate in the study was also obtained from parents or custodians of children between the ages of 1–18 years.
Children who were found to have active trachoma received two tubes of tetracycline ointment, to be applied twice daily for a period of six weeks. Females found to have trichiasis were referred to the nearest health center that provides trichiasis surgery. Persons who tested positive for W. bancrofti antigen and children who were found to be infected with S.haematobium, S.mansoni or STH were referred to the district health center in case an MDA was not planned in the district.
To determine feasibility, we took into consideration time and resources needed to conduct field activities and overall costs. All receipts were collected, and actual expenditures were recorded. The cost data were compiled into four categories: training, per diems (national and district levels including lodging), travel to the field (renting of vehicles, drivers and fuel) and supplies (medical and laboratory). Any costs of inputs that were used for multiple activities were distributed evenly among the activities. Although LF testing was only conducted once, costs for conducting the testing were included for both methodologies. The training of health personnel on diagnostic methods, salaries and data entry and analysis conducted by CDC were not included in the cost analysis.
Results for trachoma, STH and LF were analyzed at the district level for both methodologies Table 2. Results for schistosomiasis were calculated at the sub-district level for the ITM methodology and at the ecological zone level for the WHO methodology. For the ITM methodology, the results of the two villages per sub-district were combined, and for the WHO methodology, all schools were combined. The public health interventions based on the WHO thresholds Table 1, were compared for both methodologies. Feasibility, including cost, time and the human resource needs was compared for both methodologies.
In Banamba District, Mali, 1,898 persons, including 900 children (1–9 years), in 18 villages were surveyed for the ITM methodology, and 4,479 persons, including 2,738 children (1–9 years) in 35 villages were surveyed for the WHO methodology. For trachoma, eight villages already tested for the integrated method were also sampled for the WHO method. As shown in Figure 3, the results of both methodologies indicated no need for MDA for trachoma (ITM Method: 1.9%, WHO method: 4.7%, 95% CI: 2.9–6.5 or STH: 0%, 0%) within the surveyed district. The methodologies were also concordant in indicating a need for TT surgeries (ITM Method: 2.6%, WHO Method: 3.7%, 95% CI: 2.4–5.1). The LF mapping was added to validate the feasibility of the integrated mapping and not to validate the results.
Because the ITM method had as its purpose to refine the schistosomiasis mapping, the two methods differed, in indicating a need for schistosomiasis treatment: the WHO methodology indicated a need for MDA for schistosomiasis among only school-aged children in the entire area. The ITM methodology indicated that four sub-districts were in need of MDA for schistosomiasis among school-aged children only, and that five sub-districts were in need of MDA for schistosomiasis for the whole population. Treating the whole population in the sub-districts where this was indicated by the integrated mapping protocol would result in nearly 33,500 additional at-risk persons being treated than if only school-aged children in these sub-districts were treated (as indicated by the WHO protocol).
In Diourbel District, Senegal, 2,734 persons, including 1,100 children (1–9 years), in 22 villages were surveyed for the ITM methodology, and 4,614 persons, including 2,914 children (1–9 years) in 34 villages for the WHO methodology. For trachoma, one village already selected for the ITM method was also sampled for the WHO method. Both methodologies indicated the need for MDA for trachoma in the surveyed district (ITM Method: 14.9%, WHO Method: 13.4%, 95% CI: 9.8–17.0) and both methodologies indicated that there was no need for intervention for STH: 0%, 0%.
Both ITM and WHO methodologies were also concordant regarding the need for TT surgeries (ITM Method: 4.7%, WHO Method: 6.1%, 95% CI 4.4–7.7). As mentioned above, the LF mapping was added to validate the feasibility of the integrated mapping and not to validate the results.
In Mali, however, the public health interventions differed with regards to schistosomiasis treatment because the ITM method had as its purpose to refine the schistosomiasis mapping. The WHO methodology indicated a need for MDA for schistosomiasis treatment among only school-aged children in the entire area, while the ITM methodology indicated six sub-districts in need of MDA for schistosomiasis among school-aged children only, four sub-districts in need of MDA for schistosomiasis for the whole population and one sub-district which did not pass the threshold for treatment. Compared to the district based WHO protocol, the ITM methodology targeted nearly 46,000 extra at-risk individuals for treatment.
In the ITM methodology, we compared the public health intervention decisions for schistosomiasis that would result from using only the randomly or only the targeted selected village Table 3. Only in nearly half of the sub-districts (5/11 in Senegal; 5/9 in Mali), basing a treatment decision on sampling in a village considered highly endemic for schistosomiasis would have resulted in treatment for more persons than if the decision had been based on sampling in a randomly selected village. The contrary was the case for 3/11 villages in Senegal and 1/9 villages in Mali.
To compare the feasibility of the two protocols, we compared the time it took to conduct each survey. For the ITM methodology, the total number of days on the field was 46 person-days in Mali and 56 persons days in Senegal. It took approximately two hours to survey one village; in a village where LF testing was done, four hours were needed. The schistosomiasis and STH team spent an additional three hours each day preparing and reading the Kato-Katz slides. The time to travel to the villages from the health center of the sub-district varied according to geography from approximately 15 minutes to 1 hour. For the WHO methodology, the total number of days on the field was 56 person-days for Mali and 58 person-days for Senegal. Each trachoma team took approximately 4–5 hours to survey one village. The schistosomiasis and STH team surveyed one village a day with approximately two hours to collect the samples and an additional five hours to prepare and read the 50 Kato-Katz and urine samples in the district laboratory. The time to travel to the villages varied according to geography from approximately 15 minutes to over 2 hours.
The results of the cost data analysis Table 4, indicate that using the ITM methodology resulted in a 31% overall cost savings in Mali ($6,968 vs. $10,039), and a 19% overall cost savings in Senegal ($8,442 vs. $10,372). In both countries, the ITM methodology used resources more efficiently than the WHO methodology in the areas of travel, supplies and team training, as shown in Table 4.
We describe an integrated NTD mapping methodology (Integrated Threshold Mapping) that can be used as an operational tool by Ministries of Health in NTD-endemic countries to determine if the threshold needed to launch disease-specific public health intervenions has been reached. With the recent increased interest in NTDs, there is a real need for an integrated mapping approach that can provide district level data in a timely manner [9], [18], [19]. In developing this ITM methodology, we sought to balance epidemiologic rigour with field practicality, resulting in an approach that can be used only for determining where public health interventions are needed. With the exception of some minor modifications for trachoma, we have retained the key indicators and age groups used in the disease-specific WHO mapping guidelines, but have adapted the sampling methodologies. This ITM methodology reduces costs, and the need for manpower and resources, and gives MoHs a simpler and quicker way to estimate their NTD needs.
The first step in making a NTD action plan is to evaluate the existing data, including the methodology and time of the data collection, and to determine whether any factors that could influence the prevalence of the disease have changed since the data were collected. The second step is to collect data where existing data are missing or out of date; our ITM mapping protocol was designed for this purpose. Because it will be rare that all four NTDs will need mapping in any one district, the methodology consists of disease-specific modules. For each specific situation, modules can be combined to create the situation-specific NTD integrated protocol Figure 2.
To understand the value and limitations of this protocol, it is critical to understand that this protocol is not claiming to provide epidemiologically correct prevalence data because both the villages and the individuals tested were not randomly selected. First, this implies that we cannot recommend using findings from this methodology as a baseline for measuring impact at the district level. This is also not possible for the WHO recommended mapping protocols for schistosomiasis, STH and LF, but it is the case for the trachoma WHO methodology. A possible second implication could be that the WHO recommended threshold could not be used when using this methodology; but here again, this would only be an issue for trachoma, bearing that the WHO methodology does not provide epdemiological correct prevalence figures for schistosomiasis, STH and LF.
One barrier to integrated mapping of NTDs is the misconception that mapping can only be integrated if all testing is done using the same age groups. In our ITM mapping protocol, we insisted on using the program-specific age groups as indicated by WHO with the exception of some adaptation for trachoma. The advantage of using these established age groups was that the WHO thresholds were applicable and that few people had to undergo multiple tests and examinations; only in a limited number of cases were people asked to undergo more than one test or examination. In each village, we chose to use a convenience sample for reasons of field practicality. Asking people of different ages to come to a central location in each village facilitated data collection and was much more time-efficient than house-to-house visits, such as those used in the WHO trachoma protocol. As the field testing in both countries showed, the data of both methodologies resulted in the same public health recommendations.
The ITM protocol maintains the WHO-recommended disease indicators and thresholds for LF, schistosomiasis and STH, but we used a novel sampling frame for more practical field implementation. To improve the representativeness for the TF indicator, we selected a pre-determined number of children between the ages of 1–5 and 6–9. This was intended to prevent biased estimates in case the 6–9 year age group made up the majority of children being graded for trachoma because prevalence rates are highest in 1–5 year olds. Although current WHO guidelines measure trichiasis in adults of both sexes as a standard indicator, literature shows that different adult age groups have been used to determine TT prevalence [20], [21], [22]. After consultation with trachoma experts, we decided to use women above the age of 15 yrs [22].
Because the ITM methodology is focused on public health action, the sampling frame is directly linked to the implementation unit for MDA for each of the NTDs. Based on field experience, we decided to slightly modify those MDA implementation units to make public health interventions more feasible in the field, as shown in Table 2. Because a village-centered approach is difficult for implementation and supervision for a large-scale national program, we decided to use the district as the primary implementation unit for STH as it is recommended by WHO for LF and trachoma. Our methodology found the same public health intervention for trachoma and for STH. For STH, both methodologies concluded that no intervention was necessary, which confirmed the reliability of our methodology in the two countries. The LF testing was included to validate only the field feasibility of integrateing testing for four NTDs. In the case of schistosomiasis, a disease that is very focal in nature, and for which treatment medication is costly and in limited supply, we feel that it would be more appropriate to use the sub-district as implementation units to determine needs for treatment because a district-based approach may result in overtreating or undertreating subpopulations within the district [23]. Our results show that by using the sub-district as I.U., the MDA will be more targeted to people at risk. In our study, when the WHO-recommended ecological zone was used as the implementation unit, highly endemic (prevalence of >50%) or low endemic (prevalence of <10%) sub-areas of the district were not identified, but were identified by the integrated methodology. As a result, using the WHO protocol would have meant that tens of thousands of people living in highly endemic areas would not have received treatment, and that other people not in need of treatment would have been treated, as would have been the case in Senegal.
The integrated approach proved to be more efficient in cost, transport, time in the field and the use of human resources. Because transportation and per diems are variables that significantly increase the cost of mapping, we created one small team of experienced technicians already active in the national MoH program instead of using multiple disease-specific teams. We decided to limit the number of team members to five by capitalizing on their laboratory expertise in parasitic diagnostic techniques mastered during their lab technician training. Each member of the team performed multiple tasks, which increased time and energy efficiencies and also made planning and field organization easier. We also counted on the strong engagement of local staff as we saw this as capacity building of district, dispensary, and village health staff. Although the team members from the different NTD programs were initially reluctant to work as one team, because it was a new concept, this hesitancy evaporated over the course of the survey. The team members also noticed that resources were used more efficiently compared to the WHO methodologies: the ITM methodology used one vehicle to transport the team, versus 3 used in the WHO methodologies, employed a 5-member team instead of a 7-member team, and took about half as much time to implement as the WHO methodologies.
It is also important to mention that certain cost savings from integrated mapping such as time saved by only having to plan one survey compared with several different protocol meetings and logistic preparations is priceless for overburdened health staff. In addition, the workload for all the preparations and supervision can be shared among different program coordinators.
It is worth mentioning that cost savings are just a single variable that justifies conducting integrated surveys: mapping is often the first field activity of an integrated NTD program, and the creation of the team demonstrated the concept of integrating vertical programs to all levels of the MoH and this first activity can help with further collaboration for implementing integrated public health interventions.
Although the ITM methodology resulted in substantial cost savings compared to the WHO methodology, the cost depends on the number of sub-units in a district. This means, for example, that the sampling size for schistosomiasis could be much higher using the ITM methodology than using the WHO methodology. However, this would result in a more targeted MDA for schistosomiasis, which is important because there is limited availability of donated praziquantel.
The results from both the ITM and WHO methologies indicate that health workers are not always well-informed about where schistosomiasis is most prevalent. The results show that treatment decisions, based on the purposely-selected villages, did not systematically result in more treatments than those based on the randomly-selected villages.
We encountered some limitations in conducting the integrated mapping. We are comparing a convenience sampling for schistosomiasis and STH mapping with an accepted WHO convenience sampling. Ideally, we would have compared both methods to an independent, “gold standard” survey methodology. The main limitation for the trachoma mapping is that a total of nine of the 60 trachoma villages sampled were selected for both methodologies. To decrease the burden of the inhabitants by not having the same children examined twice, we included the data collected in the integrated mapping for the WHO method and added only 30 children by visiting the households. For three villages, the selection bias was limited because the villages were so small that the likelihood that all the children in the integrated method would have also been included by visiting the houses was very high, but for the other six villages, a selection bias was likely introduced because up to 62% of the sample was selected by convenience only and those persons would maybe not have been included if the sample was collected randomly.
This study shows a novel integrated mapping protocol to determine whether thresholds for public health interventions have been reached. The approach is logistically practical, cost-efficient and flexible. For schistosomiasis, our approach also results in more targeted MDAs compared with the district level implementation currently adapted by most integrated NTD program. The protocol uses mainly the age-specific disease indicators as recommended by WHO. Further, it sets the stage for all of the following integrated program activities essential for NTD elimination and control. Based on the lessons learned from the implementation in the first two countries and feedback we received form NTD colleagues, we do recognize that this novel mapping approach requires some modification to ensure that the most useful data are collected. For this reason, we are currently field testing an adapted protocol with all villages selected randomly among other minor changes.
|
10.1371/journal.pbio.2006421 | Oxytocin blocks enhanced motivation for alcohol in alcohol dependence and blocks alcohol effects on GABAergic transmission in the central amygdala | Oxytocin administration has been reported to decrease consumption, withdrawal, and drug-seeking associated with several drugs of abuse and thus represents a promising pharmacological approach to treat drug addiction. We used an established rat model of alcohol dependence to investigate oxytocin’s effects on dependence-induced alcohol drinking, enhanced motivation for alcohol, and altered GABAergic transmission in the central nucleus of the amygdala (CeA). Intraperitoneal oxytocin administration blocked escalated alcohol drinking and the enhanced motivation for alcohol in alcohol-dependent but not nondependent rats. Intranasal oxytocin delivery fully replicated these effects. Intraperitoneal administration had minor but significant effects of reducing locomotion and intake of non-alcoholic palatable solutions, whereas intranasal oxytocin administration did not. In dependent rats, intracerebroventricular administration of oxytocin or the oxytocin receptor agonist PF-06655075, which does not cross the blood-brain barrier (i.e., it would not diffuse to the periphery), but not systemic administration of PF-06655075 (i.e., it would not reach the brain), decreased alcohol drinking. Administration of a peripherally restricted oxytocin receptor antagonist did not reverse the effect of intranasal oxytocin on alcohol drinking. Ex vivo electrophysiological recordings from CeA neurons indicated that oxytocin decreases evoked GABA transmission in nondependent but not in dependent rats, whereas oxytocin decreased the amplitude of spontaneous GABAergic responses in both groups. Oxytocin blocked the facilitatory effects of acute alcohol on GABA release in the CeA of dependent but not nondependent rats. Together, these results provide converging evidence that oxytocin specifically and selectively blocks the enhanced motivation for alcohol drinking that develops in alcohol dependence likely via a central mechanism that may result from altered oxytocin effects on CeA GABA transmission in alcohol dependence. Neuroadaptations in endogenous oxytocin signaling may provide a mechanism to further our understanding of alcohol use disorder.
| Alcohol use disorder is characterized by a reduction in reward function and gain in stress function. The neuropeptide oxytocin is involved in the regulation of both reward and stress systems. We tested the hypothesis that oxytocin administration could normalize the dysregulations that occur in alcohol dependence and thereby reduce alcohol drinking in dependent rats. We demonstrated that oxytocin administered systemically, intranasally, or into the brain blocked the enhanced drinking exhibited by alcohol-dependent rats. These effects were demonstrated to be centrally rather than peripherally mediated. Oxytocin blocked this enhanced alcohol drinking at doses that did not alter non–alcohol-related behaviors or alcohol drinking in nondependent rats, suggesting the effect was specific to alcohol drinking in alcohol dependence. Ex vivo electrophysiological recordings in the central nucleus of the amygdala (CeA; a key brain region in the network of dysregulations induced by alcohol dependence) indicated that oxytocin blocked the facilitatory effects of alcohol on inhibitory signaling in dependent but not nondependent rats. These results provide compelling evidence that dysregulation in the endogenous oxytocin system is of functional relevance to a mechanistic understanding of alcohol use disorder.
| Alcohol use disorder is a global public health issue; 6% of the world’s population is subject to morbidity and mortality from alcohol [1]. Alcohol dependence (reflecting dysfunction equivalent to moderate to severe alcohol use disorder) is a chronic, relapsing disorder that develops as a result of neuroadaptations in brain reward, stress, and executive function systems controlled by neurocircuits that involve the basal ganglia, extended amygdala, and prefrontal cortex, respectively [2]. In alcohol dependence, the motivation to seek and consume alcohol is driven by reward hypofunction and stress sensitization. Both of these processes contribute to a negative emotional state that drives enhanced motivation for alcohol drinking via negative reinforcement [3,4].
Oxytocin has been hypothesized to have an important role in social behavior, fear and anxiety, and learning and memory [5]. Recent optogenetic and neuroanatomical mapping studies have advanced our understanding of the complexity of oxytocin’s neurocircuitry and the role of signaling by the endogenous oxytocin system in various behaviors [6]. Oxytocin modulates stress and reward function and has long been suggested as a putative treatment for addiction [7]. Reports that oxytocin administration can reduce drug consumption, withdrawal, and relapse associated with various drugs of abuse in preclinical models has greatly enhanced the interest in this area [8,9]. Indeed, oxytocin decreased economic demand for stimulants [10], reinstatement of stimulant seeking [11,12], opioid tolerance and withdrawal [13], and development of alcohol tolerance [7]. Particularly relevant for treating alcohol dependence, oxytocin has an anti-stress and anti-anxiety profile [14]. Preliminary clinical studies have demonstrated that intranasal oxytocin administration reduced alcohol-withdrawal–related anxiety and alcohol craving in alcohol-dependent patients in early abstinence [15], reduced alcohol craving in anxious alcohol-dependent individuals [16], and reduced neural response to alcohol cues in heavy social drinkers [17]. Oxytocin decreased drinking in nondependent rats [18,19], binge drinking in mice [20,21], and cue-induced reinstatement in postdependent rats [17]. However, oxytocin’s effects on alcohol drinking and motivation for alcohol in currently alcohol-dependent rats remains to be determined. Therefore, in the present study, we used a model of chronic-intermittent exposure to alcohol vapor to induce alcohol dependence and enhanced motivation for alcohol. This model has excellent predictive validity, has allowed dissociation of the neurobiology underlying alcohol dependent versus nondependent behavior, and has been used to determine the likely contribution of a range of neurobiological systems to alcohol use disorder [3,39].
Oxytocin has been detected in the brain following both intraperitoneal and intranasal administration in mice and rats [22–24]. Oxytocin receptors are found in the brain and in various peripheral tissues, where binding mediates oxytocin’s classic role in lactation, parturition, and sexual reflexes in males and females [25]. There are several other oxytocin binding sites in the periphery, including the gut, heart, vascular system, and vagus nerve, which all may contribute to behavioral effects of oxytocin, including inhibition of appetite and fear [25,26]. The central versus peripheral contributions to oxytocin’s putative effects on alcohol drinking also remain to be determined.
Oxytocin receptors are found in many brain regions relevant for alcohol dependence, such as the extended amygdala in rats and humans, including the central nucleus of the amygdala (CeA) [27,28]. The CeA is well described as a key structure involved in the transition to drug and alcohol dependence [2,29]. Silencing of an alcohol-withdrawal–activated CeA neuronal ensemble blocked the enhanced drinking associated with alcohol dependence [30]. The majority of CeA neurons are GABAergic, and increased GABA signaling in the CeA is linked to alcohol dependence. Intra-CeA infusion of a GABAA receptor agonist decreased alcohol drinking in alcohol-dependent rats [31]. We previously demonstrated that spontaneous and evoked GABA transmission is increased in CeA of alcohol-dependent rats [32]. Both pro- and anti-stress neuropeptides interact with CeA GABA signaling and bidirectionally modulate alcohol drinking. Oxytocin has been reported to interact with GABAergic signaling and may block the acute effects of alcohol via a direct action on GABAA receptors [33]. However, the physiological effects of CeA oxytocin in alcohol dependence are currently unknown.
Therefore, in the present study, we tested the hypothesis that oxytocin decreases alcohol drinking and motivation for alcohol specifically in alcohol-dependent rats via central and not peripheral actions. Furthermore, we hypothesized that alcohol dependence would alter how oxytocin modulates GABA signaling in the CeA.
For all behavioral and electrophysiological experiments involving anesthesia, isoflurane inhalation was used. For electrophysiological experiments, anesthesia was followed by rapid decapitation to allow brain slice preparation. All behavioral studies were approved by the Institutional Animal Care and Use Committee (ACUC) of the National Institute on Drug Abuse Intramural Research Program. All electrophysiological procedures were approved by the IACUC of The Scripps Research Institute. All procedures were conducted according to the National Institutes of Health Guide for the Care and Use of Laboratory Animals (8th edition).
Rats used in behavioral (male Wistar, n = 86; Charles River, Kingston, NY) and electrophysiology (male Sprague-Dawley, n = 87; Charles River, Raleigh, NC) experiments weighed 225 to 275 g upon arrival, were group-housed (2 to 3 per cage) in standard plastic cages in a temperature- and humidity-controlled room, and were maintained under a reverse 12 h/12 h light/dark cycle with food and water available ad libitum except during behavioral testing.
For behavioral experiments, alcohol (ethanol; Warner Graham, Cockeysville, MD) was dissolved in tap water. Oxytocin (ChemPep Inc., Wellington, FL) was dissolved in 0.9% saline. For peripheral administration, the non–blood-brain barrier-penetrant oxytocin receptor agonist PF-06655075 was initially dissolved in 10% (v/v) Self Emulsifying Drug Delivery System that consisted of 3:4:3 Miglyol 812:Cremophor RH40:Capmul MCM (v/v/v). The emulsion was completed with 90% (v/v) 50 mM aqueous phosphate buffer (pH 7.4). For intracerebroventricular administration, PF-06655075 was initially dissolved in 5% (v/v) dimethyl sulfoxide, then combined with 5% (v/v) polyethylene glycol 300, and completed with 90% (v/v) saline. PF-06655075 and the vehicle used for its peripheral administration were kindly provided by Pfizer. The non–blood-brain barrier penetrant antagonist L-371,257 (Tocris Bioscience, Ellisville, MO) was prepared in a vehicle of 5% (v/v) dimethyl sulfoxide, 5% (v/v) Cremophor, and 90% (v/v) sterile water.
For electrophysiology experiments, oxytocin and the selective vasopressin receptor 1A antagonist (d(CH2)5,Tyr(Me)2,Arg8)-Vasopressin (TMA) [34] were obtained from Tocris (Ellisville, MO). The selective oxytocin receptor antagonist desGly-NH2-d(CH2)5[D-Tyr2,Thr4]OVT (OTA) was provided by Dr. Maurice Manning (University of Toledo, OH) [34]. CGP 55845A, DL-2-amino-5-phosphonovalerate (DL-AP5), and 6,7-dinitroquinoxaline-2,3-dione (DNQX) were obtained from Tocris (Ellisville, MO). Tetrodotoxin (TTX) was purchased from Biotium (Hayward, CA). Alcohol was purchased from Remet (La Mirada, CA, USA). All drugs were dissolved in artificial cerebrospinal fluid (aCSF).
Oral alcohol self-administration experiments were conducted in standard operant chambers (Med Associates, St. Albans, VT) fitted with 2 retractable levers and a dual-cup liquid receptacle. After initial training, as previously described [29,35], the animals were allowed to lever press for alcohol (10%, w/v; 0.1 ml) and water (0.1 ml) on separate levers according to a concurrent fixed-ratio 1 (FR1) schedule of reinforcement (each lever press resulted in fluid delivery) in 30-min operant sessions. After each training session, the liquid receptacle and surrounding area were inspected to confirm the consumption of earned reinforcers. After response acquisition, rats were split into 2 groups matched by their alcohol consumption. For the remainder of these experiments, rats in the nondependent group were exposed to air without alcohol, whereas rats in the dependent group were exposed to alcohol vapor in daily cycles designed to cause intoxication (14 h vapor “on”; 200 mg/dl target blood alcohol levels) and withdrawal (10 h vapor “off”) to induce alcohol dependence, as previously described [29,35]. Dependence is characterized by somatic and motivational signs of withdrawal that include increased anxiety-like behavior, reward deficits, and enhanced motivation to self-administration of alcohol [36–39]. In all operant alcohol self-administration experiments, steady baseline consumption of alcohol was established in dependent and nondependent rats before pharmacological testing began. Operant alcohol self-administration sessions including pharmacological tests were conducted 2 to 3 sessions per week (never on consecutive days to minimize potential carry-over effects) during the 10 h “off” period, 6 to 8 h into withdrawal.
Intraperitoneal oxytocin (0, 0.125, 0.25, 0.5, and 1 mg/kg; 0.5 or 1 ml/kg) was administered 30 min prior to FR1 alcohol self-administration sessions in dependent (n = 10) and nondependent (n = 10) rats. The pretreatment time was selected based on previous work that determined peak brain oxytocin concentrations following intraperitoneal oxytocin administration [22]. The test order of intraperitoneal doses was counterbalanced using a within-subjects Latin-square design. Self-administration of alcohol and water was recorded during tests. Based on results of the FR1 test, the 0, 0.125, and 0.25 mg/kg intraperitoneal doses were then tested on a progressive ratio (PR) schedule of reinforcement, in which the number of lever presses required to obtain the next alcohol reinforcer increased progressively, as follows: 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 7, 7, 9, 9, 11, 11, 13, 13, etc. The last completed ratio (breakpoint) was used as an indication of motivation for alcohol. During the PR test, only the alcohol lever was made available. Sessions lasted 90 min or until 15 min had elapsed without a response. The pretreatment time during the PR test was the same as for the FR1 test. Again, these doses were administered in a Latin-square design, with intervening nontesting days.
To reestablish a stable baseline of intake, the same rats described above were moved to different operant chambers with matching operant manipulanda and reinforcer receptacles. They were allowed eight 30-min FR1 sessions to obtain alcohol and water (S1 Fig). Next, intranasal oxytocin was administered 1 h prior to testing responding for alcohol on FR1 and PR schedules of reinforcement (described above). The pretreatment time was selected based on previous work that determined peak brain oxytocin concentrations following intranasal oxytocin administration [22]. Again, tests were conducted in Latin-square designs with intervening nontesting days. For intranasal drug delivery, general anesthesia was rapidly induced by isoflurane (5% for 2 to 5 min), and then rats were immediately administered intranasal oxytocin (0, 0.25, 0.5, and 1 mg/kg/20 μl; based on the FR1 data, the intranasal doses of 0.5 and 1 mg/kg/20 μl were selected for testing on the PR schedule) using the rat Precision Olfactory Device (rPOD; Impel NeuroPharma, Seattle, WA) and allowed to recover before being returned to their home cage.
Based on the effects of oxytocin on alcohol drinking and the motivation for alcohol, we selected the doses of intraperitoneal (0.25 mg/kg) and intranasal (1 mg/kg) oxytocin for further behavioral testing. These doses by each route were the lowest dose by each route that significantly reduced alcohol drinking and the motivation for alcohol specifically in dependent rats. The pretreatment times were the same as used for testing on alcohol self-administration.
Behavioral tests were conducted in separate cohorts of rats to assess potential non–alcohol-specific effects of oxytocin on locomotion and grooming in an open field (dependent, n = 6; nondependent, n = 6) and motor coordination on a rotarod (dependent, n = 5; nondependent, n = 7). Two blind observers scored any instances of grooming behavior according to previously published work. A single point was given for each instance of vibration of forepaws, face washing, body grooming, body scratching, paw licking, head shaking, body shaking, and genital grooming, according to previously reported studies [40,41]. Points were totaled for each animal to yield a grooming score.
In addition to alcohol’s pharmacological effect, alcohol contains calories and has a sweet-taste component that contribute to the reinforcer efficacy of alcohol. To assess the potential role of calories and sweet taste in the ability of intraperitoneal and intranasal oxytocin to reduce alcohol consumption, we tested oxytocin on the consumption of a nonalcoholic caloric reinforcer without a sweet taste (5% maltodextrin; n = 7) and on a nonalcoholic sweet solution without caloric content (0.1% saccharin; n = 8; additional details provided in S1 Text).
Three separate cohorts of alcohol-dependent rats were used (central oxytocin administration: n = 6; peripheral oxytocin receptor agonist: n = 9; peripheral oxytocin receptor antagonist combined with intranasal oxytocin administration: n = 12). A subset of the group used for testing the peripheral oxytocin receptor antagonist combined with intranasal oxytocin administration was given 2 sessions in the absence of any treatment to reestablish a baseline and then used to test the effect of central administration of the oxytocin receptor agonist PF-06655075 (central oxytocin receptor agonist: n = 7). The central oxytocin and PF-06655075 administration cohorts were surgically implanted with a guide cannula to allow intracerebroventricular administration of these compounds.
The oxytocin receptor antagonist L-371,257, which does not cross the blood-brain barrier, was tested in combination with intranasal oxytocin administration to test the ability of peripheral antagonism to reverse the ability of intranasal oxytocin to reduce alcohol drinking in dependent rats. L-371,257 is a potent and competitive antagonist of the oxytocin receptor (pA2 = 8.4) with high affinity at both the oxytocin receptor (Ki = 19 nM) and vasopressin V1a receptor (Ki = 3.7 nM) [42]. The pretreatment time and dose (45 min; 5 mg/kg) were selected to be in excess of doses that provided blockade of oxytocin-induced uterine contractions for at least several hours following peripheral (intravenous or intraduodenal) administration in rats [42]. Doses much lower than this in rats (0.5 mg/kg intraperitoneal) [43] and mice (300 μg/kg intranasal) [44,45] have been demonstrated to provide behaviorally effective antagonism. Additionally, in previous studies, comparable doses of L-371,257 versus oxytocin (0.5 mg/kg versus 0.5 mg/kg intraperitoneal in rats [43]; 300 μg/kg versus 200 μg/kg intranasal in mice [45]) were administered. Here, we administered 5 times the dose of the antagonist compared to oxytocin (5 mg/kg versus 1 mg/kg) via a preferred route of peripheral absorption (intraperitoneal injection versus intranasal application [23]) to conservatively test for the ability of the peripherally acting antagonist to alter the actions of intranasal oxytocin. Rats were tested in a within-subjects Latin-square design with the following combinations of intraperitoneal and intranasal treatments: intraperitoneal vehicle with intranasal vehicle, intraperitoneal vehicle with intranasal oxytocin (1 mg/kg/20 μl), intraperitoneal L-371,257 (5 mg/kg/ml) with intranasal vehicle, and intraperitoneal L-371,257 (5 mg/kg/ml) with intranasal oxytocin (1 mg/kg/20 μl). Operant training and induction of dependence were conducted, as described above.
To test the effect of central versus peripheral oxytocin receptor agonism on dependence-induced alcohol drinking, we administered centrally or peripherally a long-acting, large molecule that does not penetrate the blood-brain barrier (PF-06655075). For central administration, 7 rats were surgically implanted with a guide cannula to allow intracerebroventricular administration of PF-06655075. These rats were previously used to test the peripheral oxytocin receptor antagonist combined with intranasal oxytocin administration (data shown in Fig 3B). The PF-06655075 dose of 30 μg was selected for intracerebroventricular administration to match the highest dose of oxytocin administered by the same route, a dose that blocked drinking in alcohol-dependent rats. For peripheral administration, the rats were administered 1 mg/kg of PF-06655075 subcutaneously. This dose was selected taking into account the high plasma protein binding of PF-06655075 (i.e., resulting in lower unbound concentrations than oxytocin) as well as its oxytocin receptor binding Ki. Subcutaneous administration of 1 mg/kg of PF-06655075 versus 1 mg/kg of oxytocin would be expected to produce comparable receptor occupancy (94.2% versus 99.7% at Cmax; see S1 Text for calculations based on data of Modi and colleagues [26]). Therefore, 1 mg/kg of subcutaneous PF-06655075 was expected to recapitulate the putative peripheral binding of oxytocin. This hypothesis was based on the observation that lower doses of intraperitoneal oxytocin were sufficient to block drinking in alcohol-dependent rats in the present study. Specifically, 0.25 mg/kg of intraperitoneal oxytocin reduced alcohol drinking in dependent rats to a similar extent as 30 μg of intracerebroventricular oxytocin. Finally, this systemic dose of 1 mg/kg PF-06655075 has been reported to reduce fear behavior potentially via anti-sympathetic action in the periphery [26].
Dependent and nondependent rats were deeply anesthetized with isoflurane followed by rapid decapitation and immediate removal of the brain into an ice-cold high-sucrose brain slice cutting solution (sucrose 206 mM; KCl 2.5 mM; CaCl2 0.5 mM; MgCl2 7 mM; NaH2PO4 1.2 mM; NaHCO3 26 mM; glucose 5 mM; HEPES 5 mM [pH 7.4]). Coronal slices (300 to 400 μm) containing the CeA were continuously superfused (flow rate of 2 to 4 ml/min) with 95% O2/5% CO2 equilibrated aCSF of the following composition: NaCl 130 mM, KCl 3.5 mM, NaH2PO4 1.25 mM, MgSO4·7H2O 1.5 mM, CaCl2 2.0 mM, NaHCO3 24 mM, and glucose 10 mM. Recordings were performed in neurons from the medial subdivision of the CeA. Each experimental group contained neurons from a minimum of 3 rats. GABAergic activity was pharmacologically isolated with DNQX, DL-AP5, and CGP. All drugs were applied by bath superfusion.
We recorded with sharp micropipettes filled with 3M KCl and evoked GABAergic inhibitory postsynaptic potentials (eIPSPs) by stimulating locally within the medial subdivision of CeA through a bipolar electrode. Neurons were held near their resting membrane potential (−82.4 ± 0.8 mV). We performed an input–output (I/O) protocol consisting of a range of 5 current stimulations, starting at the threshold current required to elicit an eIPSP, up to the strength required to elicit the maximum subthreshold amplitude. The middle stimulus intensity was used to monitor drug-induced changes throughout the duration of the experiment. Paired-pulse ratio (PPR) was performed at the stimulus intensity giving approximately 50% of the maximal amplitude determined in the I/O protocol.
Whole-cell voltage-clamp recordings of GABAergic spontaneous inhibitory postsynaptic currents (sIPSCs) and miniature inhibitory postsynaptic currents (mIPSCs) were from visualized CeA neurons clamped at −60 mV for the duration of the recordings. Patch pipettes (3 to 6 MΩ) were filled with an internal solution composed of the following (in mM): 145 KCl, 0.5 EGTA, 2 MgCl2, 10 HEPES, 2 Na-ATP, and 0.2 Na-GTP. In all experiments, cells with a series resistance greater than 25 MΩ were excluded from analysis, and series resistance was continuously monitored during gap-free recording with a 10-mV pulse. Cells in which series resistance changed more than 25% during the course of the experiment were excluded from analysis. All measures were performed prior to (baseline) and during drug application (details in S1 Text).
Results are presented as the mean ± standard error of the mean. The level of significance was established as p < 0.05. Statistical analyses were performed in Prism 6 (Graphpad Software, Inc., La Jolla, CA). Behavioral data were analyzed with one-way (Dose, Session, or Treatment) repeated-measures analysis of variance (R.M. ANOVA), two-way (Group × Dose) R.M. ANOVA, or by paired-samples t test. The Holms-Sidak test was used for post hoc comparisons. Electrophysiology data were analyzed with two-way ANOVA (Group × Concentration or Group × Treatment) followed by Bonferroni post hoc comparisons, one-sample t test, or independent-samples t test, as appropriate.
Passive exposure to alcohol vapor has been demonstrated to cause somatic signs of dependence in alcohol withdrawal as well as a dysregulation of reward and stress systems. The cardinal feature of the model is the increased consumption and motivation for alcohol exhibited by alcohol-dependent rats when allowed to perform an operant response for access to alcohol [3]. Prior to pharmacological testing, alcohol-dependent rats exposed to chronic-intermittent alcohol vapor exhibited significantly enhanced consumption of alcohol in comparison to nondependent rats that were exposed to air in their home cage throughout the study (S1 Fig).
Oxytocin abolished the difference in alcohol drinking between dependent and nondependent rats at doses of ≥ 0.25 mg/kg (Fig 1). A 2 × 5 (Group × Dose) R.M. ANOVA yielded a significant Group × Dose interaction (F4, 72 = 7.98, p < 0.0001). Post hoc analyses indicated that dependent rats self-administered significantly more alcohol than the nondependent rats following 0 (p < 0.0001) and 0.125 mg/kg (p < 0.001) oxytocin doses, but this difference disappeared at higher doses. Responding was significantly lowered at the 0.25, 0.5, and 1 mg/kg doses (all p < 0.0001) in dependent rats compared with the vehicle condition, whereas only the highest dose of 1 mg/kg significantly lowered lever pressing for alcohol in nondependent rats (p < 0.05). Water intake data during all tests are presented in S2 Fig.
Oxytocin (0.125 or 0.25 mg/kg) selectively eliminated the increased motivation for alcohol in dependent rats (Fig 1B). A 2 × 3 (Group × Dose) R.M. ANOVA yielded a significant Group × Dose interaction (F2, 36 = 3.59, p < 0.05). The difference in PR breakpoint between groups following vehicle treatment was significant (p < 0.05). When treated with either 0.125 or 0.25 mg/kg doses of oxytocin, dependent rats no longer differed from nondependent rats. Oxytocin (0.125 and 0.25 mg/kg) significantly reduced PR breakpoint for alcohol in dependent rats (all p < 0.01), whereas responding in the nondependent group was not altered by oxytocin.
Intranasal oxytocin dose-dependently decreased alcohol drinking in dependent rats without affecting drinking in nondependent rats (Fig 1C). The 2 × 4 (Group × Dose) R.M. ANOVA yielded a significant Group × Dose interaction (F3, 54 = 14.18, p < 0.0001). Dependent rats drank more alcohol than nondependent rats treated with saline, 0.25, and 0.5 mg/kg oxytocin (all p < 0.001). This difference was no longer observed following treatment with 1 mg/kg oxytocin. Alcohol drinking in dependent rats significantly decreased across the doses of oxytocin compared to the saline condition (all p < 0.0001), whereas drinking in the nondependent group was not significantly altered by oxytocin.
On the PR test (Fig 1D), two-way R.M. ANOVA yielded a significant Group × Dose interaction (F2, 36 = 4.27, p < 0.05). Following saline treatment, dependent rats had a significantly higher breakpoint for alcohol than nondependent rats (t54 = 4.50, p < 0.001). This difference was no longer observed following 0.5 or 1 mg/kg oxytocin administration. Oxytocin (1 mg/kg) reduced PR breakpoint in the dependent group (p < 0.001) without altering the behavior of nondependent rats.
Dependent and nondependent rats were not significantly different in their spontaneous locomotion or grooming behavior. Intraperitoneal oxytocin (0.25 mg/kg) significantly decreased locomotion in both dependent and nondependent rats (Fig 2A; Dose: F1, 10 = 14.63, p < 0.01; post hoc tests all p < 0.05), whereas intranasal oxytocin (1 mg/kg) had no effect in either group. Intraperitoneal and intranasal oxytocin treatments did not affect grooming behavior (which remained minimal throughout testing; Fig 2A inset). Dependent and nondependent rats did not significantly differ in their rotarod performance, and this performance was not significantly altered by intraperitoneal or intranasal oxytocin treatments (Fig 2B). Intraperitoneal oxytocin reduced saccharin (t6 = 2.80, p < 0.05) and maltodextrin (t6 = 3.20, p < 0.05) consumption, whereas intranasal oxytocin had no effect (Fig 2C).
Baseline sessions conducted between intracerebroventricular oxytocin test sessions did not significantly differ, so they were combined for analysis (Fig 3A). All doses of intracerebroventricular oxytocin reduced alcohol consumption in dependent rats (Dose: F3, 15 = 10.25, p < 0.001; post hoc tests: all p < 0.05).
Intranasal oxytocin treatment significantly reduced drinking in dependent rats whether the rats were pretreated with either vehicle (VEH + VEH versus VEH + OXT) or with the peripherally restricted oxytocin receptor antagonist L-371,257 (L-371,257 + VEH versus L-371,257 + OXT; Treatment: F3, 33 = 10.73, p < 0.0001; post hoc tests: all p < 0.05). L-371,257 did not significantly alter drinking by itself (VEH + VEH versus L-371,257 + VEH) nor did it reverse the ability of intranasal oxytocin to reduce alcohol drinking in dependent rats (VEH + OXT versus L-371,257 + OXT; Fig 3B).
Baseline sessions conducted between test sessions with intracerebroventricular administration of PF-06655075 did not significantly differ and thus were combined for analysis (Fig 3C). Intracerebroventricular administration of PF-06655075, which does not cross the blood-brain barrier and thus was expected to not diffuse to the periphery, significantly reduced responding for alcohol relative to baseline responding and relative to vehicle (Treatment: F2, 12 = 42.25, p < 0.0001; post hoc tests: all p < 0.0001). Vehicle administration did not significantly alter responding relative to baseline (Fig 3C). Systemic administration of PF-06655075 (expected to not reach the brain) did not significantly alter responding for alcohol relative to baseline responding or the vehicle condition nor did the vehicle itself significantly alter responding relative to baseline (Fig 3D).
We recorded from neurons in the medial subdivision of the CeA, using sharp intracellular whole cell configuration for locally evoked inhibitory GABAergic eIPSP and whole-cell patch-clamp configuration for GABAergic postsynaptic currents. We found no difference in input resistance of neurons from nondependent (154.4 ± 11.4 MΩ) and dependent (159.7 ± 7.6 MΩ) rats and no difference in spike frequency (S3 Fig) between groups. Baseline GABAergic eIPSP amplitudes stimulated locally within the medial subdivision of CeA were not different between nondependent (9.7 ± 0.7 mV) and dependent (9.9 ± 0.7 mV) animals (S4 Fig). No differences were observed in PPR of eIPSPs at 100 ms interstimulus interval in nondependent (1.06 ± 0.08) and dependent (1.01 ± 0.09) neurons.
In CeA neurons of nondependent rats, 100 nM oxytocin (10 to 15 min) did not alter eIPSP amplitudes, whereas 500 nM and 1,000 nM significantly decreased amplitudes to 83.4% ± 4.9% (t10 = 3.40, p < 0.01) and 74.1% ± 7.7% (t4 = 3.37, p < 0.05) of baseline, respectively (Fig 4B). Notably, in dependent rats, oxytocin at all 3 concentrations did not affect eIPSPs. A two-way ANOVA (Group × Concentration) yielded a significant Group effect (F1,44 = 4.55, p < 0.05). Oxytocin did not alter input resistance of nondependent (baseline: 144.3 ± 13.8 MΩ, oxytocin: 148.3 ± 13.7 MΩ) or dependent (baseline: 148.2 ± 10.6 MΩ, oxytocin: 141.8 ± 12.9 MΩ) CeA neurons, suggesting that oxytocin effects on eIPSP amplitudes were not due to excitability changes but due to decreased GABAergic transmission. Oxytocin (500 nM) significantly increased the PPR of eIPSPs (baseline PPR: 0.98 ± 0.11; oxytocin PPR: 1.35 ± 0.14; t10 = 2.56, p < 0.05) in nondependent rats, suggesting a possible decrease in the presynaptic release of GABA (as changes in PPR are inversely related to changes in release) [46]. Oxytocin did not alter PPR in dependent rats (baseline PPR: 1.17 ± 0.22, oxytocin PPR: 1.03 ± 0.07).
Acute alcohol (44 mM; maximal dose [47]) application significantly increased eIPSP amplitudes in CeA of both nondependent (to 140.8% ± 8.5% of baseline; t7 = 4.79, p < 0.01) and dependent rats (to 133.9% ± 13.4% of baseline; t6 = 2.53, p < 0.05; Fig 4C), indicating a lack of tolerance for the acute effects of alcohol on stimulated GABAergic signaling, as previously reported [47]. Additionally, alcohol did not alter input resistance of nondependent (baseline: 169.0 ± 19.2 MΩ, alcohol: 167.6 ± 15.8 MΩ) or dependent (baseline: 167.6 ± 10.4 MΩ, alcohol: 161.4 ± 11.4 MΩ) neurons. Because oxytocin also binds to vasopressin receptors and there is a CeA subpopulation of neurons sensitive to vasopressin binding (specifically, via the V1a but not V1b vasopressin receptor), we always preapplied the selective vasopressin receptor 1A antagonist TMA (500 nM) to isolate specific oxytocin receptor-mediated actions [34,48]. TMA had no effect on eIPSP amplitudes (nondependent: 8.51 ± 1.08 mV and dependent: 9.68 ± 1.31 mV without TMA) and did not affect the magnitude of the alcohol-induced increase of eIPSPs in both nondependent (to 148.6% ± 10.5% of baseline; t5 = 4.64, p < 0.01) and dependent rats (129.8% ± 10.6% of baseline; t6 = 2.80, p < 0.05; S5 Fig).
To determine potential interactions between oxytocin and acute alcohol, in most of the neurons that received 500 nM oxytocin, we applied alcohol in the presence of oxytocin. In this subset of neurons (7 of 9) from nondependent slices, coapplication of alcohol only induced a moderate, nonsignificant increase of eIPSP amplitudes (oxytocin: 83.4% ± 4.9%; oxytocin + alcohol: 101.7% ± 8.3% of baseline; Fig 4C). In the subset of neurons (12 of 15) from dependent slices, alcohol did not further increase eIPSPs (oxytocin: 107.6% ± 9.3%; oxytocin + alcohol: 123.0% ± 16.8% of baseline; Fig 4C), suggesting a blockade of oxytocin-induced decreased GABA transmission in dependent rats. A two-way ANOVA (Group × Treatment) yielded a significant main effect of Treatment (F2,53 = 6.33, p < 0.01), and a Bonferroni post hoc test indicated a significant difference between the effects of alcohol and oxytocin (t53 = 3.35, p < 0.01). Similarly, oxytocin at 100 and 1,000 nM blunted the alcohol-induced facilitation of eIPSPs in both nondependent (84.2% ± 11.0% and 82.1% ± 13.2% of baseline, respectively) and dependent slices (124.0% ± 18.6% and 125.5% ± 15.3% of baseline, respectively). Overall, oxytocin blunted the alcohol-induced increase of GABA transmission (see Fig 4C).
To further investigate the pre- versus postsynaptic action of oxytocin on GABA signaling, we performed whole-cell patch-clamp recordings of sIPSCs and mIPSCs in CeA neurons. Generally, changes in IPSC frequency reflect altered transmitter release, and changes in amplitude or kinetics reflect alterations in postsynaptic GABAA receptor sensitivity. However, altered amplitude may also reflect a mix of pre- and postsynaptic effects [49,50].
We first investigated sIPSCs, finding no significant differences between nondependent and dependent rats in sIPSC frequency (1.1 ± 0.2 and 0.8 ± 0.1 Hz, respectively), amplitude (82.0 ± 8.4 and 66.6 ± 5.8 pA, respectively), rise time (2.6 ± 0.1 and 2.8 ± 0.1 ms, respectively), or decay time (8.0 ± 0.8 and 9.4 ± 1.0 ms, respectively). We next used the vasopressin receptor 1A antagonist TMA and found no effects of TMA alone in either nondependent or dependent rats (S6 Fig), as observed for evoked GABAergic responses. In CeA neurons of nondependent rats, application of 500 nM oxytocin significantly decreased the amplitude of sIPSCs to 85.2% ± 6.7% of baseline (t11 = 2.22, p < 0.05) and increased rise times to 110.5% ± 4.7% of baseline (t11 = 2.24, p < 0.05) with no changes in frequency or decay times (Fig 5A), indicating mainly postsynaptic actions of oxytocin to decrease GABAA receptor function. In dependent rats (Fig 5A), oxytocin also significantly decreased sIPSC amplitude to 83.1% ± 5.0% of baseline (t8 = 3.37, p < 0.01), with no changes in sIPSC frequency or kinetics.
We next investigated mIPSCs in the presence of the sodium channel blocker TTX to isolate action potential independent currents. Generally, mIPSC analysis reveals specific effects of drugs on vesicular release of GABA, which results from exocytosis of neurotransmitter containing vesicles in a manner independent of action-potential–induced release mechanisms. In addition, sIPSCs and mIPSCs likely result from distinct presynaptic neurotransmitter release mechanisms, e.g., vesicle fusion machinery, spatial segregation of the vesicles and/or vesicle populations, and synaptic vesicle pool dynamics [51]. Similar to sIPSCs, we did not find any baseline differences between nondependent and dependent rats in mIPSC frequency (0.8 ± 0.1 and 0.9 ± 0.3 Hz, respectively), amplitude (53.2 ± 4.0 and 60.7 ± 12.3 pA, respectively), rise time (2.5 ± 0.1 and 2.5 ± 0.1 ms, respectively), or decay time (5.5 ± 0.6 and 6.2 ± 0.8 ms, respectively). Oxytocin did not alter mIPSC frequency, amplitude, or kinetics for neurons from nondependent or dependent rats (Fig 5B), suggesting no effect of oxytocin on this form of GABA release in the medial subdivision of CeA.
We next examined the interaction of oxytocin and acute alcohol on CeA sIPSCs in nondependent and dependent rats. In 8 of the 12 CeA neurons from nondependent rats that received 500 nM oxytocin, and 7 of the 9 neurons from dependent rats, we coapplied alcohol (44 mM) and compared effects between groups on sIPSC measures (Fig 5D–5G). A two-way ANOVA (Group × Treatment) yielded a significant effect of Treatment (F1,13 = 4.87, p < 0.05) and a Group × Treatment interaction (F1,13 = 9.84, p < 0.01) for sIPSC frequency (Fig 5D). The Bonferroni post hoc comparisons indicated that alcohol with oxytocin significantly increased sIPSC frequency compared with oxytocin alone in nondependent rats (t13 = 3.91, p < 0.01). Therefore, oxytocin blocked the alcohol-induced increase in GABA release only in CeA neurons of alcohol-dependent rats. Additionally, a two-way ANOVA (Group × Treatment) yielded a significant Group × Treatment interaction (F1,13 = 20.88, p < 0.001) for sIPSC decay time (Fig 5G). The Bonferroni post hoc comparisons indicated that alcohol with oxytocin significantly increased decay times compared with oxytocin alone in nondependent rats (t13 = 4.86, p < 0.001).
As previously shown [52], we found that in nondependent and dependent rats, acute alcohol alone significantly increased sIPSC frequency to 145.5% ± 13.4% of baseline (t5 = 3.40, p < 0.05) and 131.4% ± 8.6% of baseline (t9 = 3.65, p < 0.01), respectively (Fig 5H), with no changes in sIPSC amplitude or kinetics, suggesting increased action-potential–dependent GABA release. We compared the effects of alcohol alone with the effects of alcohol and oxytocin using two-way ANOVAs (Group × Treatment) for each sIPSC measure (Fig 5H–5K). We found a significant main effect of Treatment (F1,27 = 9.00, p < 0.01) and a significant Group × Treatment interaction (F1,27 = 5.61, p < 0.05) for sIPSC rise times (Fig 5J). The Bonferroni post hoc comparisons indicated a significant difference in the effects of alcohol with oxytocin versus alcohol alone for nondependent neurons.
Finally, we found that the selective oxytocin receptor antagonist desGly-NH2-d(CH2)5[D-Tyr2,Thr4]OVT alone did not alter sIPSCs (S7 Fig), suggesting no basal activity of these receptors on spontaneous GABAergic transmission in dependent rats. Notably, in neurons from dependent rats, oxytocin in the presence of OTA had no effect on sIPSCs when compared to baseline (Fig 5L and 5M). In a subset of these neurons (8 out of 11), we applied alcohol in the presence of both OTA and oxytocin, resulting in a significant increase in sIPSC frequency (t7 = 3.01, p < 0.05; Fig 5M). We found a similar effect of OTA to rescue the alcohol effect on eIPSPs from nondependent neurons (t7 = 2.70, p < 0.05; S8 Fig). These results confirm that oxytocin receptor antagonism blocked postsynaptic effects of oxytocin and restored the acute alcohol-induced facilitation of GABA release in the CeA of alcohol-dependent rats.
Oxytocin delivered via intraperitoneal, intranasal, and intracerebroventricular routes blocked the enhanced motivation for alcohol drinking that developed in alcohol-dependent rats. Intraperitoneal and intranasal oxytocin at certain doses blocked the increased alcohol consumption and motivation for alcohol in dependent rats, without impacting these behaviors in nondependent rats. Intranasal oxytocin did not disrupt spontaneous locomotion, grooming behavior, motor coordination, or consumption of sweet or caloric palatable solutions. Central administration of oxytocin produced a dose-dependent reduction in alcohol drinking in dependent rats, and this effect was replicated by central administration of an oxytocin receptor agonist (PF-06655075) that does not cross the blood-brain barrier and therefore was expected to not diffuse to the periphery. However, peripheral administration of PF-06655075 (expected to not reach the brain) had no effect. Additionally, peripheral administration of an oxytocin receptor antagonist that does not cross the blood-brain barrier did not reverse the ability of intranasal oxytocin to reduce alcohol drinking in dependent rats. Together, the data suggest a central mechanism for oxytocin’s actions on alcohol drinking. Ex vivo electrophysiology recordings indicated that oxytocin inhibits spontaneous action-potential–dependent GABAergic transmission in CeA slices from both dependent and nondependent rats. However, oxytocin’s ability to dose-dependently reduce evoked network GABAergic activity was absent in slices from dependent rats. Furthermore, oxytocin blunted acute alcohol-induced facilitation of evoked GABAergic responses in both nondependent and dependent rats but blocked alcohol-induced facilitation of spontaneous action-potential–dependent GABA release only in CeA slices from alcohol-dependent rats, suggesting differential GABA network effects in nondependent and dependent rats produced by oxytocin.
Systemic oxytocin blocked the enhanced motivation for alcohol observed in dependent rats, indexed by escalated alcohol drinking and increased breakpoint in a PR test, at doses that did not change the behavior of nondependent rats. These findings are consistent with previous reports that oxytocin can decrease alcohol drinking in nondependent mice and rats [20,21] and that oxytocin is particularly effective in decreasing cue-induced reinstatement of alcohol seeking in postdependent rats compared with nondependent rats [17]. The present findings are unique in several ways from previous work in rats [17]. They demonstrate the following: (1) Oxytocin can reduce alcohol drinking and the enhanced motivation to “work” for alcohol in currently alcohol-dependent rats. This difference is critical, because it provides a novel indication for oxytocin’s potential therapeutic value. Alcohol drinking during acute (i.e., currently alcohol-dependent rats) and protracted abstinence (i.e., alcohol seeking in rats with a history of alcohol dependence) are proposed to model distinct phases and aspects of alcohol use disorder, with distinct underlying neurocircuitry [2]. The results of the present study indicate that oxytocin may have the potential to reduce heavy drinking in moderate to severe alcohol use disorder. (2) Oxytocin’s anti-drinking effects in rats can be achieved by intranasal administration, using a device designed to allow noninvasive drug delivery across the blood-brain barrier [53]. (3) Intranasal administration reduced alcohol consumption and motivation for alcohol in dependent rats without causing nonspecific locomotor, grooming, motor coordination, and consumption of nonalcoholic sweet or caloric palatable solutions, suggesting that oxytocin’s effects on alcohol drinking are specific to the pharmacological effects of alcohol. (4) The present study provides evidence that peripheral receptors did not make a major contribution to the effect of intranasal oxytocin. More specifically, intracerebroventricular administration of a large molecule (PF-06655075) that does not cross the blood-brain barrier or intracerebroventricular administration of oxytocin significantly reduced alcohol drinking in dependent rats. However, PF-06655075 administered systemically at a dose matching the highest tested dose of systemic oxytocin (4 times the dose sufficient to block drinking in alcohol-dependent rats) did not have an effect. Therefore, oxytocin’s effect of reducing alcohol drinking in the present model of alcohol dependence is likely centrally mediated. Further supporting this hypothesis, application of the nonbrain penetrant antagonist L-371,257 did not reverse the ability of intranasal oxytocin to reduce alcohol drinking in alcohol-dependent rats, suggesting that intranasal administration of oxytocin reduces alcohol drinking in dependent rats independent of oxytocin receptor binding in the periphery.
Despite the precise mechanism remaining unknown, there is accumulating evidence that peripherally applied oxytocin can cross the blood-brain barrier in adult rodents. As suggested by others [22,23], there is likely a mechanism for direct transport of intranasally applied oxytocin across the blood-brain barrier in adult rodents. Tanaka and colleagues applied oxytocin by intravenous, intraperitoneal, and intranasal routes in rats to determine the peripheral and central levels of oxytocin resulting from each route of administration [23]. The authors used the intraperitoneal administration data to determine the amount of centrally detected oxytocin that can be expected to result from peripherally circulating levels. They used this information to account for centrally detected oxytocin following intranasal administration that may be the result of oxytocin transferring first from the nasal epithelium into the peripheral circulation, then into the brain. It was concluded that, under the conditions of intranasal administration, >95% of the oxytocin detected in the brain is expected to result from direct transport across the blood-brain barrier [22,23]. Indeed, oxytocin detected in the brain was higher following intranasal administration compared with intravenous or intraperitoneal administration, despite intranasal administration resulting in far lower oxytocin plasma concentrations. Furthermore, oxytocin concentrations were especially high in the olfactory bulb after intranasal administration, suggesting a direct route of entry [23]. Bustion and colleagues demonstrated that radio-labeled oxytocin administered in mice intranasally was detected throughout the brain, i.e., dissociating exogenously applied deuterated oxytocin from endogenous oxytocin, thus confirming direct entry [24]. Again, particularly elevated levels were detected in the olfactory bulb, suggesting a point of entry at the nasal epithelium.
Although intraperitoneal oxytocin administration decreased locomotion in the open field and the consumption of a nonalcoholic, noncaloric sweet solution and a nonalcoholic, unsweetened caloric solution, intranasal administration did not. Neither intraperitoneal nor intranasal oxytocin administration altered grooming behavior in the open field or performance in the rotarod task, indicating that the intraperitoneal effect on locomotion is likely not a result of altered motor coordination. A possible explanation for nonspecific effects of intraperitoneal oxytocin administration is that that oxytocin may have “side effects” that result from binding at peripheral sites (e.g., gut, heart, vascular system, and/or vagus nerve) that disrupt behavior in general and that were induced following intraperitoneal oxytocin administration in the present study. Congruent with this account is the observation that intraperitoneal oxytocin suppressed water consumption during tests for oxytocin’s effect on alcohol consumption. In contrast, oxytocin treatment did not alter water consumption when administered intranasally, intracerebroventricularly, or intranasally in combination with L-371,257, suggesting that peripheral “side effects” were avoided during these tests (S2 Fig). Intraperitoneal administration of oxytocin, unlike the peripherally restricted agonist PF-06655075, was able to block alcohol drinking in dependent rats, similarly to intranasal oxytocin administration. Although the transfer of peripherally circulating oxytocin to the central compartment is expected to be very limited, a rapid, dramatic spike in plasma-oxytocin concentrations has been noted following intraperitoneal administration that may allow pharmacologically relevant central concentrations of oxytocin to be achieved, especially following application of supraphysiological oxytocin doses, as used in the present study [18,20,22,23,54]. A recent study found deuterated oxytocin administered intravenously was detected centrally in rhesus macaques (i.e., dissociating exogenous labeled oxytocin from endogenous oxytocin, confirming direct transfer [55]). Although intraperitoneal administration of oxytocin induced some side effects in the present study, the action of intraperitoneal oxytocin that causes reduction of alcohol drinking in dependent rats is likely the same central mechanism as engaged following intranasal administration. Nevertheless, the present data suggest that administration by intranasal compared with intraperitoneal route may result in more favorable pharmacokinetics for achieving central over peripheral oxytocin exposure [22,23].
To further test the hypothesis of central oxytocin action, we demonstrated that intracerebroventricular infusion of oxytocin and PF-06655075 (expected to not diffuse to the periphery) reduced alcohol drinking in dependent rats. Systemic administration of PF-06655075 (expected to not to cross the blood-brain barrier) did not affect alcohol intake in dependent rats. Finally, the peripherally restricted oxytocin receptor antagonist L-371,257 did not alter the effects of intranasal oxytocin in reducing alcohol drinking in dependent rats. Together, these data suggest that peripheral receptors make a minimal contribution to intranasal oxytocin’s effects on alcohol drinking.
Next, we examined the effects of oxytocin on GABAergic transmission in the CeA, a key brain region of dysregulation in alcohol dependence [3,56–59]. To gain mechanistic insight into the contribution of the oxytocin system on CeA GABAergic transmission in the context of acute and chronic alcohol, here, we examined both spontaneous action-potential–dependent and evoked GABAergic transmission wherein the network activity is intact, as well as action-potential–independent transmission. We found that oxytocin decreases GABA signaling in the CeA of both dependent and nondependent rats, but its effects varied between different modes of GABAergic transmission. In alcohol-nondependent rats, oxytocin decreased evoked GABA responses by decreasing evoked GABA release, an effect that is no longer observed in dependent rats, suggesting neuroadaptations of the CeA oxytocin system in dependence. In contrast, oxytocin decreased GABAA receptor function in both nondependent and dependent rats. Similar to previous studies in the medial subdivision of CeA [48], oxytocin did not affect action-potential–independent GABAergic transmission in either group. Evoked GABA responses are generated by delivering a controlled electrical stimulation locally within the CeA, whereas spontaneous events reflect inhibitory signaling across the broader CeA synaptic network. Miniature currents result from spontaneous presynaptic vesicle fusion independent of sodium entry and more precisely determine pre- and postsynaptic drug effects occurring at the terminals. In addition, spontaneous and evoked forms of GABA transmission may represent distinct forms of GABA transmission, resulting from distinct presynaptic neurotransmitter release mechanisms (e.g., vesicle fusion machinery, spatial segregation of the vesicles and/or vesicle populations, synaptic vesicle pool dynamics) [51]. Of note, oxytocin had no effect on mIPSCs or cellular excitability, therefore its effects on sIPSC amplitude and rise time, as well as in eIPSP amplitude and PPR, suggest synaptic effects rather than cell excitability. These synaptic effects likely occur upstream within the local intact GABAergic network that is shared by spontaneous and evoked forms of synaptic transmission, whereas miniature effects are more localized to the specific terminals in the medial subdivision of the CeA where oxytocin receptors may not be present [48]. These considerations are critical and need to be taken into account when comparing with oxytocin electrophysiological effects reported on by others in the lateral subdivision of the CeA [48,60,61]. In contrast, these lateral CeA studies reported effects of oxytocin in increasing excitability of lateral CeA GABAergic neurons that project to the medial CeA, resulting in an increase in sIPSC frequency and decreased excitability in the medial CeA [48]. Although the lack of oxytocin effect on mIPSCs in the medial CeA is consistent with our results, we observed that oxytocin decreased GABAergic transmission in the medial CeA without changing cellular excitability. There are two likely reasons for these discrepancies. The first is the use of the native peptide oxytocin in our studies as opposed to the selective peptide oxytocin-receptor agonist (Thr⁴,Gly⁷)-Oxytocin (TGOT) [34]. Although we pharmacologically block the predominant vasopressin receptor in the CeA, native versus selective peptide agonists for the oxytocin receptor may exhibit functional selectivity at the oxytocin receptor or off-target effects. The second possibility is the heterogeneity and interconnectivity of GABAergic neurons in the CeA. The medial CeA contains GABAergic neurons that project out of the CeA as well as synapse locally. In one previous study [48] of the TGOT responsive cells, only about half of the TGOT inhibited neurons were excited by vasopressin. Additionally, Viviani and colleagues reported that projection-specific populations of medial CeA GABAergic neurons were unresponsive to TGOT [60]. Therefore, it is possible that the majority of the neurons from which we recorded were from a subpopulation of GABA neurons that did not receive direct innervation of oxytocin-excited lateral CeA GABAergic neurons. Additionally, from our evoked experiments, the electrical stimulation likely excited neurons in the lateral and medial CeA, and our responses were a composite of these effects. Therefore, future experiments will determine specific cell types to understand the distinct oxytocin effects on the different components of the synaptic network.
Understanding the physiological actions of oxytocin in the CeA may benefit by comparison to the pro-stress, pro-drinking effects of corticotropin-releasing factor (CRF) and the anti-stress, anti-drinking effects of neuropeptide Y (NPY). The pro-stress neuropeptide CRF is elevated in the CeA, likely mediated by glucocorticoid receptors [62,63] in alcohol dependence, and infusion of CRF1 or glucocorticoid receptor antagonists in the CeA suppress drinking specifically in alcohol-dependent rats [38,62,64–66]. In contrast, the anti-stress neuropeptide NPY is decreased in the CeA in alcohol dependence, and intra-CeA infusion of NPY suppresses alcohol drinking specifically in dependent rats [67,68]. We have reported that CRF, similar to alcohol, robustly enhances GABAergic transmission in CeA of rats [38,66,68]. However, NPY and nociceptin decreased presynaptic GABA release in CeA, normalizing the enhanced GABAergic transmission observed in alcohol dependence [68]. Both NPY and nociceptin also block the presynaptic acute alcohol-induced facilitation of CeA GABA release, whereas oxytocin blocked GABA release in dependent rats only. We previously reported that alcohol may act presynaptically through voltage-gated calcium channels to increase action-potential–dependent GABA release in the CeA and that alcohol dependence disrupts this mechanism, shifting alcohol actions to CRF1 receptors [52]. This may explain why oxytocin has no effect on eIPSPs in dependent animals if alcohol dependence dysregulates calcium channels and oxytocin effects work through action-potential–induced calcium-dependent mechanisms. Additionally, oxytocin, via action at oxytocin receptors, may interfere with the CRF1 receptor-mediated mechanism in dependent rats to blunt the effects of alcohol. Therefore, oxytocin’s actions in the CeA are similar but not identical to anti-stress neuropeptides like NPY [67]. Oxytocin’s actions are particularly complex given the pre- and postsynaptic interactions with alcohol. Future studies will be needed to investigate the functional role of oxytocin in its interactions with other pro-stress and anti-stress systems in the CeA.
One important limitation of the present work is that the studies were conducted exclusively in male rats, especially considering sex-specific oxytocin distribution and behavioral mediation [69,70] and sex differences in alcohol drinking [35,71]. It will be critical to test the effect of oxytocin in female subjects to determine the extent to which the present conclusions can be generalized to females. Second, the behavior and electrophysiology experiments were performed in Wistar and Sprague Dawley rats, respectively. Note that, in general, we find the different rat strains to be better suited to one set of experiments or the other (Wistar for behavior and Sprague-Dawley for electrophysiology). There may be strain differences in the sensitivity of rats to oxytocin treatment (e.g., MacFadyen and colleagues reported that 0.1 mg/kg oxytocin reduced drinking in nondependent Sprague-Dawley rats [18]). However, we included internal controls in behavioral and electrophysiological experiments and have previously shown that baseline CeA synaptic activity and effects of drugs are similar between these strains [72]. As such, we do not find problems conceptually linking the data sets. Lastly, although we tested the oxytocin effects on CeA GABAergic signaling in the presence of a V1a antagonist to rule out the contribution of CeA vasopressin receptors in oxytocin’s effects on CeA GABAergic signaling, the contribution of vasopressin receptors to the observed behavioral effects were not evaluated in the present study. There is evidence suggesting a role of central vasopressin receptors in social and aggressive behaviors, and affective states, including anxiety-like behavior [54]. We would expect that agonism of these receptors by oxytocin would have pro-stress, pro-drinking effects rather than contributing to a reduction in alcohol drinking in dependent rats in this model. Consistent with this hypothesis, Edwards and colleagues reported that V1b antagonism in dependent rats decreased their alcohol drinking [73]. Similar effects have also been reported in Sardinian alcohol-preferring rats [74] and alcohol-dependent humans [75].
In summary, the present study reports that oxytocin via central rather than peripheral action reduced alcohol consumption and motivation for alcohol in an animal model of alcohol dependence. These effects may result from dependence-induced alterations in oxytocin and GABA systems in reward- and stress-related extrahypothalamic brain regions such as the extended amygdala. Intranasal oxytocin administration appears to be advantageous in terms of specificity in reducing alcohol-motivated behavior compared with intraperitoneal administration. Therefore, the present work highlights the oxytocin system as a target for understanding the plasticity of brain stress and anti-stress systems in the etiology of alcohol use disorders. Targeting this system, possibly by intranasal administration, may provide novel pharmaceutical interventions for the treatment of alcohol use disorder.
|
10.1371/journal.pcbi.1005426 | Patient-specific modeling of individual sickle cell behavior under transient hypoxia | Sickle cell disease (SCD) is a highly complex genetic blood disorder in which red blood cells (RBC) exhibit heterogeneous morphology changes and decreased deformability. We employ a kinetic model for cell morphological sickling that invokes parameters derived from patient-specific data. This model is used to investigate the dynamics of individual sickle cells in a capillary-like microenvironment in order to address various mechanisms associated with SCD. We show that all RBCs, both hypoxia-unaffected and hypoxia-affected ones, regularly pass through microgates under oxygenated state. However, the hypoxia-affected cells undergo sickling which significantly alters cell dynamics. In particular, the dense and rigid sickle RBCs are obstructed thereby clogging blood flow while the less dense and deformable ones are capable of circumnavigating dead (trapped) cells ahead of them by choosing a serpentine path. Informed by recent experiments involving microfluidics that provide in vitro quantitative information on cell dynamics under transient hypoxia conditions, we have performed detailed computational simulations of alterations to cell behavior in response to morphological changes and membrane stiffening. Our model reveals that SCD exhibits substantial heterogeneity even within a particular density-fractionated subpopulation. These findings provide unique insights into how individual sickle cells move through capillaries under transient hypoxic conditions, and offer novel possibilities for designing effective therapeutic interventions for SCD.
| Sickle cell disease is a genetic blood disease that causes vaso-occlusive pain crises. Here, we investigate the individual sickle cell behavior under controlled hypoxic conditions through patient-specific predictive computational simulations that are informed by companion microfluidic experiments. We identify the different dynamic behavior between individual sickle RBCs and normal ones in microfluidic flow, and analyze the hypoxia-induced alteration in individual cell behavior and single-cell capillary obstruction under physiological conditions.
| In research investigations of hematological disorders, most experiments are performed on groups of cells with the underlying assumption that all of the cells in a particular type are identical. However, recent evidence reveals that individual cells within the same population may differ drastically in size, shape, mechanical properties and protein levels, and these variations can have important consequences for the health and biological function of the entire cell population [1]. A representative case is sickle cell disease (SCD), one of the most common inherited genetic blood disorders affecting more than 270,000 new patients each year [2, 3].
SCD has been characterized as the first molecular disease [4], being linked to the mutation of a single nucleotide in the hemoglobin molecule. The primary pathophysiological event in SCD is the polymerization of sickle hemoglobin (HbS) into long fibers upon deoxygenation (DeOxy) [5, 6]. The fibers distort RBCs into irregular and heterogeneous shapes—e.g. granular, elongated, oval, and crescent (classic sickle) shapes [7, 8]. The hypoxia-affected RBCs are also heterogeneous in their cell density in a range of less than 30 g/dL to more than 46 g/dL [9], which are usually fractioned into four arbitrary cell density subpopulations (fractions I-IV) in vitro analysis [7]. Heterogeneous cell fractions engender heterogeneity in cell rigidity [10–13]. These hypoxia-affected RBCs are more sticky and stiff, causing frequent painful episodes of vaso-occlusion and depriving oxygen from tissues and organs [10,14]. The decrease in RBC deformability contributes to impaired blood flow and other pathophysiological origins of the disease. However, the clinical expression of SCD is heterogeneous, as the hypoxia-affected RBCs do not all behave in the same way all the time, and the variance is considerable even within a same density-fraction, making it hard to predict the risk of a vaso-occlusive crisis [15–17]. This poses a serious challenge for disease management.
Precision medicine [18], which accounts for individual variability, is an emerging approach for treatment and prevention of disease [19,20]. Developing such an approach, however, inevitably requires resolving various heterogeneity-related issues, both at whole cell population and single-cell levels [21]. Although developments in quantitative, in vitro microfluidic assays provide greater understanding of cell dynamics under hypoxic conditions in patients’ blood samples [16], there is a critical need to develop single-cell level assays to assess sickle RBC behavior under transient hypoxic conditions for better therapeutic interventions [22]. These considerations lead to the motivation for the present work whose aim is to address the following question: To what extent does morphological sickling affect cell dynamics in microcapillaries under transient hypoxia, with consequences for hemodynamics and vaso-occlusion?
In order to gain a better understanding of vaso-occlusive crisis, it is necessary to obtain direct and real-time observations of the traversal individual sickle RBCs through microcapillaries and of alterations in cell biodynamics and biorheology in response to controlled changes in oxygen (O2) concentration. However, owing to the phenotypic heterogeneity of SCD, existing experimental capabilities do not readily provide this information. As a result, there is a critical need to develop predictive, patient-specific cell models to quantify alterations in cell biodynamics under transient hypoxic conditions. Such models could be validated by recourse to a variety of well-controlled and independent in vitro experiments.
In this article, we investigate the dynamic behavior of individual sickle cells in real time, in a capillary-like microenvironment, and perform quantitative analysis of hypoxia-induced alteration in cell behavior and response to obstruction of capillaries by combining predictive simulations with controlled and quantitative information obtained from microfluidic experiments.
De-identified SCD blood samples from two patients with HU therapy (on-HU) and two patients without HU therapy (off-HU) were selected for this study, following institutional review board (IRB) approvals from the National Institutes of Health (NIH) and Massachusetts Institute of Technology (MIT). All samples were collected into EDTA anticoagulant and stored at 4°C during shipping and storage. Table 1 shows selected hematologic and hemorheologic parameters in these four blood samples. The experiments were performed exactly in the same way as those in Ref. [23], but with a specific focus on (1) correlations between cell shape, mean corpuscular hemoglobin concentration (MCHC), transit velocity and trajectories through the microgates, and (2) movements of individual cells with and without disturbances from adjacent blockages.
After washing twice with PBS (Thermo Scientific) at 821 × g for 5 min at 21°C, the RBCs in the blood sample were separated into four density fractions, using a stepwise gradient medium prepared with OptiPrep medium (Sigma Aldrich, Saint Louis, MO, USA) and Dulbecco’s PBS (HyClone Laboratories, Inc., South Logan, UT, USA). The estimated MCHC values were 27.3, 30.9, 34.9, and > 45.0 g/dL for fraction I-IV, respectively (Table 2). Fraction I (SS1, reticulocyte rich) and fraction II (SS2, discocyte rich) have moderate MCHC values, which are similar to those of healthy RBCs. Fractions III (SS3) and IV (SS4) mainly comprised rigid discocytes and irreversible sickle cells (ISCs), respectively, with their MCHC values considerably higher than those of healthy RBCs. The fractionated RBCs were washed with PBS and re-suspended in RPMI-1640 medium with 1% (wt/vol) Bovine Serum Albumin (Sigma-Aldrich, Saint Louis, MO, USA) for cell sickling measurement in a polydimethylsiloxane (PDMS)-based microfluidic hypoxia assay [23]. This platform provided measurements of cell sickling under controlled O2 concentrations at 37°C, including a fully Oxy state (20 vol% O2), and a hypoxic condition, in which the O2 concentration decreased from 20 vol% to below 5 vol% within 15 s and maintained at 2 vol% for the rest periods. Patient-specific data relevant to the present work along with cell morphological data are provided in Table A in S1 Text for predictive simulations.
In order to investigate the behavior of sickle RBCs in each density-fractionated subpopulation under transient hypoxic conditions, we have developed a unified modeling framework based on dissipative particle dynamics (DPD). Our computational framework is predicated on recent developments [23] in microfluidics that quantify in vitro single RBC dynamic response and its heterogeneity under controlled oxygen partial pressures in blood samples from SCD patients. Specifically, the geometry of microfluidic channel (Fig 1) and profile of transient hypoxia (Fig 1 and Figure A in S1 Text) are the same as in the experimental setup [23]. The solid walls of the microfluidic channel in DPD simulations are modeled by layers of frozen DPD particles, and a force-adaptive is employed for fluid particles to control their density fluctuations [24]. An external body force is exerted on each fluid particle to generate a flow in the microfluidic channel. In this study, the externally applied body force works in the direction of flow (x-direction) on the fluid particles.
We studied the dynamics of sickle RBCs under transient hypoxia using a multiscale RBC model [25]. For completeness, the model is briefly summarized below, whereas details of this MS-RBC model are available in Refs. [25,26]. Here, the membrane of MS-RBC is modeled by a two-dimensional triangulated network with Nv vertices, which are connected by Ns viscoelastic bonds to impose proper membrane mechanics. In this study, the elastic energy of the bond is taken as
Vs=∑j∈1…,Ns[kBTlm(3xj2−2xj3)4p(1−xj)+kp(n−1)ljn−1],
(1)
where lj and lm are the equilibrium (initial edge) length and maximum extension of spring j, p is the persistence length, kp is a constant factor of the spring, and n is an exponent. The bending energy of the RBC membrane is written as
Vb=∑j∈1…,Nskb[1−cos(θj−θ0)],
(2)
where kb is the bending constant, θj and θ0 are the instantaneous and spontaneous angles between two adjacent triangles with a shared common edge j. In addition, the volume and surface area of RBC are controlled to mimic the incompressible interior fluid and the area-preserving cell membrane. The corresponding energy is taken as
Va+v=∑j∈1…,Ntkd(Aj−A0)22A0+ka(A−A0tot)22A0tot+kv(V−V0tot)22V0tot,
(3)
where ka and kd are the global and local area constraint coefficients, kv is the volume constraint coefficients, A0 is the triangle area, and the terms A0tot and V0tot are the total area and volume in equilibrium, respectively.
The RBC membrane interacts with the fluid and wall particles through DPD forces, and the system temperature is maintained by the DPD thermostat. The surrounding external fluids and cytosol are modeled by collections of free coarse-grained particles and their separation is enforced through bounce-back reflections at the moving surface of RBC membrane. In addition, we consider a positive correlation between the number of intracellular coarse-grained particles and the MCHC value, i.e., for a denser RBC with a higher MCHC value, we include more coarse-grained particles inside the cell. The MS-RBC model has been validated by comparisons with a number of available experiments that examine the mechanics and biorheology of healthy and diseased RBCs [25–31]. Such connections between simulations and experiments have also indicated that the results developed here are not dependent on the details of refinement of the simulation in that they are independent of the level of coarse-graining for Nv ≥ 500 [25,26]. In this study, we employ a MS-RBC model with a level of coarse-graining (Nv = 500) that facilitates computationally efficient simulations of RBCs in microfluidic channel.
Sickle RBCs undergo various morphological transitions, due to the polymerization of intracellular HbS molecules, from the normal biconcave shape to an irregular sickled shape, in the form of granular, elongated and crescent shapes. Instead of directly modeling the formation of HbS polymer fibers and the resulting cell morphology, we consider an “effective surface tension” (stretching force) applied on the cell membrane to mimic the cell distortion exerted by the growing HbS polymer domain [31,32]. We develop a two-step RBC morphological sickling model, as described below.
1) We first apply surface tension on the cell membrane to transform the discocyte RBC into a granular, elongated or crescent shape. Here, we define the directions along the two long axes of RBC as x- and y-directions, and the third direction, perpendicular to the long axes, as the z-direction. We choose four anchor points located at the maximum/minimum values in the x/y directions (Fig 1), where the intracellular polymer fibers can potentially interact with the red cell membrane. Different combinations of the stretching force applied on the anchor points cause different sickle cell shapes: when the stretching force is exerted on two diametrically opposite anchor points (for example, points A and C), an elongated or a crescent cell shape is obtained. Similarly, if the stretching force is exerted on all the four anchor points, a granular shape results. The values of stretching forces for obtaining different shapes of sickle RBCs are shown in Table B in S1 Text. We define the distorted shape as the equilibrium state of the sickle RBC with minimum free energy. Specifically, we adjust the equilibrium length, lj, of each spring j to the edge length, lj,REF, of the final distorted state to eliminate local stress on the cell membrane.
2) Subsequently, we employ the kinetic description for sickling RBCs [33] to model the cell morphological sickling process under transient hypoxia. Here, we define the two parameters, lj and lj,REF, as the reference length of the spring j, respectively, in the RBC under normal Oxy state (~20 vol% O2) and fully DeOxy state (~2 vol% O2). Once the oxygen partial pressure, pO2, is lower than a critical value pO2,c for cell sickling, we assume that a sufficiently large stretching force, fj(t), is applied on each spring so that its instantaneous edge length, lj(t), is essentially equal to the expected edge length,
lj,ref(t)=lj{1−[lj,REFlj−1][pO2,c−pO2(t−tD)pO2,c]},
(4)
where tD is delay time for cell sickling. In the present study, the parameters of pO2,c, pO2(t), and tD are directly obtained or calculated from experiments [23]. Through this approach, we are able to dynamically adjust the bond lengths of the elastic springs to affect the sickling and unsickling processes of RBCs while maintaining the cell membrane free of any total external force.
Membrane stiffening of sickle RBCs is closely related to their morphologic change [11]. It seems likely that a step increase in rigidity occurs at the same time as morphological sickling. Similarly, the shear modulus, μref(t), of the hypoxia-affected cell membrane is taken as
μref(t)=μ0{1−[μREFμ0−1][pO2,c−pO2(t−tD)pO2,c]},
(5)
where μREF is the shear modulus of sickle RBC, described below. The cell unsickling process is analogous to the cell sickling process, by considering the delay time (tD,R) of cell unsickling. This kinetic cell sickling model, which accounts for (1) cell morphological change, (2) cell sickling and unsickling delay times, and (3) cell membrane stiffening under transient hypoxia, offers an effective method of investigating the sickling and unsickling processes of RBCs in response to changes in O2 concentration under both static and flow conditions (S1 Video).
In this study, we modeled a healthy RBC using the multiscale RBC model with the following parameters: number of RBC vertices Nv = 500; RBC area A0 = 135.2 μm2 and volume V0 = 92.4 μm3; RBC membrane bending modulus kc,0 = 2.4 × 10−19 J; shear modulus μ0 = 4.7 pN∙μm-1; cytosol viscosity η0 = 1.2 cp. In contrast to a healthy RBC, the SCD RBC is characterized by extensive impairment in deformability, the extent of which is dependent on the cell density and oxygen partial pressure. Sickle RBCs in different cell density subpopulations exhibit different membrane elasticity in the Oxy state (O2 concentration ~ 20 vol%) and the DeOxy state (O2 concentration < 5 vol%). Previous studies have shown that the intracellular polymerization of HbS upon DeOxy leads to significant increases in cytosol viscosity and membrane elasticity [34–39]. The cytosol viscosity, ηcytosol, under DeOxy state could be two orders of magnitude greater than η0 [34,39]. In our previous sensitivity study, we have shown that the blood dynamics is nearly independent of cytosol viscosity if ηcytosol > 50η0 [32]. We would expect that the cytosol viscosity also plays a major role in determining sickle cell behavior in a capillary-like microenvironment under transient hypoxia. However, we find the cytosol viscosity in the densest cell fractions IV is only around 4.5 η0, because our model does not explicitly include the intracellular HbS polymer fibers. Here, we used an effective shear modulus of RBCs, which accounts for the effects of (1) membrane stiffening of sickle RBCs, (2) high membrane tension induced by the growth of intracellular HbS polymers, and (3) elevated cytosol viscosity. In the Oxy state, we select experimentally determined shear modulus data [35–37], and set the effective shear modulus to μREF = 1.0μ0, 1.2μ0, 1.5μ0 and 3.0μ0 for RBCs in fractions I, II, III and IV, respectively. In the DeOxy state, we consider four distinct types of sickle RBCs with different cell membrane mechanical properties based on experimental observations [11,40]. For an SS1 deformable cell with a low MCHC value, the measured shear modulus increased by up to 10 times after the occurrence of sickling [11], and hence we set the effective shear modulus μREF = (5–10) μ0. For an SS2 cell with a mild MCHC value, in which the effective shear modulus is increased by one or two orders of magnitude compared to that of healthy RBCs [11,40], we set μREF = (50–100) μ0. For an ISC of the SS4 type, its effective shear modulus is at least two or three orders of magnitude greater than the value of healthy RBCs [11], so we set μREF = (1000–2000) μ0. For a SS3 rigid cell with a higher MCHC value, we set μREF = (250–500) μ0. In addition, for the hypoxia-unaffected RBCs in each cell density fraction, we assume comparable deformability with the population in the same cell density fraction under the Oxy state. Similar parameters have been used as inputs for DPD simulations to quantify the shear-independent rheological behavior of blood flow in SCD [32]. We also investigated the functional dependence of shear viscosity of sickle RBC suspensions on the effective shear modulus, and demonstrated that the sickle RBC suspension behaves as a Newtonian fluid only if the effective shear modulus of sickle RBCs increases by two or three orders of magnitude [31]. The bending rigidity kc, following prior work [31,32], is kept constant in the model as the hemoglobin concentration increased.
The cell sickling delay time due to intracellular HbS polymerization is demonstrated to be the primary determinant of clinical severity in SCD [41]. The delay time of cell sickling is extraordinarily sensitive to solution conditions, particularly to HbS concentration. From prior results obtained from in vitro microfluidics experiment on sickle RBCs [23] and in vivo studies [41], we set the mean tD ≈ 8.7 s, 19.8 s and 23.6 s for granular, elongated and classic sickle shaped RBCs, respectively, for the off-HU group blood samples. In the microfluidic experiments, the delay time for sickling for the on-HU group blood samples varied from 28 to 100 s [23], which is much longer than that for the off-HU group. Based on the experimental results, here we set the mean tD to be in the same range for the on-HU group. The cell unsickling delay time, tD,R, is also directly obtained from the microfluidic experiments. According to Du et al. [23], the cell unsickling process after reoxygenation (ReOxy) was much faster (< 20 s) than the cell sickling process, and the delay time distribution of cell unsickling was not significantly different between the two different groups. Based on the experimental results, here we set the mean tD,R to be in the range of 10–15 s.
All simulations were carried out with a representative volume of 180 μm × 60 μm × 5 μm that comprised a total of 30 periodic obstacles with a fluid particle number density of 3 obtained by dividing the fluid particle number of the model system by its volume. Capillary obstruction statistics were collected by running 200 independent simulations and the collective behavior is found to be independent of the initial position and initial orientation of RBCs. For the case of flow under Oxy state (without obstruction), steady state is reached with an average velocity of ~ 120 μm/s. The simulations are performed using the USERMESO functional package that was written based on LAMMPS [42]. The time integration of the motion equations is computed through a modified velocity–Verlet algorithm with λ = 0.5 and time step Δt = 0.001 τ where τ is a characteristic time in DPD units. A typical simulation performed in the current study involves one million time steps and a computing time, on average, of about 20,000 CPU core hours on the Blue Gene/P system at the Argonne Leadership Computing Facility (ALCF).
Individual sickle cells show marked heterogeneity, and the variance is considerable even within the same density-fraction. This leads to the question: how do individual sickle RBCs from different density-fraction behave differently from healthy ones when they travel though microcapillaries under transient hypoxic conditions? Here, we investigate the motion of individual RBCs flowing in a capillary-inspired microchannel that consists of parallel periodic obstacles forming 15-μm-long, 4-μm-wide and 5-μm-high microgates. Fig 2 and S2 Video in supplementary material, show typical dynamic motion of RBCs travelling through the microfluidic channel under cyclic hypoxia. This figure and the accompanying video reveal that all RBCs are easily deformed and pass readily through the microgates. Following a decrease in O2 concentration, some RBCs become sickled and get trapped at the microgates; other RBCs can still deform and squeeze through the microgates (Fig 2). When hypoxia continues for a prolonged period, more RBCs become sickled and are obstructed at the microgates. After the O2 levels are restored, these rigid RBCs remain trapped for a few seconds due to a delay in the cell unsickling process before regaining their deformability; after this delay period, they once again easily traverse the microgates (Fig 2). The simulated individual cell behavior under controlled hypoxia conditions is in qualitative agreement with experimental observations [23]. In addition, we calculated the transit velocity of individual sickle RBCs under transient hypoxia and compared them to experimentally measured data, see Fig 2. We found that the transit velocities obtained from experiments and simulations are mutually consistent, under both Oxy and DeOxy states.
A direct means of triggering the early stages of a vaso-occlusive event is possible by following trajectories of sickle RBCs under controlled transient hypoxia at the single-cell level. For the purpose of illustration and to help with the analysis, we track the trajectories of individual sickle RBCs flowing in the capillary-inspired microchannel, both from patient-specific predictive simulations and companion microfluidic experiments, see Fig 3. These figures show that the sickle RBCs have two different types of motion, i.e., translational and flipping motion, before they arrest at the microgates. These different types of motion may lead to transient or even permanent occlusion of the microcapillary. We observe that a sickle RBC favors moving along the flow direction (translational motion) when it undergoes cell sickling (Fig 3 and corresponding S3–S5 Videos). Specifically, for elongated and crescent-shaped (classic sickle) cells, more than three quarters of the cells follow the translational motion. The flow of an RBC approaching the entrance of a microgate is disturbed because of the size mismatch between the cell and microgate. While a deformable cell is able to squeeze through the microgate, a stiffened (sickled) cell loses its ability to deform dynamically. Instead, it undergoes a continuous rotation until its long axis is nearly parallel to the flow, allowing a part of the cell to enter into the microgate. These cells are usually arrested at the microgates in a parallel manner, i.e., the sickle RBC tends to align with flow streamlines, as shown in Fig 3. Our observations show that such a blockage is initially transient (Fig 3; S3–S5 Videos), due to a relatively smaller contact area of the trapped sickle RBC than the cross-sectional area of the microgates. The trapped sickle RBC may move slowly through the microgates (S3 Video), or eventually stop at the microgates, causing persistent obstruction to RBC flow (S4 and S5 Videos), if the sickle RBCs become stiff, e.g., an ISC in fraction IV. Interestingly, if a sickle RBC moves alongside with other RBCs, while travelling or being trapped, it may also experience considerable rotation because it is subject to a velocity gradient even when the sickle RBC aligns with flow streamlines. Such a cell is more likely to flip, leading to occasionally longitudinal or vertical blockage (sickled RBC aligns perpendicular to flow direction) (Fig 3 and corresponding S3 Video and S6 Video). It should be noted that the vertical blockage is much more likely to become persistent blockage, with more serious implications for vaso-occlusion in SCD patients.
We also track the dynamic behavior of individual sickle RBCs at different density-fractionated subpopulations under controlled transient hypoxic conditions. Deformable sickle RBCs in fractions I and II appear to take a preferred path (if the adjacent microgates in the flow direction are not fully blocked): they twist and turn along a serpentine path once they spot trapped cells ahead of them, see Fig 4. However, the stiff sickle RBCs in fractions III and IV just flow toward the trapped sickle RBCs and eventually stop nearby (S7 and S8 Videos). The difference in cell behavior between deformable and stiff sickle RBCs is probably due to the chaotic motion of RBCs caused by cell-cell interactions and flow perturbations near the obstructions. For a deformable sickle RBC, fluid flow can change the cell shape substantially and drift it along the flow direction; however, a stiff sickle RBC is also disturbed by the fluid flow near the obstruction.
SCD exhibits heterogeneous morphologies, which depend on the DeOxy rate [10]. Gradual DeOxy is known to result in predominantly elongated- and crescent-shaped RBCs, whereas rapid DeOxy results in less distorted granular-shaped RBCs. Cell morphological sickling is thus identified by visibly changes in cell shape and texture associated with transient hypoxic conditions. We performed computational simulations of patient-based samples with different cell shape count information under the Oxy and DeOxy states. We calculated the transit velocity of sickle RBCs in the microfluidic channel and compared the results to those measured in experiments (Fig 5). The results show that the RBC shape plays an important role in cell traversal through microgates: with the same cell rigidity values under the Oxy state, sickle RBCs with a disc-shape have the lowest transit velocity (v ~ 98 μm/s). By contrast, sickle RBCs with a crescent shape have the highest transit velocity (v ~ 136 μm/s). Therefore, the transit velocity of sickle cells with different shapes is statistically significantly different. The reason for this might stem from the different cross-sectional areas of sickle RBCs with different shapes. In our simulations, the cross-sectional area of sickle RBCs with disc, oval, and crescent shapes are about 32 μm2, 26 μm2, 23 μm2, respectively, which are greater than that of the microgate opening (20 μm2). Thus, the sickle RBC has to deform during its traversal through the microgates. When we applied a fixed pressure gradient in each simulation to force the cells to cross the microgate, the granular-shaped ones suffer the largest deformation, resulting in the lowest transit velocity. The simulation results, albeit counter-intuitive, are quantitatively consistent with the experimental observations of sickle RBCs transiting in the microfluidic channel.
It is known that the loss of deformability of RBCs play a crucial role in vaso-occlusion in SCD [11,14,40]. We perform simulations of sickle RBCs in different density-fractionated subpopulations in order to verify the significant role of cell deformability in determining the dynamic and rheological characteristics of individual sickle RBCs. Here, we consider three distinct types of sickle RBCs under channel flow, i.e., the hypoxia-affected RBCs in fractions I (SS1), II (SS2), and IV (SS4). Our simulations indicate that sickle RBCs show increased flow resistance under the DeOxy state, especially for the sickle RBCs in fraction IV; some sickle RBCs (for example, the granular and disc-shaped ones) in this fraction always are obstructed at the microgates (Fig 5). In addition, considering the sickle RBCs in the same density-fractioned subpopulation, (hence with nearly the same shear modulus), the present results also show an obvious difference in transit velocity for sickle RBCs with different shapes. This confirms the hypothesis that SCD exhibits substantial heterogeneity even within the same density-fractionated subpopulation.
There is currently no universal cure for SCD patients, but symptoms can be managed with fluids, oxygen and medication. HU has a number of characteristics of an ideal drug for SCD patients [43,44]. It increases HbF production and reduces the occurrence of sickling-related complications [45], partially attributed to improved cell hydration [46] and cell deformability [47,48]. However, HU treatment is also associated with an elevated MCV [49,50]. This dual influence of HU on RBC structural and mechanical properties may significantly affect cell traversal through the microgates. In order to develop a quantitative assessment of the efficacy of HU treatment on the dynamic behavior of individual sickle RBCs, we examine how the enlarged sickle RBCs move in shear flow.
We study the dynamic behavior of individual sickle RBCs using patient-specific hematological values for individual patients with SCD after treated with HU. When the MCV values are compared between the off-HU and on-HU groups, we find that they are always higher in the on-HU groups (see the MCV values in Table 1), consistent with previous observations [49,50]. Our simulations indicate that sickle RBCs travelling through the microgates are sensitive to MCV, and that the correlation between cell transit velocity and MCV is enhanced with the decrease in the size of the microgates. As shown in Fig 5, an increase in MCV from 83 fl to 103 fl results in a slight decrease in cell transit velocity when individual sickle RBCs travel through 4-μm-wide microgates. The average cell transient velocity decreases significantly, however, when the individual sickle RBCs pass through 3-μm-wide microgates. These results reveal the importance of MCV as a determinant of individual sickle RBCs passage through the smallest capillary.
The hypoxia-affected RBCs in SCD cause blockages at the microgates under the DeOxy state. Hence, we performed simulations of patient-based samples to predict the single-cell capillary obstruction and compare the results against experimental data [23].
The capillary obstruction ratio, is defined as the ratio of the total number of trapped sickle RBCs at the microgates to the total number of RBCs in the microfluidic channel during the DeOxy state. This ratio is now examined from the four SCD blood samples including on-HU and off-HU groups. Fig 6 shows the values of capillary obstruction ratio obtained from DPD simulations and experiments. In general, the off-HU group exhibits a significantly higher capillary obstruction ratio than the on-HU group. Such increases are likely caused primarily by changes in sickled fraction. As shown in Table A in S1 Text, 20.2% and 39.0% of RBCs sickle in the off-HU/S-P-I group and off-HU/S-P-II group, respectively, among which 15.8% and 39.0% of hypoxia-affected RBCs fall within cell density fractions III and IV, respectively. In the on-HU groups, fewer RBCs become sickled than those in the on-HU groups, i.e., there are only 9.4% and 3.1% of sickled RBCs in the on-HU/S-P-III group and the on-HU/S-P-IV group, respectively, among which only 6.3% and 3.1% of hypoxia-affected RBCs fall within cell density fractions III and IV, respectively. As we demonstrated in an earlier section, the sickled RBCs show increased flow resistance under the DeOxy state and the densest ones are usually unable to traverse the microgates. Thus, a relatively large number of sickle RBCs, especially in the denser cell fractions, can cause a marked increase in single-cell capillary obstruction. The results confirm that cell deformability plays a key role in RBC traversal in small microfluidic channels. In addition, we find that the predictions of all DPD simulation cases fall within the range of experimental data. The discrepancy between predictions and experiments could arise from an underestimation of the cell morphological sickling count of individual sickle RBCs.
In summary, in this paper, we demonstrate the unique capabilities benefits of combining dynamic microfluidic experiments with multiscale simulations for characterizing the complex behavior of individual sickle RBCs in a capillary-like microenvironment under transient hypoxia. We show that hypoxia-affected RBCs undergoing sickling significantly alter cell behavior. We also monitor the dynamic behavior of both hypoxia-affected and hypoxia-unaffected RBCs as they travel through the capillary-like microenvironment under cyclic hypoxia. Taken together, these experiments and corresponding systematic particle-based simulations elucidate the effects of irregular geometry, decreased cell deformability, and elevated cell volume on the biodynamic behavior of individual sickle RBCs and their roles in single-cell capillary obstruction. They provide a quantitative measure of the heterogeneity associated with SCD at the single-cell level. Hence, the particle-based simulations and comparisons with available independent experiments offer a powerful means for real-time monitoring of in vitro behavior of individual sickle RBCs under controlled transient hypoxic conditions, providing an objective way of assessing the effectiveness of targeted drug therapy aimed at easing or preventing vaso-occlusive crisis associated with SCD.
Our model does not explicitly include the intracellular HbS polymer fibers. Hence, it does not model the interaction between cell membrane and polymer fibers, and the potential influences on morphological distortion of RBCs and the attendant alteration in their mechanical properties. This deficiency could be addressed in future work by recourse to a hybrid model that encompasses the molecular and cellular scales by combing the MS-RBC model with particle-based HbS polymer models developed recently [51–53]. In addition, it would require further computational validation and extensive testing of these patient-specific models against clinical and experimental studies to make future predictions more reliable. Such simulations from these patient-specific predictive models would be useful for testing known biomarkers and discovering new biomarkers.
|
10.1371/journal.ppat.1000617 | MUC1 Limits Helicobacter pylori Infection both by Steric Hindrance and by Acting as a Releasable Decoy | The bacterium Helicobacter pylori can cause peptic ulcer disease, gastric adenocarcinoma and MALT lymphoma. The cell-surface mucin MUC1 is a large glycoprotein which is highly expressed on the mucosal surface and limits the density of H. pylori in a murine infection model. We now demonstrate that by using the BabA and SabA adhesins, H. pylori bind MUC1 isolated from human gastric cells and MUC1 shed into gastric juice. Both H. pylori carrying these adhesins, and beads coated with MUC1 antibodies, induced shedding of MUC1 from MKN7 human gastric epithelial cells, and shed MUC1 was found bound to H. pylori. Shedding of MUC1 from non-infected cells was not mediated by the known MUC1 sheddases ADAM17 and MMP-14. However, knockdown of MMP-14 partially affected MUC1 release early in infection, whereas ADAM17 had no effect. Thus, it is likely that shedding is mediated both by proteases and by disassociation of the non-covalent interaction between the α- and β-subunits. H. pylori bound more readily to MUC1 depleted cells even when the bacteria lacked the BabA and SabA adhesins, showing that MUC1 inhibits attachment even when bacteria cannot bind to the mucin. Bacteria lacking both the BabA and SabA adhesins caused less apoptosis in MKN7 cells than wild-type bacteria, having a greater effect than deletion of the CagA pathogenicity gene. Deficiency of MUC1/Muc1 resulted in increased epithelial cell apoptosis, both in MKN7 cells in vitro, and in H. pylori infected mice. Thus, MUC1 protects the epithelium from non-MUC1 binding bacteria by inhibiting adhesion to the cell surface by steric hindrance, and from MUC1-binding bacteria by acting as a releasable decoy.
| The bacterium Helicobacter pylori can cause peptic ulcer disease, gastric adenocarcinoma and MALT lymphoma. H. pylori colonize the mucosal surface of the stomach, where adherence helps the bacteria to remain in the neutral and protected niche under the mucus layer, and helps it withstand the continuous mucus washing of the mucosal surface. Adherence is also thought to mediate much of the H. pylori mediated disease. The cell-surface mucin MUC1 is highly expressed on the mucosal surface and limits the density of H. pylori in a murine infection model. We now demonstrate that the majority of H. pylori strains can bind to human MUC1 and that release of MUC1 following binding limits adhesion to the cell surface. Furthermore, MUC1 protects the epithelium from non-MUC1 binding bacteria by acting as a physical barrier to adhesion to other cell surface molecules. Thus, appropriate expression and function of MUC1 is likely to limit development of disease ensuing from chronic H. pylori infection.
| The bacterium Helicobacter pylori can cause peptic ulcer disease, gastric adenocarcinoma and MALT lymphoma [1]. H. pylori is estimated to cause approximately 70% of all gastric cancers, and gastric cancer is the second most common cause of cancer related deaths. H. pylori infection and the H. pylori induced pathologies, chronic atrophic gastritis and gastric cancer, are all associated with an increase in epithelial apoptosis [2],[3],[4]. One mechanism by which H. pylori can induce apoptosis is by the delivery of the protein CagA into epithelial cells by a type IV secretion system [5],[6]. This process subsequently activates multiple intracellular signaling cascades inducing an apoptotic response [5],[6] that has been suggested to promote gastric carcinogenesis by a compensatory increase in gastric epithelial cell proliferation [4]. Supporting this notion, there are more proliferating cells in inflamed mucosa under H. pylori infestation than in H. pylori free areas of the mucosa [7]. Furthermore, it has been shown that in response to chronic Helicobacter felis infection in mice, bone marrow–derived cells can home to and repopulate the gastric mucosa, replacing dead or exhausted epithelial stem cells and contribute over time to metaplasia, dysplasia, and cancer [8].
Adherence of some H. pylori to the mucosal surface is likely to help the bacterial population remain in the neutral and protected niche under the mucus layer, and help it withstand the continuous mucus washing of the mucosal surface. Adherence of H. pylori is dependent on the expression of bacterial adhesins and cognate host glycans, displayed by glycoproteins and glycosphingolipids in gastric epithelium and also by mucins in the gastric mucus layer [9],[10],[11]. Several adhesins have been implicated in H. pylori binding: the Blood group Antigen Binding Adhesin (BabA) binds to the fucosylated ABO/Lewis b antigen (Leb), and the Sialic Acid Binding Adhesin (SabA) binds to the sialyl-Lewis x (sLex) and sialyl-Lewis a antigens (sLea) [12],[13]. In the human stomach, the Leb blood-group antigen is mainly expressed by the surface epithelium and on the MUC5AC secreted mucin [11]. Expression of sialylated Le antigens are common in infected and inflamed gastric mucosa [14],[15],[16].
Under the mucus layer, the cell-surface associated mucins are highly expressed glycoproteins on the apical surface of all mucosal epithelial cells. Because of their long filamentous nature, cell surface mucins are likely to be the first point of direct contact between host tissue and organisms that penetrate the secreted mucus layer. MUC1 is the most highly expressed cell surface mucin in the stomach [17]. Most mucins exhibit considerable genetic polymorphism due to variability in their numbers of tandem repeat peptides, which results in proteins of widely divergent lengths. Several studies have linked MUC1 polymorphisms with susceptibility to H. pylori-induced disease, such as gastritis and gastric cancer [18],[19], suggesting a direct effect of MUC1 polymorphisms on the development of Helicobacter-associated pathology. We have shown that mice deficient in Muc1 are more susceptible to infection by both H. pylori [20] and Campylobacter jejuni [21]. However, the mechanism by which this mucin limits bacterial pathogenesis has not been elucidated. In this study, we have characterized mechanisms by which MUC1 limits H. pylori colonization and decreases ensuing mucosal pathology.
The human gastric epithelial cell line MKN7 forms a contiguous polarized monolayer (epithelial resistance 250 ohm/cm2) and expresses MUC1 on the apical membrane surface as occurs in normal gastric epithelium (Figure 1A). Using double determinant ELISA we measured MUC1 in MKN7 cells and human gastric juice (Figure 1B), demonstrating that MUC1 is shed from the gastric cell surface in vivo. MUC1 from the MKN7 cells carried the Leb and sLea oligosaccharide structures (Figure 1C, D) but not sLex (Figure 1E) demonstrating that MKN7 MUC1 possesses ligands for both the SabA and BabA H. pylori adhesins. The glycosylation of MUC1 in gastric juice varied between individuals. Shed MUC1 from all three individuals tested carried Leb (Figure 1C), MUC1 from one individual carried sLex (Figure 1E) but none of the three gastric juice samples we investigated carried sLea on MUC1 (Figure 1D). This is consistent with interindividual differences in glycosylation demonstrated for secreted gastric mucins such as MUC5AC [11].
H. pylori strains differ in their expression of the BabA and SabA adhesins and consequently in their ability to bind to the Leb, sLea and sLex carbohydrate structures expressed on gastric mucins [11],[13]. The J99 wild type strain possesses both adhesins, and bound to the adhesion ligands Leb and sLex (Figure 2A). J99 bound to MUC1 isolated from the human gastric epithelial cell line MKN7 (Figure 2B) as well as to MUC1 shed into human gastric juice in both samples that were tested (Figure 2C), as analyzed using a double determinant ELISA. This demonstrates that H. pylori can bind to both cell associated and shed MUC1. In contrast, when an isogenic J99 double mutant lacking both BabA and SabA was used, no specific attachment of the bacteria to MUC1 was detected (Figure 2B and C). A single J99 SabA deletion mutant resulted in a small non-significant decrease in MUC1 binding and a single J99 BabA deletion mutant showed intermediate MUC1 binding (Figure 2B). This demonstrates that binding to MUC1 occurred via MUC1 oligosaccharides bound by both the BabA and SabA adhesins. Since the healthy stomach normally contains a very low amount of sialylated structures, further binding studies were performed with Leb positive and sLex negative MUC1 obtained from a healthy gastric surgical resection specimen. The BabA positive H. pylori strains P466 and CCUG17875/Leb also bound to MUC1, whereas the BabA negative strains 75ΔBabA1/A2, CCUG17874, 26695, P1, P1-140 and M019 did not (Figure S1). Thus, non-MUC1 binding H. pylori strains also exist. However, 80% of H. pylori strains are BabA positive [12], and the majority of H. pylori strains thus have the ability to bind to MUC1.
The demonstration that H. pylori binds to MUC1 and that MUC1 can be shed from the epithelial surface indicated that MUC1 could act as a releasable decoy upon bacterial adherence. To determine whether MUC1 release can be triggered simply by particulate binding, in the absence of microbial molecular signalling, we used 1 µm beads coated with either an antibody against the extracellular domain of MUC1 or with an isotype control antibody. Immunohistochemistry of cell layers incubated with the beads demonstrated that the MUC1-antibody-coated beads attached to the cell surface but that they were not endocytosed (data not shown). Adherence of beads coated with the anti-MUC1 antibody to confluent MKN7 epithelial cells was significantly higher at 24 h than at 4.5 h and then decreased (Figure 3A). This pattern is consistent with a loss of beads from the cell surface once the MUC1 binding sites on the beads are filled and/or the MUC1 on the epithelial cell surface is depleted. The amount of beads in the conditioned culture supernatant followed the opposite pattern, demonstrating that the decrease of beads attached to the cells is due to release into the supernatant rather than bleaching of the fluorescence (Figure 3B). Furthermore, the amount of cellular MUC1 α-subunit (extracellular domain) was decreased in cells treated with anti-MUC1 coated beads compared to isotype control coated beads, as demonstrated by capture ELISA (Figure 3C). This was not due to the adherent beads sterically hindering antibody access to MUC1 on the cell surface, as the same decrease was shown by Western blotting of cell lysates (Figure 3D). MUC1 was also detected in the bead-containing conditioned media from cells treated with anti-MUC1 coated beads, but not with isotype control coated beads (Figure 3D). These results demonstrate that when inert bacterial sized particles bind to MUC1, the MUC1 extracellular domain is released from the epithelial cell surface.
Co-cultures of confluent MKN7 cells and H. pylori were established in microaerobic conditions at 1∶1–1∶20 (mammalian cell∶bacteria:) as a model of the interaction between the gastric epithelium and H. pylori that penetrate gastric mucus. Under microaerobic conditions the mammalian cells are not stressed and the H. pylori survive, proliferate and show a closer physiology to that seen in vivo than the typically used aerobic co-cultures [22]. The use of confluent cultures is important in modelling the gastric epithelium to avoid H. pylori binding to the basolateral surface of cells, as occurs with the typically used non-confluent AGS culture system [23]. MUC1 is polarised to the apical membrane in these MKN7 cultures (Figure 1A). After 44 h of co-culture we used flow cytometry to measure the amount of cell surface and total MUC1 within individual viable MKN7 cells. Co-culture with the H. pylori strain J99wt depleted MKN7 cells of approximately 40% of both the total pool of the extracellular domain of MUC1 (Figure 4A) and of the cell surface located extracellular domain of MUC1 (Figure 4B). To explore whether H. pylori SabA or BabA adhesin mediated binding to MUC1 was required for depletion of MUC1 we used the J99ΔBabAΔSabA mutant as the single mutants were able to bind MUC1 (Figure 2B). Co-culture with the double adhesin mutant did not deplete cellular or cell surface MUC1. These results are consistent with decreased MUC1 production, increased degradation in the cell or release of MUC1 following binding of the live MUC1-binding bacteria in a similar manner to that shown with the inert 1 µm beads.
Further experiments over 1–24 h showed that the decrease in cellular MUC1 began early in co-culture (Figure 5A; cellular MUC1 was lower in infected cells at 1, 6, 12 and 24 h). However, MUC1 mRNA levels increased progressively in culture but were not significantly affected by the presence of H. pylori (Figure 5B). The rapid decrease in MUC1 protein with sustained levels of MUC1 mRNA, suggest that loss of protein rather than decreased MUC1 production is the likely explanation for MUC1 depletion from the cell. Despite the unchanged mRNA levels and the decreasing cellular MUC1, the amount of free MUC1 in the conditioned medium was significantly lower (by about 40%) in infected cultures at 1, 2, 4 and 6 h (Figure 5C). Decreased MUC1 in the culture medium in infected cultures suggests that substantial amounts of MUC1 was either bound to the bacteria (which were removed by centrifugation prior to ELISA) and/or was degraded. In a further experiment, total and cell surface MUC1 was assessed by flow cytometry after 6 and 24 h of co-culture. After 6 h of co-culture H. pylori strain J99 wt did not affect the level of MUC1 on the cell surface, however, it led to depletion of 44% of cell surface MUC1 after 24 h infection compared to uninfected MKN7 cells (Figure 5D). Bacteria recovered from the culture medium were stained for the presence of MUC1 using an antibody reactive with the extracellular domain. MUC1 was detected bound to H. pylori strain J99wt but not to the J99ΔBabAΔSabA mutant using both flow cytometry (Figure 5E) and confocal microscopy (Figure 5F–I, Figure S2). Binding to H. pylori strain J99wt was greatest at 6h (Figure 5E) suggesting the bacteria may be capable of degrading or shedding bound MUC1. Confocal microscopy showed that MUC1 binding to bacteria was often focally intense rather than evenly distributed on the bacterial surface (Figures 5F,G). Taken together these data are consistent with progressive shedding of MUC1 from the cell surface and binding of shed MUC1 to bacteria dependent on the BabA and SabA adhesins.
Release of MUC1 can occur either by (a) breakdown of the non-covalent association of the α- and β-subunits, possibly in response to shear stress or conformational changes, and/or (b) action of extracellular proteases/sheddases. The proteases ADAM17 and MMP-14 have previously been shown to act as MUC1 sheddases in endometrial cells [24],[25]. As both of these proteases are expressed by MKN7 cells we repeated co-culture experiments with and without knockdown of expression of either and both ADAM17 and MMP-14 using siRNA. Successful knockdown of mRNA expression of the proteases was achieved over a time period from 8 h (Figures 6A,B) to 24 h (data not shown) after infection. Cellular MUC1 decreased with infection at 24 h and free MUC1 in the culture medium also decreased at 8 h which is consistent with other experiments (presented as the proportion of uninfected controls for all siRNA conditions in Figure 6C and D, respectively). There were no significant alterations in the amount of cellular MUC1 in either uninfected or infected MKN7 cells following knockdown of ADAM 17 or MMP-14. However, there were significant increases in the amount of shed MUC1 following knockdown of MMP-14 at 8 but not 24 h of infection, although no changes were seen following knockdown of ADAM17 (Figure 6D). These data indicate that MMP-14 is partially involved in the MUC1 shedding process in MKN7 cells.
Transfection of MUC1 siRNAs into MKN7 epithelial cells reduced cell surface MUC1 expression by 80% with the 1∶1 siRNA and by 85% with the 1∶3 siRNA, compared to scrambled siRNA (Figure 7A). H. pylori J99wt, J99ΔCagA or J99ΔBabAΔSabA (8×105 CFU H. pylori/well) were co-cultured with confluent transfected MKN7 monolayers for 0.5, 2, 4.5 and 20 h. While the J99wt and J99ΔCagA strains bound MKN7 cells to a similar level, binding of the J99ΔBabAΔSabA adhesin mutant was reduced at all time points (as determined by staining the co-cultures with an anti-H. pylori antibody, Figure 7B–E; p<0.001). After 30 min of co-culture, binding of J99 (both with and without the SabA and BabA adhesins) to MKN7 cells was higher when MUC1 was depleted by siRNA (Figure 7B). By 2 h, and through to 20 h, siRNA knockdown of MUC1 had no significant effect on adhesion to the epithelial cells of the J99wt and J99ΔCagA strains that bind MUC1 via the SabA and BabA adhesins (see Figure 2). In contrast, MUC1 protection against adhesion of the J99ΔBabAΔSabA strain (which did not bind MUC1, see Figure 2) to the epithelial cells was maintained to at least 20 h. When considered together with the changes in MUC1 expression shown in Figures 4 and 5, these results indicate that: (a) while binding via the BabA and/or SabA adhesins increases H. pylori adhesion to gastric epithelial cells, these bacteria can bind epithelial cells via mechanisms other than these adhesins; (b) during early infection, MUC1 inhibits H. pylori binding to epithelial cells occurring via both adhesin-dependent and -independent mechanisms; (c) as infection progresses MUC1 is depleted from the epithelial cell surface of cells infected with H. pylori strains carrying the BabA and SabA adhesins and is no longer effective in inhibiting attachment; and (d) in cells that are infected with H. pylori strains that cannot bind to MUC1, MUC1 continues to be protective against adhesion to the epithelial cells, most likely by steric hindrance of bacterial interactions with non-MUC1 receptors expressed on the epithelial cell surface.
To analyze cell death we used flow cytometry in combination with annexinV/7-AAD staining which can distinguish healthy cells, cells in early apoptosis, late apoptosis and necrosis/very late apoptosis. The extent of H. pylori induced apoptosis in MKN7 cells was dependent on the concentration of bacteria in the co-cultures. In MKN7 cells co-cultured for 20 h with 6×105 CFU/mL H. pylori (MOI 1∶1), J99wt induced 35% early apoptosis, 8% late apoptosis and 5% necrosis as determined by annexinV/7AAD staining. J99ΔCagA induced similar levels of early apoptosis, but no late apoptosis or necrosis suggesting a delayed induction of apoptosis, while J99ΔBabAΔSabA had no affect on the viability of MKN7 cells (Figure 8A–D, left columns). At a 10-fold higher concentration (6×106 H. pylori CFU/mL, MOI 10∶1), all the isogenic strains induced more apoptosis and necrosis compared with the lower concentration of H. pylori, albeit in a similar pattern; J99wt induced more cell death than J99ΔCagA, and J99ΔCagA induced more cell death than J99ΔBabAΔSabA (Figure 8A–D, right panel). Thus, both BabA/SabA mediated adhesion to the epithelial cell and CagA contribute to epithelial cell death, but to get the maximal effect of CagA, H. pylori needs to bind to the epithelial cell via the BabA and SabA adhesins.
In an experiment to evaluate the importance of MUC1 in H. pylori induced apoptosis, transfection of siRNAs into MKN7 cells reduced MUC1 levels by 93% and 97% with the 1∶1 and 1∶3 sequences respectively, compared to scrambled siRNA. Cultures with depleted MUC1 had fewer healthy cells and more apoptotic cells, irrespective of whether or not they were infected (Figure 9A–D). However, the proportion of necrotic cells was slightly lower in the cells with reduced MUC1.
Finally, to ascertain whether MUC1/Muc1 can also influence H. pylori induced epithelial apoptosis in vivo we analysed apoptosis in wild-type (Muc1+/+) and Muc1−/− mice infected with H. pylori-SS1 as previously described [20]. While no difference was observed 1 week post-infection, Muc1−/− mice had a 3-fold increase in the mean proportion of TUNEL-positive apoptotic cells in their gastric mucosa after 8 weeks of H. pylori infection, compared to Muc1+/+ mice (Figures 9E and S3). Thus, MUC1/Muc1 protects against apoptosis both in vitro and in a murine H. pylori infection model.
Previously we have shown that mice deficient in Muc1 are more susceptible to infection by H. pylori [20] both with regard to the level of colonisation and the degree of pathology that develops. We have now extended these observations by examining the mechanism by which this mucin can limit infection by H. pylori. Here we demonstrate that MUC1 protects the epithelial cell from both bacterial adhesion and apoptosis. When the pathogen does not bind to MUC1, the 200–500 nm long extracellular domain of the mucin appears capable of physically distancing the bacteria from the host cell surface, thus sterically inhibiting adhesion to other potential cell surface ligands. When the pathogen does bind to MUC1, the extracellular domain of the mucin is released from the epithelial surface, thereby acting as a releasable decoy and preventing prolonged adherence. In the process H. pylori are coated with MUC1 via, as we demonstrate here, interactions between the mucin and the BabA and SabA adhesins. We propose that this would further prevent anchorage to the mucosal surface by blocking these key adhesins. Limiting attachment of the bacteria to the epithelial surface would also be expected to reduce pathogenicity, by restricting the functional activity of secretion systems such as that encoded by the Cag pathogenicity island, which delivers proinflammatory mediators into epithelial cells [26].
An isogenic CagA deletion mutant induced less cell death in the polarized MKN7 cultures we established in microaerophilic conditions, consistent with previous studies in other cell lines [27]. However, in our experiments which are the first to test the influence of SabA and BabA adhesins on apoptosis empirically, ablation of both adhesins had a substantially greater impact on H. pylori induced apoptosis than loss of CagA. Thus, both adhesion to the epithelial cell and CagA affect viability, but to get the maximal effect of CagA, H. pylori needs to bind to the epithelial cell via the lectin adhesins. Interestingly, at very high non-physiological bacterial infection densities, similar to those used by most researchers, these nuances were lost. Infection with BabA positive H. pylori strains has been associated with higher lymphocytic infiltration, increased epithelial proliferation, and the presence of glandular atrophy and intestinal metaplasia in human antral biopsies [28]. Similarly, severe neutrophil infiltration and atrophy (important indicators of more severe pathology) are associated with the expression of functional SabA [29]. Here, we showed that H. pylori induces apoptosis and necrosis in gastric epithelial cells in a dose dependent manner, and that an isogenic mutant strain lacking the BabA and SabA adhesins had diminished ability to induce cell death. This indicates that these adhesins are major determinants of cell surface binding and that it is the amount of adherent H. pylori that determines the impact on epithelial cell viability.
To inhibit bacterial access to the epithelial cells, the mucosal surfaces are covered with a mucus layer primarily composed of secreted mucins. This mucus layer is continuously secreted and transports away trapped material. Both salivary and gastric mucins have a high and specific binding capacity for H. pylori [9],[10],[11], and it is likely that this mucus layer keeps the majority of H. pylori away from the epithelial cell surface. For example, in the human-like rhesus monkey model [30], monkeys secreting mucins with less H. pylori binding capacity develop higher H. pylori density infections and gastritis [14]. Similarly, humans with primary Sjogren's syndrome, who produce less mucins, have more H. pylori-associated pathology [31], suggesting that the ability of secreted mucins to bind to H. pylori protects the gastric epithelium. In cultured MKN7 cells we observed a progressive depletion of MUC1 as it was shed off the cell surface. In human patients with chronic gastritis, depletion of the MUC1 extracellular domain α-subunit, but not the transmembrane β-subunit, has been reported [32]. Thus, it is likely that during chronic infection rapid shedding of the extracellular domain of MUC1 occurs in vivo, producing a similar result to that we observed in vitro.
We have shown that H. pylori adherence to live gastric epithelial cells is increased when the bacterial strain carries the BabA and SabA adhesins. However, adherence to live epithelial cells occurs even when these adhesins are absent, albeit to a reduced level. In contrast, binding to MUC1, as we have shown previously for other mucins [10],[11], is dependent on the presence of H. pylori adhesins. During the initial contact with gastric cells, MUC1 inhibits adhesion of both MUC1-binding and non MUC1-binding H. pylori to epithelial cells. H. pylori bind to MUC1 isolated from epithelial cells as well as to MUC1 shed into the gastric juice of human patients. In this study we have shown that, in vitro, H. pylori carrying these adhesins caused the gastric epithelial cells to shed MUC1 which bound to the bacteria. Inert bacterial sized beads coated with MUC1 antibodies also caused shedding of MUC1, demonstrating that MUC1 can act as a releasable decoy following engagement by particulate ligands in a Toll like receptor independent manner. Our demonstration of progressive depletion of cellular MUC1 is consistent with a previous study where MUC1 expression decreased in Kato III cells after 4 h of co-culture with H. pylori (with unknown adhesion properties) [33]. Although the study with Kato III cells showed recovery of MUC1 after 24 h of co-culture, the recovery of MUC1 may be because this study added 1000-fold higher concentrations of H. pylori than used in our study, and their H. pylori died during the experiment according to the authors possibly due to the use of aerobic culture conditions [22],[33],[34].
We have demonstrated that MMP-14 but not ADAM17 is the relevant protease involved in the endogenous shedding of MUC1 from MKN7 cells. This is consistent with previous reports that found that MUC1 shedding was sensitive to MMP-14 depletion in endometrial cells [24]. However, MMP-14 only partially affected MUC1 shedding during infection and had no influence on MUC1 shedding in non-infected cells, indicating that other factors are involved in MUC1 release. Although cleavage by another unknown MUC1 sheddase cannot be definitively excluded, the most likely scenario is disassociation of the non-covalent interaction between the transmembrane and extracellular domains at the SEA module, a site of cleavage during synthesis found in most cell surface mucins [35],[36]. Disassociation could occur due to conformational changes in MUC1 following binding or due to shear forces following binding to the highly motile bacteria. Additionally, secreted isoforms of MUC1 expressed via alternative splicing [37], could potentially be shed without prior H. pylori binding and contribute to the MUC1 pool in the mucus and gastric juice. The pattern of focal binding of MUC1 to bacteria that we observed by confocal microscopy is consistent with binding of MUC1 to regions of the bacteria coming in contact with the cell surface, followed by MUC1 shedding and bacterial detachment. However, this pattern could also arise if the SabA and BabA adhesins are focally expressed or if they are aggregated on the bacterial surface following ligation by MUC1.
MUC1 confers resistance to apoptosis induced by genotoxic drugs, the Campylobacter jejuni cytolethal distending toxin and oxidative stress in vitro [21],[38]. We found that Muc1−/− mice infected with H. pylori strain SS1 had substantially more apoptotic cells in their gastric mucosa than Muc1+/+ mice. In vitro, cultures with MUC1 knockdown had a higher proportion of apoptotic cells, both when infected and not infected, indicating that epithelial cell apoptosis is down regulated by MUC1 as a general function, not only as a response to stressors. Paradoxically, however, necrosis was lower in cells with reduced MUC1. A similar increase of damage induced apoptosis and delay of secondary necrosis has previously been described for mono(ADP-ribosyl)transferases, which control signal transduction pathways in response to cell damage during cell repair and apoptosis [39]. MUC1 increases ß-catenin levels in the cytoplasm and nuclei of carcinoma cells by blocking its degradation, resulting in an increase in cell proliferation [40],[41]. Similarly, NF-κB regulates genes that control cell proliferation and cell survival, and NF-κB is constitutively active in some cancers. MUC1 interacts directly with the IκB kinase complex, resulting in degradation of the NF-κB inhibitor IκBα [42]. Thus, high expression of MUC1, as occurs in the stomach and is commonly found in human cancers, confers increased proliferation and resistance to apoptosis both via ß-catenin and the NF-κB pathway. NF-κB is also the master regulator of inflammatory responses, including responses to H. pylori. In addition to a possible effect on NF-κB, it is possible that MUC1 modulates responses to PAMPs mediated by TLRs. In respiratory cells, MUC1 binding of bacterial flagellin (TLR5 ligand) appears to repress TLR5-mediated activation of inflammation [43],[44]. Thus, in addition to limiting bacterial attachment to the cell surface, MUC1 may modulate inflammatory responses possibly limiting unnecessary responses when bacteria fail to stably bind to the cell surface. Further work is required to fully define the influence of MUC1 on epithelial cell responses to the presence of H. pylori.
Human population studies have found associations between MUC1 allele polymorphisms and susceptibility to the development of gastric adenocarcinoma and H. pylori-associated gastritis [18],[19]. VNTR length polymorphisms have been used to characterize these alleles and short VNTR alleles which encode MUC1 proteins with shorter extracellular glycosylated domains are associated with disease. Our data are consistent with the interpretation that shorter forms of MUC1 are less efficient at sterically inhibiting attachment or acting as releasable decoys, thereby allowing increased bacterial binding to the epithelial surface and exacerbation of pathology. Experiments where cells were transfected with MUC1 expressing differing VNTR lengths suggest that longer alleles are more efficient at binding H. pylori [45]. However, it is possible that these VNTR length polymorphisms are surrogate markers for SNP's encoding functional changes in other domains (for example, SNPs affecting cytoplasmic domain signaling or cleavage of the extracellular domain) or promoter polymorphisms affecting the level of MUC1 expression during infection. More comprehensive genetic epidemiological studies are warranted to further define the nature of the MUC1 risk alleles.
Our experiments show that MUC1 inhibits the H. pylori binding to epithelial cells that occurs via both the BabA and SabA adhesins and non-adhesin mediated binding. When the pathogen does not bind to MUC1, the mucin sterically inhibits adhesion to other potential cell surface ligands. When the pathogen does bind to MUC1, the extracellular domain of the mucin is released from the epithelial surface, thereby acting as a releasable decoy and preventing prolonged adherence. Demonstration of the mechanism by which MUC1 limits gastric H. pylori infection is a model paradigm for elucidation of the function of the family of cell surface mucins which decorate the apical membrane surface of all mucosal epithelial cells, and for exploration of their contribution to preventing infectious and inflammatory disease.
The collection of gastric juice was obtained after written informed consent from patients undergoing upper gastrointestinal endoscopy (approved by the Mater Health Services' Human Research Ethics Committee, Approval No. 396A). All procedures involving animals were reviewed and approved by Institutional animal care and use committees (University of Melbourne; AEEC No. 03219)
Dewaxed and rehydrated formalin-fixed sections (4 µm) were treated with 10 mM citric acid, pH 6 at 100°C for 20 min and then with 3% (v/v) hydrogen peroxide for 30 min at room temperature. The sections were washed 3 times between all subsequent steps in 0.15 M NaCl, 0.1 M Tris/HCl buffer (pH 7.4) containing 0.05% Tween-20. Non-specific binding was blocked by protein block (Dako) for 30 min. The sections were incubated with an anti-MUC1 antibody (CT2, gift from Prof. S. Gendler, Scottsdale, USA) diluted 1∶50 in Antibody Diluent (Dako) for 1 h, then incubated with Broad Spectrum Poly HRP Conjugate (Zymed Laboratories Inc, San Fransisco, USA) for 10 min and with diaminobenzidine for 10 min. The sections were counterstained with Harris's haematoxylin.
Three antibodies against MUC1 were used in this study: The CT2 antibody is against the cytoplasmic tail of MUC1, whereas the BC2 and BC3 antibodies are against the extracellular domain of MUC1, and react with a VNTR repeat epitope that is exposed even in fully glycosylated MUC1 [46]. These antibodies all react with mature glycosylated MUC1 on the apical surface of human gastric epithelium [46].
Gastric juice was obtained after informed consent from patients undergoing upper gastrointestinal endoscopy. The gastric juice was adjusted to pH 7 with Tris and mixed with equal volume of 150 mmol/L NaCl, 1% NP40, 0.5% deoxycholic acid, 0.1% SDS, 50 mmol/L Tris, pH 7.5 containing Complete protease inhibitor (Roche Diagnostics GmbH, Mannheim, Germany) (RIPA buffer) and incubated for 5 min under agitation at 4°C, centrifuged at 10,000 g for 10 min at 4°C and the supernatant used for the binding assays. Cells from the gastric cancer cell line MKN7 (Riken, Japan) were lysed in RIPA buffer for 10 min under agitation at 4°C, lysates centrifuged at 10,000 g for 10 min at 4°C and the high molecular weight fraction purified from the supernatant using a Microcon centrifugal filter device (cut off 100 kDa). Microtitre plates (Polysorb, Nunc, Denmark) were coated overnight at 4°C with either the BC2 antibody or the isotype control antibody 401.21 against α-gliadin (100 µL/well, 4 µg/mL in phosphate-buffered saline, PBS). The plates were then washed 3 times in PBS with 0.05% Tween-20 and unbound sites blocked with 1% BSA in PBS for 1 h at room temperature. The wells were then incubated with cell lysate or gastric juice extract diluted 1∶5 in Blocking reagent for ELISA (Boehringer Mannheim, Germany) containing 0.5% BSA and 0.05% Tween-20 (dilution buffer) for 2 h with orbital shaking, washed as above, incubated with primary antibody: anti-MUC1 (clone BC3), anti-Leb (clone LE2, Biotest, Dreieich, Germany), anti-sialyl-Lea (clone CA19-9, NeoMarkers, Freemony, CA, USA) or anti-sialyl-Lex (AM3, gift from Dr C. Hanski, University Medical Center Charite, Berlin, Germany) diluted to 1 µg/mL, 1∶200, 1∶1000 and 1∶20, respectively. The plates were washed and incubated with HRP-conjugated anti-mouse IgM (Jackson ImmunoResearch Laboratories, Inc, USA). HRP activity was determined using 2,2′-Azinobis(3-ethylbenzothiazoline)-6-sulphonic acid as a substrate (0.550 g/L in citrate phosphate buffer pH 4.3) by measuring absorbance at 405 nm. Assays were performed in triplicate.
H. pylori were harvested and washed twice by centrifugation at 2,500 g in PBS containing 0.05% Tween-20. 100 µl H. pylori (OD600 nm = 0.90) was incubated with 500 ng FITC labelled Leb or sialyl-Lex HSA-conjugates for 30 min in PBS containing 0.5% human albumin and 0.05% Tween-20. Fluorescence was measured after washing the bacteria twice. Assays were performed in triplicate.
Binding of H. pylori to MUC1 was determined by sandwich ELISA using biotinylated bacteria. Polysorb Microtitre plates (Nunc, Denmark) were coated overnight at 4°C with either the BC2 antibody against the extracellular domain of human MUC1, or the isotype control antibody 401.21 (100 µL/well, 4 µg/mL in PBS). The plates were then washed 3 times in PBS with 0.05% Tween-20 and unbound sites blocked with 1% BSA in PBS for 1 h at room temperature. The wells were then incubated with cell lysate or gastric juice extract diluted 1∶5 in Blocking reagent for ELISA (Boehringer Mannheim, Germany) containing 0.5% BSA and 0.05% Tween-20 (dilution buffer) for 2 h with orbital shaking, washed as above, then incubated with a suspension of biotinylated SS1, J99wt or J99ΔBabA/SabA (OD600 = 0.15 diluted 1∶10 in dilution buffer) for 2 h at 37°C with orbital shaking. The plates were washed and incubated with streptavidin-HRP (diluted 1∶1000 in dilution buffer) for 1 h at room temperature and HRP activity determined as above. To get a sufficient amount of MUC1 to analyze binding with a range of H. pylori strains, archived purified mucins from a previously published study on tissue from patients with no history of peptic ulcer disease undergoing elective surgery for morbid obesity was used [11]. MUC1 was isolated from whole gastric wall using isopycnic density gradient centrifugation followed by gel chromatography of the mucin containing fractions to separate MUC1 from the oligomeric mucins as previously described [9], and analyzed by ELISA [9].
The gastric epithelial cell line MKN7 (Riken Cell Bank, Japan) was cultured in RPMI containing 10% FCS, 2 mM L-glutamine, 100 units/mL penicillin G sodium and 100 ug/mL streptomycin. Mucin expression and glycosylation in this cell line has been described previously [22]. For co-culture with H. pylori the medium was changed to antibiotic free medium 20 h prior to infection. The co-culture experiments were performed on confluent MKN7 cells transferred to microaerobic conditions (5% O2, 15% CO2, 80% N2) at the start of the co-culture. MKN7 cell viability is not compromised during these conditions [22], and the microaerobic conditions are similar to the actual pO2 and pCO2 in tissues of the human body [47].
To assess integrity of monolayer cultures MKN7 cells were cultured on snapwell tissue culture inserts, which were then were mounted in vertical Ussing chambers (exposed area 1.13 cm2). The basolateral side of the membrane was immersed in 115.8 mM NaCl, 1.3 mM CaCl2, 3.6 mM KCl, 1.4 mM KH2PO4, 23.1 mM NaHCO3, 1.2 mM MgSO4 (KREB) solution containing 5.7 mM Na-pyruvate, 5.1 mM Na-L-glutamate 10 mM and D-glucose, whereas the apical compartment was immersed in KREB's solution containing 5.7 mM Na-pyruvate, 5.13 mM Na-L-glutamate and 10 mM D-mannitol. The solutions were gassed with 95% O2 and 5% CO2 at a temperature of 37°C and pH 7.4 throughout the whole experiment. Epithelial resistance (Rp) was measured using square-pulse analysis. 5 V, 3 ms pulses were generated by a square pulse generator (Medimet, Gothenburg, Sweden) via a current limiting resistor (36 kΩ) connected to a platinum electrode and applied across the sample.
H. pylori were grown on Brucella agar supplemented with 10% bovine blood, 2% Vitox (Oxoid), 10 µg/mL vancomycin (Sigma), 5 µg/mL trimethoprim (Sigma) and 4 µg/mL amphoteracin B (Sigma) for 4 days in 5% O2 and 15% CO2 at 37°C. H. pylori strains J99 wild type (wt) bind Leb and sialyl-Lex, CCUG17875/Leb and P466 bind Leb but not sialyl-Lex, the 17875BabA1::kanbabA2::cam-mutant (75ΔbabA1A2) and CCUG17874, bind sialyl-Lex but not Leb, whereas the remaining strains; the isogenic J99 adhesion mutant lacking the BabA and SabA adhesins (J99babA::camsabA::kan [13], referred to as J99ΔBabAΔSabA), 26695, P1 and P1-140 do not bind sialyl-Lex or Leb [11] (provided by Prof. Thomas Boren, Umeå University, Sweden). The J99 CagA deletion mutant (J99ΔCagA) was made by natural transformation with a plasmid containing the cloned CagA from strain 26695 that was knocked out by insertion of the CatGC cassette into the singular Bg/II site in the middle of the cagA gene (provided by Prof Steffen Backert, Institut fur Medizinische Microbioligie, Magdeburg, Germany). Clones were selected after culture on 6 µg chloroamphenicol/mL agar. Disruption of CagA was verified by PCR using the 5′-AAAGGATTGTCCCTACAAGAAGC-3′ and 5′-GTAAGCGATTGCTCTTGCATC sequences. The concentration of H. pylori was estimated by measuring OD600, and then the amount of CFU in the inoculum was determined by counting colonies from serial dilutions cultured for 5 days. The Multiplicity Of Infection (MOI, pathogen∶host cell) was determined based on the CFU of the H. pylori and a density of 1.5×105 MKN7 cells/cm2 cells at the day of confluency, which is the day we infected the cells.
The bacteria were washed twice in 0.2 mol/L carbonate buffer pH 8.3 (2×109 bacteria/mL), 125 µg/mL biotin-XX-NHS was added and the mixture rotated in the dark for 15 min at room temperature. To verify that the BabA adhesin remained functional, adherence to the Leb conjugate was measured (Figure 2).
105 fluorescent beads (Fluospheres NeutrAvidin labelled 1 µm microspheres, Molecular probes) coated with an antibody against the extracellular domain of MUC1 (BC2 antibody) or isotype control (401.21) were added to 96 well plates with confluent MKN7 epithelial cells (N = 9). After incubation for 4.5 h, 24 h and 44 h, cultures were washed 3 times with ice cold PBS and fluorescence was quantified in a FLA5100 (Fujifilm).
RIPA lysates were subjected to SDS-PAGE in a 4–12% gradient gel. Western blots were cut in the middle (∼70 kDa) and the upper half was stained with an antibody against the MUC1 extracellular domain (BC2) and the lower half was probed for β-actin and detected with fluorescent probes on the Licor Odyssey instrument.
Total RNA was prepared using the RNeasy Mini Kit (Qiagen, Valencia, CA, USA). The quantity and quality of the RNA was determined by spectrophotometry (ND-1000; NanoDrop Technologies Inc., Wilmington, DE). Total RNA (1 µg) from each sample was used for first strand cDNA synthesis using SuperScript™ III reverse transcriptase (Invitrogen) following the manufacturer's instructions. Real-time PCR was monitored by SYBR® Green I fluorescence (Invitrogen) using Platinum ® Taq DNA-Polymerase (Invitrogen) with 3 mM MgCl2, 0.2 µM primers, 200 µM dNTPs, and 0.5 U polymerase per reaction (25 µl) under primer-specific conditions. The following experimental protocol for PCR reaction (40 cycles) was performed on a Rotor-Gene 3000 cycler (Corbett Research, Sydney, Australia): denaturation for 15 min at 95°C, followed by 40 amplification cycles at 94°C (20 s), annealing under primer-specific conditions (30 s), and extension for 45 s at 72°C. Primers with the following sequences were chosen: GAPDH: forward: 5′- CCTGTACGCCAACACAGTGC -3′, reverse: 5′- ATACTCCTGCTTGCTGATCC -3′, annealing temperature 60°C. MUC1: forward: 5′- CCCCTATGAGAAGGTTTCTGC-3′, reverse: 5′- ACCTGAGTGGAGTGGAATGG -3′, annealing temperature 60°C. ADAM-17: forward: 5′-ACCTGAAGAGCTTGTTCATCGAG -3′, reverse: 5′-CCATGAAGTGTTCCGATAGATGTC-3′, annealing temperature 60°C. MMP-14: forward: 5′-CCATCATGGCACCCTTTTACC-3′ reverse: 5′-TTATCAGGAACAGAAGGCCGG-3′; annealing temperature 60°C. All Primers were obtained from GeneWorks (Hindmarsh, SA, Australia). To confirm the specificity of the amplified DNA, a melting curve was determined at the end of each run. The reaction efficiency was determined with a dilution series of cDNA containing the PCR products. Genes were normalized to the unregulated housekeeping gene GAPDH and the results were expressed as ratio of target gene and GAPDH expression (arbitrary units). Control experiments were also performed to ensure that GAPDH expression was not differentially regulated under the experimental conditions employed.
siRNA specific for ADAM17 and MMP-14 as well as scrambled siRNA were chemically synthesized (Dharmacon Research, Lafayette, USA) as a mixture of four siRNAs targeting different regions of the same gene to enhance the silencing performance (21 mers, SMARTpool). MKN 7 cells with a confluence of 70–80% were transfected with either 250 nM of ADAM17 siRNA or MMP 14 siRNA individually or combined both together using Lipofectamine 2000 Reagent (Invitrogen) according to the manufacturer's instructions. 250 nM or 500 nM of scrambled control siRNA were used as negative control. After 48 h of transfection, the MKN7 cells were co-cultured with H pylori J99 wild type for a further 8 and 24 h, and the uninfected MKN 7 cells were also cultured for a further 8 and 24 h as a negative control. The MKN7 cells were harvested after 56 and 72 h transfection, the level of knockdown of the ADAM 17 or MMP-14 was detected by quantitative RT-PCR, and the influence of the treatment on MUC1 shedding was monitored by ELISA.
100 nM 2′-hydroxyl DsiRNA against the target sequence NNGUUCAGUGCCCAGCUCUAC (1∶1) and NNGCACCGACUACUACCAAGA (1∶3) or siCONTROL non-targeted siRNA#2 (Dharmacon) were transfected into MKN7 cells using Lipofectamine 2000. Four days after transfection the level of knockdown was measured using the median fluorescence intensity determined by flow cytometry (below). The 1∶3 siRNA generally gave higher knockdown than the 1∶1 siRNA.
Cell viability was analysed by collecting non-adherent cells together with attached cells harvested with trypsin and counting cell suspensions by flow cytometry on cells stained with 1 µg/mL 7-aminoactinomycin D (7AAD) and Annexin-V-PE Apoptosis Detection Kit I (BD Pharmingen) according to the manufacturers instructions. For MUC1 detection, cells harvested with trypsin and stained with 1 µg/mL 7AAD were either stained without fixation (staining of cell surface structures) or fixed in 1% paraformaldehyde for 5 min on ice and then permeabilized with 0.5% saponin (intracellular and extracellular staining). The cells were then incubated with the anti-MUC1 antibody BC2 or isotype control 401.21 at 3 µg/mL in 1% BSA in PBS for 60 min at 4°C and then with anti-mouse antibody conjugated to Alexa fluor 488 (Invitrogen). For detection of intracellular antigens, washes and antibody incubations were performed in the presence of 0.5% saponin. After staining, all cells were fixed with 1% paraformaldehyde. The analysis was gated to exclude 7AAD positive cells and assessment of staining was performed on a LSRII Flow Cytometer (BD Biosciences, San Jose, USA) using the Diva software (BD Biosciences, San Jose, USA). Bacteria were recovered from the culture medium of MKN7 cells by first sedimenting non-adherent mammalian cells (300 g, 5 min) and then sedimenting bacteria (5000 g, 10 min). Bacteria were then stained with BC2 or 401.21 at 10 µg/mL as above and gated using the SSC and FSC pattern of broth cultured H. pylori.
Bacteria prepared and stained for flow cytometry as above were smeared onto charged glass slides, stained with DAPI (0.1 µg/ml) for 15 min, washed with PBS, mounted in Prolong Gold (Invitrogen) and examined using a Zeiss LSM510 confocal microscope with multitracking detecting DAPI (excitation 405 nm, detection 420–480 nm) and FITC (excitation 488 nm detection, LP 505 nm) fluorescence separately.
H. pylori J99wt, the J99ΔCagA or J99ΔBabAΔSabA were co-cultured at a concentration of 8×105 CFU H. pylori/well in a 96 well plate. Co-cultures were washed 3 times with ice cold PBS and then fixed with 4% paraformaldehyde for 20 min on ice. After washing, wells were blocked for 1 h with 1% BSA in PBS and then incubated with mouse polyclonal anti-Helicobacter antisera [20]. The plates were washed and incubated with HRP-conjugated anti-mouse IgG (Jackson ImmunoResearch Laboratories, Inc, USA). The plates were washed and HRP activity determined using TMB by measuring absorbance at 450 nm.
All procedures involving animals were reviewed and approved by Institutional animal care and use committees (University of Melbourne; AEEC No. 03219). The mouse infection tissue samples were archived material from a previously published study [20]. Female aged matched 129/SvJ wild type and 129/SvJ Muc1−/− mice were infected intra-gastrically once with 107 H pylori suspended in 0.1 mL Brain Heart Infusion (Oxoid).
The assay was performed according to the manufacturer's instructions (Roche), except that the reaction was diluted 1∶4 in 0.1 M sodium cacodylate buffer, pH 7.3 to decrease the background. The total number of TUNEL positive cells per 10 randomly selected fields of view in the entire gastric mucosa was counted at 20× magnification.
For normally distributed data the Students t-test was used to compare groups. For analyses where a normal distribution could not be demonstrated, including where the number of replicates was low, the non-parametric Mann Whitney U test was used to compare groups. The ANOVA test was used when comparing 3 or more groups, and to ascertain that the multiple testing did not add to the chance of finding statistically significant differences, the Tukey's or Bonferroni's post hoc tests were used.
|
10.1371/journal.pgen.1005516 | Photoreceptor Specificity in the Light-Induced and COP1-Mediated Rapid Degradation of the Repressor of Photomorphogenesis SPA2 in Arabidopsis | The Arabidopsis COP1/SPA E3 ubiquitin ligase is a key negative regulator that represses light signaling in darkness by targeting transcription factors involved in the light response for degradation. The COP1/SPA complex consists of COP1 and members of the four-member SPA protein family (SPA1-SPA4). Genetic analysis indicated that COP1/SPA2 function is particularly strongly repressed by light when compared to complexes carrying the other three SPAs, thereby promoting a light response after exposure of plants to extremely low light. Here, we show that the SPA2 protein is degraded within 5–15 min after exposure of dark-grown seedlings to a pulse of light. Phytochrome photoreceptors are required for the rapid degradation of SPA2 in red, far-red and also in blue light, whereas cryptochromes are not involved in the rapid, blue light-induced reduction in SPA2 protein levels. These results uncover a photoreceptor-specific mechanism of light-induced inhibition of COP1/SPA2 function. Phytochrome A (phyA) is required for the severe blue light responsiveness of spa triple mutants expressing only SPA2, thus confirming the important role of phyA in downregulating SPA2 function in blue light. In blue light, SPA2 forms a complex with cryptochrome 1 (cry1), but not with cryptochrome 2 (cry2) in vivo, indicating that the lack of a rapid blue light response of the SPA2 protein is only in part caused by a failure to interact with cryptochromes. Since SPA1 interacts with both cry1 and cry2, these results provide first molecular evidence that the light-regulation of different SPA proteins diverged during evolution. SPA2 degradation in the light requires COP1 and the COP1-interacting coiled-coil domain of SPA2, supporting that SPA2 is ubiquitinated by COP1. We propose that light perceived by phytochromes causes a switch in the ubiquitination activity of COP1/SPA2 from ubiquitinating downstream substrates to ubiquitinating SPA2, which subsequently causes a repression of COP1/SPA2 function.
| Plants have evolved photoreceptors that initiate a signaling cascade to adjust growth and development to the ambient light environment. The CUL4-dependent COP1/SPA E3 ubiquitin ligase is a key negative regulator of light signaling whose function is repressed by light. Recent research has identified mechanisms that are common to both phytochrome and cryptochrome photoreceptors. Here, we have identified a mechanism of light-induced COP1/SPA repression that is specific to phytochrome photoreceptors. We show that the SPA2 protein is very rapidly degraded in red, far-red and blue light in a phytochrome-dependent fashion. We further show that SPA2 degradation in the light depends on COP1 and on the interaction of SPA2 with COP1. Hence, our results suggest a light-induced degradation of SPA2, but not of COP1, by the COP1/SPA2 ubiquitin ligase. The human ortholog of COP1, which functions without the plant-specific SPA proteins, is known to be regulated by autodegradation following DNA damage. Hence, autodegradation of components of this E3 ligase is a regulatory mechanism used in both humans and plants.
| As sessile organisms plants continuously monitor the ambient light conditions and adjust their growth and development with the aim to optimize growth and—ultimately—seed production in a competitive environment. Plants sense the intensity, color, direction and periodicity of light. Responses to these light parameters include seedling deetiolation (inhibition of hypocotyl elongation, opening of cotyledons and apical hook, greening), phototropism, shade avoidance, the accumulation of anthocyanins and the induction of flowering in particular day lengths [1].
To sense the light, plants have evolved several classes of photoreceptors [1,2]. The phytochrome photoreceptors sense red light (R) and far-red light (FR) and exist in two R/FR photointerconvertible conformations. Among the five phytochromes in Arabidopsis (phyA-phyE), the relatively light-stable phyB is the primary phytochrome controlling FR-reversible responses to R. These responses are also named low fluence responses (LFR). phyA is rapidly degraded in R and primarily mediates high-irradiance responses (HIR) to continuous FR (FRc) and very low fluence responses (VLFR) [3,4]. Blue light (B) is sensed by cryptochromes, phototropins and the ZEITLUPE family, but also by phyA. Cryptochromes are encoded by two genes in Arabidopsis, CRY1 and CRY2. Both mediate seedling deetiolation in B, while primarily cry2 is responsible for B-induced flowering in long days [5,6]. For both, phytochromes and cryptochromes, mutant photoreceptor variants have been identified that are constitutively active and thus signal also in darkness [7–10]. Recently, UVR8 was identified as the long-sought UV-B receptor [11,12].
In Arabidopsis, the phytochrome and cryptochrome photoreceptors act to inhibit a key repressor of light signaling that prevents light responses in darkness. This repressor, the CONSTITUTIVELY PHOTOMORPHOGENIC1/SUPPRESSOR OF PHYA-105 (COP1/SPA) complex, functions as an E3 ubiquitin ligase which ubiquitinates positively-acting light signaling intermediates, mainly transcription factors, thereby targeting them for proteolytic degradation in the 26S proteasome. In the light, photoreceptors directly interact with the COP1/SPA complex, leading to its inactivation which subsequently allows the target transcription factors to accumulate and to initiate vast reprogramming of gene expression [13,14]. The degradation of the light-labile photoreceptors phyA and cry2 is also in part dependent on COP1 and/or SPA genes [15–18]. The Arabidopsis COP1/SPA complex is likely a tetramer consisting of two COP1 and two SPA subunits [19]. COP1 is a single-copy gene in higher plants, while SPA proteins are encoded by a small gene family of four genes in Arabidopsis (SPA1-SPA4) and 2 genes in rice [13,20]. Mutations in either COP1 or all four SPA genes lead to constitutive photomorphogenesis in Arabidopsis, with seedlings showing the features of light-grown seedlings in complete darkness [21,22]. While cop1 null mutants arrest growth at the seedling stage, spa null mutants are viable. cop1 spa quintuple null mutants can complete embryogenesis, indicating that the COP1/SPA complex is not necessary for embryogenesis [23]. Apart from controlling seedling growth, the COP1/SPA complex also plays an important role during other light-induced responses, such as anthocyanin biosynthesis, elongation responses during shade avoidance, leaf expansion and the suppression of flowering under non-inductive short-day conditions. These responses are mediated through a number of COP1/SPA substrates including CO, HFR1, PAP1, PAP2 and BBX family proteins [24–32]. Moreover, COP1/SPA is a positive regulator in UV-B mediated photomorphogenesis [11,12]. The four SPA genes have overlapping but also distinct functions in controlling the various light responses during plant development [22,24–26,33].
The COP1/SPA complex acts as part of a CULLIN4 (CUL4)-based E3 ubiquitin ligase. CUL4-associated E3 ligases consist of CUL4, RBX1, DDB1 as well as a variable WD repeat protein which recognizes the substrate and binds DDB1 [34,35]. The WD repeat proteins COP1 and SPA are substrate adaptors in CUL4-DDB1COP1/SPA E3 ligase(s) [36]. Both COP1 and SPAs contain a central coiled-coil domain responsible for the formation of the COP1/SPA complex via homo- and heterodimerization [19,37,38]. In their C-termini, both COP1 and SPAs carry a WD-repeat domain which mediates interaction with substrates as well as with DDB1 [36,39]. The N-termini of COP1 and SPA are distinct, with COP1 harboring a RING finger domain and SPA proteins carrying a kinase-like domain [40,41].
Light is the key factor controlling COP1/SPA activity. Genetic studies showed that the SPA2 protein is particularly strongly inactivated by light when compared to the other three SPAs, making SPA2 a particularly interesting SPA when analyzing light-mediated inhibition of COP1/SPA activity [22,42]. How light inactivates the COP1/SPA complex is not fully understood. Evidence indicates that phytochrome and cryptochrome photoreceptors converge on COP1/SPA to promote light signaling in R, FR and B. Such light-induced inactivation of COP1/SPA occurs via multiple mechanisms. First, after light exposure, COP1 translocates from the nucleus into the cytoplasm [43,44]. Second, the B-dependent interaction of cry1 with SPA1 reduces the COP1/SPA1 interaction [45–47]. Similarly, an interaction of light-activated phytochromes A and B with members of the SPA family reduces the interaction between COP1 and SPA proteins [48,49]. For cry2, B acts to promote the interaction of cry2 with COP1 [50]. A third mechanism which reduces COP1/SPA activity in FRc-grown plants involves the degradation of SPA1 and SPA2 in the proteasome [42]. Here, we have analyzed the molecular mechanism of SPA2-degradation in different light qualities and uncover a photoreceptor-specific mechanism of light-induced COP1/SPA repression via COP1.
To investigate the dynamics and wave-length dependency of light-induced SPA2 degradation, we determined SPA2 protein levels in dark-grown seedlings that were briefly exposed to R, FR or B. These seedlings expressed HA-tagged SPA2 under the control of the 5′ and 3′ regulatory sequences of SPA2 (SPA2::SPA2-HA) [42]. The SPA2 promoter expresses at the same level in dark-grown and light-exposed seedlings [42,51]. Therefore, light-induced differences in SPA2-HA protein levels in these lines are due to changes in protein stability, as shown previously [42]. Exposure of dark-grown seedlings to a short, 200-second pulse of R (Rp) was sufficient to strongly reduce SPA2-HA protein levels within 5 min after subsequent transfer to darkness (Fig 1A). Ten minutes after the Rp, there was barely any SPA2-HA protein detectable. Similarly, when dark-grown seedlings were irradiated with a pulse of FR (FRp) or B (Bp), SPA2-HA protein abundance decreased to a very low level. The response time to FRp and Bp was also very rapid, but slightly longer when compared to Rp.
To determine whether the native SPA2 protein behaves like the SPA2-HA protein, we analyzed SPA2 protein levels in wild-type seedlings using an α-SPA2 antibody. Because SPA2 levels are very low, we enriched the protein preparations through nuclear extracts to detect the constitutively nuclear-localized SPA2 protein [22,42]. Fig 1B shows that a pulse of FR, R or B rapidly and strongly reduced SPA2 protein abundance. Again, Rp was more effective in reducing SPA2 levels than FRp and Bp. We subsequently asked what fluences are necessary for the reduction of SPA2 protein levels. Fluences of 0.002 μmol m-2 of R, i.e. a 200-s-pulse of R with a fluence rate of 10−5 μmol m-2 s-1, was sufficient to reduce SPA2 protein levels to almost undetectable levels (Fig 1C), indicating that degradation in R is extremely sensitive to light and likely involves a VLFR. FRp and Bp were again less effective than Rp (Fig 1C).
We subsequently asked whether R, FR and B also cause a decrease in SPA1 levels. Here, the induction of SPA1 gene expression by light [40] precluded a specific analysis of protein stability using α-SPA1 antibodies. Therefore, we used transgenic lines expressing SPA1-HA under the control of the constitutive SPA2 promoter. These lines showed a strong reduction in SPA1-HA abundance in FR, R and B (Fig 2).
We asked which photoreceptor(s) are responsible for degradation of SPA2 in different light qualities and quantities. To this end we investigated SPA2 levels in various photoreceptor mutants. Degradation of SPA2 in response to FRc was fully abolished in a phyA mutant, in both Col and RLD accessions (Fig 3A). Similarly, a pulse of FR had no effect on SPA2 protein levels in a phyA mutant (Fig 3B). Hence, phyA is responsible for SPA2 degradation in FR.
After a pulse of R with low fluence rates, SPA2 protein levels were not reduced in a phyA mutant nor in a phyA phyB double mutant. A phyB mutant, in contrast, showed a reduction in SPA2 levels after Rp and thus exhibited a similar response as the wild type (Fig 3C). The phyA-requirement for a response to low Rp confirms that this treatment initiates a phyA-perceived VLFR [3]. After a pulse with higher fluence rates of R, only the phyA phyB double mutant lacked a reduction in SPA2 protein abundance when compared to dark-grown seedlings (Fig 3D). Hence, phyA and phyB mediate the degradation of SPA2 after high Rp. This suggests that both VLFR and LFR responses trigger SPA2 degradation in red light. R/FR reversibility is a hall-mark of an LFR [4]. Indeed, SPA2 degradation after Rp was reversible by a pulse of FR in a phyA mutant background which would lack the VLFR (Fig 3E).
Deetiolation in blue light is mediated by the cryptochromes cry1 and cry2 as well as by phyA. We therefore investigated B-induced degradation of SPA2 in cry1 cry2 and in phyA mutants. After a pulse of B, the decrease in SPA2 levels was abolished in phyA mutant seedlings but was normal in the cry1 cry2 mutant (Fig 4A). These results indicate that a pulse of B only triggered phyA-mediated SPA2 degradation. Also after irradiation with continuous B of very high fluence rates (50 μmol m-2 s-1) for 30 min high SPA2 levels were retained in phyA mutant seedlings. In cry1 cry2 mutant seedlings, SPA2 levels were again strongly reduced similar to wild-type seedlings (Fig 4B). Only after prolonged irradiation with B of high fluence rates for 24 h, SPA2 levels decreased in a phyA-deficient mutant (Fig 4C). These results show that the rapid B-induced reduction in SPA2 levels is exclusively mediated by phyA. Only after very long irradiation with B of high fluence rates other photoreceptor(s) become active in reducing SPA2 levels.
In an attempt to uncover a possible role of cryptochromes in the response to long-term B irradiation, we analyzed SPA2 protein levels in a cry1 cry2 phyA-201 triple mutant background (Ler accession). SPA2 protein levels still decreased in this mutant after prolonged exposure to blue light (S1A Fig). However, SPA2 levels in phyA-201 also decreased after FRc (S1B Fig). Hence, degradation of SPA2 in the cry1 cry2 phyA-201 triple mutant may either be due to residual phyA activity or, alternatively, the regulation of SPA2 stability may be different in the Ler accession than in the Col and RLD accessions.
Constitutively active photoreceptor variants have been described that initiate light signaling even in darkness. We therefore investigated whether these photoreceptor variants also cause a constitutive reduction in SPA2 protein abundance, i.e. also in dark-grown seedlings. To this end, we analyzed SPA2 protein levels in transgenic lines expressing the constitutively active phytochrome mutants phyBY276H and phyAY242H [7]. As reported previously, phyBY276H-expressing seedlings showed very strong constitutive photomorphogenesis, both in the PHYB wild-type and the phyB-5 mutant background [7] (Fig 5A). These phyBY276H lines showed very low SPA2 protein levels in dark-grown seedlings (Fig 5B), suggesting that the SPA2 protein is destabilized in darkness by the constitutively active phyB photoreceptor.
phyAY242H-expressing seedlings exhibit weaker constitutive photomorphogenesis than phyBY276H-expressing seedlings [7]. These seedlings had a shorter hypocotyl and a partially opened hook, especially in the phyA mutant background, when compared to the wild type (Fig 5A). This phenotype was somewhat weaker than reported previously which is likely due to the younger age of our seedlings and the absence of sucrose in the culture medium when compared to [7]. SPA2 abundance in dark-grown seedlings was strongly reduced in lines expressing phyAY242H in a phyA background, while it was similar to wild type in lines expressing phyAY242H in a PHYA wild-type background (phyAY242H/Ler) (Fig 5B). Hence, SPA2 levels were constitutively reduced in the presence of phyAY242H, but this effect is outcompeted by the presence of wild-type phyA. In summary, these mutations in both phyA and phyB cause constitutive degradation of SPA2 in darkness.
Expression of the N-terminal 406 amino acids of phyA fused to an artificial dimerization domain has also been shown to cause constitutive photomorphogenesis in darkness [10] (Fig 5C). However, this phyA variant did not alter SPA2 protein levels in darkness (Fig 5D), indicating that this constitutively active phyA variant was not capable of inducing SPA2 degradation in darkness.
Fusion of the cry1 C-terminal extension (CCT1) to an artificial dimerization domain (GUS) leads to a constitutively active cry1 photoreceptor. Similarly, a cry1G380R variant is constitutively active. Hence, seedlings expressing CCT1 or cry1G380R exhibit strong constitutive photomorphogenesis in darkness [8,9]. SPA2 protein levels were unaltered in dark-grown GUS-CCT1- and cry1G380R-expressing seedlings when compared to the wild type, despite the constitutive photomorphogenesis displayed by these seedlings (Fig 5E and 5F). Hence, none of the constitutively active cry1 variants affected SPA2 protein levels in darkness. This is in agreement with the primary roles of phytochromes in SPA2 degradation.
Since rapid degradation of SPA2 in B was exclusively dependent on phyA, we predicted that phyA is of particular importance in inactivating SPA2 function in B. To test this hypothesis, we generated a phyA-deficient spa1 spa3 spa4 phyA mutant which only expresses functional SPA2 among the four SPA proteins. Hence, we can observe the effect of light on SPA2 activity in the absence of any other SPAs, and in the presence or absence of phyA. We had shown previously that spa1 spa3 spa4 mutant seedlings etiolate normally in darkness but are very hypersensitive to R, FR and B when compared to the wild type, thus resembling a spa quadruple mutant already at extremely low fluence rates of light [22,42] (Fig 6A–6C). Hence, SPA2 is sufficient for full repression of photomorphogenesis in darkness but is extremely effectively inactivated by light. In B, spa1 spa3 spa4 phyA mutant seedlings displayed much longer hypocotyls than spa1 spa3 spa4 mutant seedlings, indicating that the lack of phyA dramatically reduced the responsiveness of the spa1 spa3 spa4 mutant to B. The hypocotyl length of the spa1 spa3 spa4 phyA mutant in B was very similar to that of the phyA single mutant. Hence, in the absence of phyA, the mutations in SPA1, SPA3 and SPA4 had no detectable effect (Fig 6A). In Rc, the phyA mutation abolished the hypersensitivity of the spa1 spa3 spa4 mutant to lower fluence rates of Rc but not to higher fluence rates of Rc (Fig 6B). This is consistent with our finding that SPA2 degradation in lower fluence rates of R requires phyA, while in higher fluence rates of R phyB in addition to phyA mediates SPA2 degradation. As expected, the responsiveness of spa1 spa3 spa4 mutant seedlings to FRc was fully dependent on phyA (Fig 6C). Taken together, these results show that the hypersensitivity of the spa1 spa3 spa4 mutant to B fully depends on phyA. This agrees with our observation that rapid SPA2 degradation in B was exclusively dependent on phyA. Since the spa1 spa3 spa4 phyA mutant retained responsiveness to B, as indicated by the inhibition of hypocotyl elongation in B of higher fluence rates, additional phyA-independent mechanisms of SPA2 inactivation by B exist. These are likely mediated by the cryptochromes.
The lack of cryptochrome activity in B-induced SPA2 degradation might be caused by a failure of cryptochromes to rapidly interact with SPA2. Indeed, FRET/FLIM studies in transfected tobacco leaves failed to show an interaction between cry2 and SPA2. Similarly, recombinantly produced cry2 and SPA2 did not interact in in vitro pull-down assays [15]. On the contrary, SPA2 was shown to weakly interact with cry2 in B in the yeast two-hybrid system [50]. For cry1, no significant interaction with SPA2 was observed in the yeast two-hybrid assay [46]. To reinvestigate this question in planta, we conducted co-immunoprecipitation experiments using transgenic Arabidopsis seedlings expressing SPA2-HA and, as a positive control, SPA1-HA (Fig 7). To obtain similar protein levels of SPA1-HA and SPA2-HA in B, SPA1-HA was expressed under the control of the weaker SPA2 promoter (SPA2::SPA1-HA) and SPA2-HA from the stronger SPA1 promoter (SPA1::SPA2-HA). Moreover, seedlings were treated with proteasome inhibitor to reduce SPA degradation in B. Fig 7A shows that upon B-exposure both SPA1-HA and SPA2-HA co-immunoprecipitated higher-mobility cry1 isoforms which are formed in B. Hence, B induced the formation of a SPA2/cry1 complex, as it was previously reported for a SPA1/cry1 complex [46,47]. In addition, a lower-mobility cry1 which likely represents the non-phosphorylated isoform of cry1 showed weak constitutive interactions with SPA1-HA and SPA2-HA in B and darkness. The association of higher-mobility cry1 with SPA2 was very rapid. It occurred within 5 min of B-exposure (S2 Fig). cry2, in contrast, was not co-immunoprecipitated by SPA2-HA, neither in darkness nor in B. The positive control SPA1-HA showed the expected B-dependent association with cry2 (Fig 7B). In summary, in B, SPA2 associates with cry1 but not with cry2 in planta.
To identify the E3 ubiquitin ligase that mediates SPA2 degradation in the light, we asked whether the COP1/SPA E3 ligase itself may be responsible for ubiquitination of SPA2. We therefore investigated SPA2 protein levels in the hypomorphic cop1-4 mutant and in the cop1-5 null mutant, using light conditions that cause full degradation of SPA2. In a cop1-4 hypomorphic background, considerable SPA2 protein levels were retained in seedlings irradiated with FRc (Fig 8A and 8B). Hence, the FRc-induced reduction in SPA2 protein abundance was strongly attenuated, but not abolished, by the partial-loss-of-function cop1-4 mutation. We subsequently analyzed SPA2 protein levels in the cop1-5 null mutant. Because cop1 null mutants arrest growth at the very early seedling stage and, moreover, mostly fail to break the seed coat during germination, we could not obtain enough tissue for nuclear-enriched protein preparations which are necessary to detect the native SPA2 protein with α-SPA2 antibodies. We therefore crossed the SPA2::SPA2-HA transgene into a cop1-5 mutant background and detected the SPA2-HA protein using α-HA antibodies. This transgene-encoded SPA2-HA fully mimics function and behavior of the native SPA2 protein [42] (this study). As shown in Fig 8C and asreported above, SPA2-HA protein levels in the progenitor SPA2::SPA2-HA line decreased to almost undetectable levels upon irradiation with Rc. In a homozygous cop1-5 mutant background, in contrast, SPA2-HA levels were not reduced in Rc when compared to darkness. As an additional control, we also determined SPA2-HA protein levels in COP1 wild-type siblings that segregated in a progeny derived from the cross of cop1-5 with the SPA2::SPA2-HA line. In these siblings, SPA2-HA protein levels decreased upon Rc irradiation as in the progenitor SPA2::SPA2-HA line. Hence, the Rc-induced reduction in SPA2-HA protein abundance was fully dependent on COP1.
Because the cop1 null mutations severely affect seedling growth and cause growth arrest, we wished to exclude the possibility that premature lethality is an indirect reason for the lack of SPA2 degradation in cop1-5. To do so, we made use of previous findings showing that degradation of phyA-Pfr is in part COP1-independent [16,18] and thus should occur in cop1-5. Indeed, phyA levels strongly decreased upon Rc irradiation (Fig 8C). Hence, the cop1-5 tissue used was clearly still capable of light perception and light response. The analysis of phyA abundance indicated that phyA levels were considerably lower in cop1-5 than in the wild type in both dark-grown and light-exposed tissues. The reasons for this are unknown. It may relate to the low efficiency of protein extraction using cop1-5 when compared to using the wild type. Hence, normalization to HSC70 levels may be unreliable. Consistent with this idea, SPA2-HA levels were also unexpectedly lower in cop1-5 than in the wild type.
Since the four SPA proteins heterodimerize in the tetrameric COP1/SPA complex [19], we asked whether the presence of SPA1, SPA3 and SPA4 affects SPA2 protein levels in the light. The light-induced reduction in SPA2 protein levels was also dramatic in the spa1 spa3 spa4 mutant, but slightly higher SPA2 protein levels consistently remained in FRc in spa1 spa3 spa4 when compared to the wild type (Fig 8D). Hence, the COP1/SPA2 complex which forms in the spa1 spa3 spa4 mutant is sufficient to allow SPA2 degradation in the light. Whether SPA2 is required for its own degradation cannot be determined from the presented experiments. The finding that the other three SPAs slightly increase SPA2 degradation in FRc hints at the possibility that SPA2 is involved in its own degradation.
Our finding that COP1 is required for SPA2 degradation in the light suggests that SPA2 is directly ubiquitinated by the COP1 or COP1/SPA2 ubiquitin ligase. If so, it is expected that interaction of SPA2 with COP1 is necessary for SPA2 degradation to occur. To test this hypothesis, we expressed a SPA2 deletion derivative that lacks the COP1-interacting coiled-coil domain under the control of the native SPA2 promoter (SPA2::ΔCC SPA2-HA; Fig 9A). Indeed, the ΔCC SPA2-HA protein failed to co-immunoprecipitate COP1 in extracts of transgenic plants, confirming that ΔCC SPA2-HA does not incorporate into a COP1/SPA complex (Fig 9C). Consistent with this finding, the ΔCC SPA2-HA transgene did not complement the spa1 spa2 spa3 mutant phenotype, whereas the full-length SPA2-HA transgene did (S3 Fig). ΔCC SPA2-HA protein abundance did not change in response to light (Fig 9B). The levels of full-length SPA2-HA, in contrast, decreased to undetectable levels in FRc. This difference in the behavior of the SPA2-HA and ΔCC SPA2-HA proteins is not due to any differences in SPA2-HA and ΔCC SPA2-HA transcript levels because transcript levels were not regulated by light, as expected for a gene expressed from the SPA2 promoter (S4 Fig). These results show that the COP1-interacting coiled-coil domain of SPA2 is necessary for SPA2 degradation in the light.
The four SPA proteins are components of the COP1/SPA E3 ubiquitin ligase and have redundant but also distinct functions in regulating plant growth and development in response to the light environment. The phenotypic analysis of spa mutants showed that SPA2, among the four SPA proteins, exhibits the greatest difference in activity between dark- and light-grown seedlings and is therefore a particularly interesting SPA protein when investigating light-induced inactivation of COP1/SPA activity [22,42]. Here, we have analyzed the molecular mechanism of SPA2 degradation in different light qualities and have uncovered a photoreceptor-specific mechanism of light-induced COP1/SPA repression via COP1.
Our results demonstrate that the SPA2 protein is degraded very rapidly, i.e. within 5–15 min after dark-grown seedlings were exposed to a brief pulse of R, FR or B. Since COP1 function depends on SPA proteins, this rapid, light-induced degradation of SPA2 provides a very effective mechanism to inactivate COP1/SPA2 activity in light-grown plants. We and others have shown previously that COP1 levels do not significantly change in response to R, FR or B [41,42]. Hence, light does not affect the stability of the whole COP1/SPA2 complex but only that of SPA2. This shows that the presence of SPA2 in the COP1/SPA2 E3 ubiquitin ligase provides a means for light-induced inactivation of the COP1/SPA2 complex. Though both phytochrome and cryptochrome photoreceptors inactivate COP1/SPA function in the respective light qualities [14], we found that the rapid degradation of SPA2 specifically required phytochromes not only in R and FR, but also in B. Thus, this mechanism of rapid COP1/SPA2 inactivation is specific to phytochrome action. In summary, our analysis shows that a photoreceptor-specific mechanism of COP1/SPA2 inactivation developed during evolution. Evidence indicates that multiple mechanisms have evolved that inactivate COP1/SPA function in the light. Another mechanism of inactivation was found to be common to phytochromes and cry1 since phyA, phyB and cry1 induce a dissociation of COP1 from SPA1 in R or B, respectively [46–49]. A third mechanism, the light-induced exclusion of COP1 from the nucleus also occurs in R, FR and B and is primarily mediated by phyA, phyB and cry1 in FR, R and B, respectively [52]. On the other hand, B-control of COP1 nuclear abundance was found to also require biosynthesis of the phytochrome chromophore [53], suggesting an essential role of phytochromes also in B. In total, evidence indicates that photoreceptor-specific mechanisms and common mechanisms induced by both phy and cry photoreceptors co-act to allow an appropriate response to a changing light environment.
Our results show that rapid SPA2 degradation in R involves a phyA-dependent VLFR and a phyB-dependent LFR which is also reversible by FR. In FR, SPA2 degradation was fully dependent on phyA. This demonstrates that the responsiveness of the SPA2 protein to R and FR directly correlates with our current knowledge on phyA and phyB activities in R and FR [3,4] and thus appears to be an immediate output of light-induced phytochrome action. Previous findings showing that SPA2 directly interacts with phyA and phyB [49] are in good agreement with this conclusion. phyA is also a well-known B-photoreceptor that together with cry1 and cry2 is responsible for seedling deetiolation in B [4]. The particular biological significance of phyA in B-induced repression of SPA2 function is supported by our finding that the extreme hypersensitivity to B in spa1 spa3 spa4 triple mutants which only have functional SPA2 was indeed fully dependent on phyA. We therefore suggest that light inactivates COP1/SPA2 function in B primarily through rapid, phyA-induced degradation of SPA2. Residual SPA2 protein that escapes degradation may be inactivated by additional mechanisms, such as cry1-mediated dissociation from COP1, as it has been described for SPA1 [46,47], and phyA-mediated dissociation from COP1 [49]. The latter, however, has not been analyzed in B so far.
Since the SPA1 protein is also degraded in R, FR and B, albeit with lower efficiency than SPA2, a SPA1-containing COP1 complex may also be inactivated through phytochrome-mediated degradation of SPA1, i.e. via the same or a very similar mechanism as the light-induced degradation of SPA2. Interestingly, the mutant phenotypes of spa single mutant seedlings defective in SPA1, SPA3 or SPA4 are also fully dependent on phyA, even in R. These single mutants etiolate normally in darkness, but exhibit hypersensitivity in the light in a PHYA wild-type background only [33,54,55]. The mechanistic reason for this observation has so far remained unknown but could be explained by a phyA-mediated de-stabilization of these SPA proteins in light-grown seedlings. Hence, a stabilization of SPA1, SPA3 and SPA4 in a phyA mutant background might lead to the complete rescue of the spa single mutant phenotypes.
The failure of other B receptors than phyA, such as cryptochromes, to cause rapid degradation of SPA2 in B is not due to a general lack of SPA2-cry interactions in vivo. However, our results demonstrate that SPA2 only associates with cry1 and not with cry2 in B-treated seedlings. Hence, the lack of a cry2-SPA2 interaction is likely in part responsible for the observed stability of SPA2 in B-treated phyA mutant seedlings. On the other hand, our results also show that SPA2 rapidly interacts with cry1 in B without causing rapid SPA2 degradation. Based on this finding we conclude that the failure of cry1 to cause rapid degradation of SPA2 is not due to a lack of a SPA2-cry1 interaction, especially since the SPA2-cry1 interaction is observed rapidly in vivo, i.e. within 5 min of B irradiation. Thus, cry1 interacting with SPA2 in B does not induce rapid degradation of SPA2; cry1 action thereby strongly differs from phytochrome actions on the SPA2 protein.
In contrast to SPA2 which only interacted with cry1 in our in vivo co-immunoprecipitation experiments, SPA1 interacted with both cryptochromes, as shown previously [46,47,50]. Hence, SPA1 and SPA2 clearly differ in their interaction capacity with cry2. cry2 was shown to interact with the N-terminal domain of SPA1 [50]. Though we do not know the cry2-interacting domain in SPA2, it is possible that the relatively high sequence divergence between the N-terminal domains of SPA1 and SPA2 might be the cause for their differential interaction capacities with cry2. cry1, in contrast, interacts with the WD-repeat domains of SPA1 and SPA2 [47], and this domain is highly conserved between SPA1 and SPA2 [55].
The mechanism of SPA2 degradation may essentially reflect ubiquitination by the COP1 (or COP1/SPA2) E3 ubiquitin ligase or the action of another E3 ligase. Recently, the COP1-interacting E3 ubiquitin ligase COP1 SUPPRESSOR1 (CSU1) was reported to de-stabilize COP1 and SPA1 in darkness, but not in the light. SPA2, SPA3 and SPA4 protein levels were not altered in csu1 mutants, neither in dark-grown nor in light-grown seedlings [56]. It is therefore unlikely that CSU1 is involved in the light-dependent degradation of SPA2. Indeed, light-induced SPA2 degradation was absent in a cop1-5 null mutant. Hence, ubiquitination of SPA2 by COP1 or the COP1/SPA2 ubiquitin ligase is the likely mechanism. This is supported by our finding that a ΔCC SPA2 deletion derivative which does not interact with COP1 in vivo is not degraded in the light. We therefore propose that light influences the E3 ligase activity of COP1/SPA2 in two ways: it inhibits COP1/SPA2 E3 ligase activity towards its substrate transcription factors, while it enhances COP1 (or COP1/SPA2) (auto)-ubiquitination activity towards SPA2 and, possibly, SPA1 as well (Fig 10). However, we cannot fully exclude the possibility that SPA2 is ubiquitinated by an indirect COP1-dependent mechanism. For example, COP1 might be a scaffolding protein required for SPA2 degradation or control the activity of another E3 ubiquitin ligase. Whether SPA3 and SPA4 protein stability is controlled by light remains to be determined. In humans, DNA damage increases COP1 autodegradation by ATM-mediated phosphorylation of COP1, followed by stabilization of the COP1 substrate p53 as a cell cycle check point [57]. Though the phosphorylated residue in human COP1 is not conserved neither in Arabidopsis COP1 nor in the SPA proteins, this finding shows that autodegradation of components of this E3 ligase is a regulatory mechanism used in both humans and plants.
Wild-type Arabidopsis thaliana accessions Col-0, RLD and Ler were used in this study. Photoreceptor mutants phyA-211 (Col-0) [58], phyA-101 (RLD) [59], phyB-1 (introgressed into RLD) [60,61], phyA-101 phyB-1 (RLD), phyA-201 (Ler) [58], cry1 cry2 (Ler) and cry1 cry2 phyA-201 (Ler) [62] were described previously. The transgenic lines with constitutively active photoreceptors expressed the phytochromes AY242H and BY276H [7], PHYA406-YFP-DD/NLS [10], CRY1G380R [8] or GUS-CCT1 [9]. The transgenic lines SPA2::SPA1-HA 28, SPA2::SPA1-HA 70, SPA1::SPA2-HA 64 and SPA2::SPA2-HA 32 were described previously [42]. The mutants spa1-7 spa2-1 spa3-1 and spa1-7 spa3-1 spa4-1 [51] were used whenever no allele information is provided. spa1-100 spa3-1 spa4-3 [23], cop1-4 [63] and cop1-5 [64] were described. The spa1-7 spa3-1 spa4-1 phyA-211 quadruple mutant was generated by crossing the spa1-7 spa3-1 spa4-1 triple mutant with the phyA-211 single mutant and was confirmed in the F2 and F3 progenies by the phyA phenotype and a genotypic analysis using molecular markers that can distinguish between mutant and wild-type spa alleles.
To obtain SPA2::SPA2-HA cop1-5 (-/-) seed, the transgenic line SPA2::SPA2-HA 32 was crossed with cop1-5 (+/-). Transgenic homozygous cop1-5 seeds were selected in a segregating F4 population based on their black seed phenotype which was scored using a stereo microscope. Seeds with normal seed color served as a control that is homo- or heterozygous for the wild-type COP1 allele.
LED light sources and seedling growth conditions were as described previously [22,54]. Growth conditions for the SPA2::SPA2-HA cop1-5 (-/-) experiment were as follows: after stratification of imbibed seeds for 3 days at 4°C, seeds were irradiated with white light for 3 h to break the dormancy and were subsequently kept in darkness for another 21 h. Seeds were then transferred from darkness to Rc (40 μmol m–2 s–1) for 6 h.
SPA2::ΔCC SPA2-HA lines express a deletion derivative lacking the amino acids 580–702 in the SPA2 protein. To generate the construct, two PCR fragments were amplified from the full-length SPA2 ORF lacking the stop codon using the primer pairs SC_SPA2deltaCC_ApaI_F1 and SC_SPA2deltaCC_R1 or SC_SPA2deltaCC_F2 and SPA2deltaN-NotI-R. Both PCR products were purified, combined and subsequently used as templates for amplifying the ΔCC SPA2 sequence using the primers SC_SPA2deltaCC_ApaI_F1 and SPA2 delta N NotI R, thereby also introducing a 5’ ApaI restriction site and a 3’ NotI restriction site. After PCR-amplification of ΔCC SPA2, the resulting fragment was introduced into the pJET1.2 vector (Thermo Scientific). After sequencing of the insert to confirm the correct sequence, the deletion construct was digested with ApaI and NotI and ligated into the ApaI and NotI sites of the pBS vector carrying the SPA2 5’ and 3’ regulatory sequences as described previously [42], resulting in the SPA2::ΔCC SPA2 construct in pBS. The 3xHA tag with stop codon was subsequently cloned into the NotI site and the complete insert was cloned into the pJHA212 binary vector [65] as described in [42] to generate SPA2::ΔCC SPA2-HA. This construct was transformed into spa1-7 spa2-1 spa3-1 mutant plants by floral dipping. T2 plants were used for analysis.
In order to detect the native SPA2 protein using an α-SPA2 antibody [42], nuclear proteins were enriched from seedlings as described previously [66].
Approximately 200 mg of seedlings or, for cop1-5 related experiments, approximately 20 μl volume-equivalents of imbibed seeds were homogenized to a fine powder using liquid nitrogen. Lysis buffer [50 mM Tris pH 7.5, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 0.1% Triton X-100, 5 mM DTT, 1% protease inhibitor cocktail (Sigma-Aldrich), 10 μM MG132] was added to the ground tissue at a ratio of 150 μl per 100 mg tissue. The mixtures were thawed on ice and centrifuged at 20.000 g at 4°C for 12 min. 5x Laemmli buffer was added to the supernatant to a final concentration of 1x before heating at 96°C for 5 min. Protein concentrations were determined by Bradford assay (Bio-Rad).
For separating nuclear-enriched protein extracts by SDS-PAGE, equal volumes of nuclear-enriched extracts were loaded. To separate total protein extracts, equal amounts of protein were resolved by SDS-PAGE. Protein samples were subsequently blotted onto PVDF membranes. After blotting, membranes were blocked with Rotiblock (Roth) reagent and incubated with the respective primary antibody followed by a horseradish peroxidase (HRP)-conjugated secondary antibody. HRP activity was detected using the SuperSignal West Femto Maximum Sensitivity kit (Thermo Scientific) and visualized by a LAS-4000 Mini bioimager (GE Healthcare Life Sciences). Signal intensities were quantified using Multi-Gauge software (GE Healthcare Life Sciences). Commercial antibodies used were HRP-conjugated α-HA (Roche), α-Histone H3 (Abcam), α-HSC70 (Stressgen), α-α-Tubulin (Sigma-Aldrich), α-rabbit IgG-HRP (Sigma-Aldrich) and α-mouse IgG-HRP (Sigma-Aldrich). α-SPA2 and α-COP1 antibodies were described previously in [42]. α-cry1 [67] and α-cry2 [68] antibodies were used to detect cry1 and cry2, respectively.
Co-immunoprecipitation experiments were performed using μMACS Anti-HA Starting Kits (Miltenyi Biotec) according to the manufacturer’s protocol with minor modification. Total proteins were extracted as described above. Protein lysates were incubated with 10 μl μMACS Anti-HA MicroBeads. After incubation on ice for 30 min, the mixture was applied onto prepared μ Columns which were placed in the magnetic field of μMACS Separator attached to a MACS MultiStand. The columns were washed four times with lysis buffer and once with Wash Buffer 2 provided by the kit. Elution was performed at 95°C with Elution Buffer according to the manufacturer’s manual. For cry1 and cry2 pull-down experiments, seedlings were pre-infiltrated with 100 μM MG132 and 10 μM clasto-Lactacystin β-lactone twice, 15 min each, before light treatment. Furthermore, five times more protein extract was used for the SPA2-HA immunoprecipitation than for the SPA1-HA immunoprecipitation.
Seedlings were flattened on the surface of solid MS plates and photographed with a Nikon D5000 digital camera. Images were analyzed by ImageJ 1.43u (Wayne Rasband, National Institutes of Health) to obtain hypocotyl lengths.
Total RNA isolation, DNase I treatment, first-strand cDNA synthesis and qRT-PCR were performed as described in [42]. Primers used to amplify HA-tag and UBQ10 were previously described [42]. Two biological replicates were included. Relative transcript levels were calculated using the ΔΔCt method with UBQ10 as a normalization transcript.
COP1 (At2g32950), SPA1 (At2g46340), SPA2 (At4g11110), SPA3 (At3g15354), SPA4 (At1g53090), cry1 (AT4G08920), cry2 (AT1G04400), phyA (AT1G09570), phyB (AT2G18790).
|
10.1371/journal.pntd.0006003 | Albendazole and ivermectin for the control of soil-transmitted helminths in an area with high prevalence of Strongyloides stercoralis and hookworm in northwestern Argentina: A community-based pragmatic study | Recommendations for soil-transmitted helminth (STH) control give a key role to deworming of school and pre-school age children with albendazole or mebendazole; which might be insufficient to achieve adequate control, particularly against Strongyloides stercoralis. The impact of preventive chemotherapy (PC) against STH morbidity is still incompletely understood. The aim of this study was to assess the effectiveness of a community-based program with albendazole and ivermectin in a high transmission setting for S. stercoralis and hookworm.
Community-based pragmatic trial conducted in Tartagal, Argentina; from 2012 to 2015. Six communities (5070 people) were enrolled for community-based PC with albendazole and ivermectin. Two communities (2721 people) were re-treated for second and third rounds. STH prevalence, anemia and malnutrition were explored through consecutive surveys. Anthropometric assessment of children, stool analysis, complete blood count and NIE-ELISA serology for S. stercoralis were performed.
STH infection was associated with anemia and stunting in the baseline survey that included all communities and showed a STH prevalence of 47.6% (almost exclusively hookworm and S. stercoralis). Among communities with multiple interventions, STH prevalence decreased from 62% to 23% (p<0.001) after the first PC; anemia also diminished from 52% to 12% (p<0.001). After two interventions S. stercoralis seroprevalence declined, from 51% to 14% (p<0.001) and stunting prevalence decreased, from 19% to 12% (p = 0.009).
Hookworm’ infections are associated with anemia in the general population and nutritional impairment in children. S. stercoralis is also associated with anemia. Community-based deworming with albendazole and ivermectin is effective for the reduction of STH prevalence and morbidity in communities with high prevalence of hookworm and S. stercoralis.
| Soil-transmitted helminth (STH) infections are a relevant public health problem in resource restricted settings due to their potential to perpetuate poverty, since chronic infections are associated with learning and grow impairment in children and reduced productivity in adults. The current strategy for STH control gives a key role to preventive chemotherapy of risk groups (preschool and school age children and women of childbearing age) with anthelmintic drugs. The drugs recommended for regular deworming are albendazole or mebendazole. This strategy does not target Strongyloides stercoralis, an STH resistant to the recommended drugs in single doses. The efficacy of ivermectin, alone or in combination, for the treatment of Strongyloides stercoralis infection has been reported in controlled trials. We conducted a pragmatic study aiming to assess the effectiveness of community based preventive chemotherapy with albendazole plus ivermectin for the control of STH prevalence and morbidity, in endemic communities of Northwestern Argentina. We found high baseline prevalence of hookworm and Strongyloides stercoralis and significant nutritional and hematological morbidity associated with these infections. After three rounds of preventive chemotherapy with albendazole and ivermectin we observed a significant decline in the prevalence of infection and in the prevalence and severity of morbidity.
| Soil-transmitted helminth (STH) infections are the most prevalent Neglected Tropical Diseases (NTD) worldwide affecting over 2 billion people. Four nematode species (Ascaris lumbricoides, Trichuris trichiura, Necator americanus and Ancylostoma duodenale) are the most common STH infections of humans [1,2]. Due to their common biological characteristics and risk factors, geographic overlap and anthelmintic treatment of choice, the recommendations from the World Health Organization (WHO) for STH control target the four species together [3]. Strongyloides stercoralis is an STH with similar distribution but with some distinctive characteristics regarding its diagnosis and therapy that have prevented its inclusion in the current guidelines for STH control; however, it could be targeted with updated comprehensive control strategies [4,5].
STH morbidity is of public health concern because of the population affected, the high prevalence in low income countries and the long lasting consequences of the infection, all contribute to the economic impact of the disease and perpetuation of poverty [6]. Chronic STH infection has been associated with cognitive impairment in school age children (SAC) and negative impact on motor and language development of preschool age children (PSAC) [7–9]. Negative effects of STH infection on nutritional status of children have also been described such as stunting, reduced weight gain and specific micronutrient deficiencies (i.e. iron and vitamin A) [10]. A major consequence of STH infection is iron-deficiency anemia, of significant relevance in infants and pregnancy outcomes. STH infections and other tropical infectious diseases, such as schistosomiasis and malaria, are implicated in the etiology of iron-deficiency anemia in lesser-developed countries [11,12]. Among them, hookworm infection has been strongly associated with the development of iron-deficiency anemia, due to chronic intestinal blood loss [13].
Comprehensive STH control strategies include health education, improvements in water, sanitation and hygiene, and Preventive chemotherapy (PC) through mass drug administration (MDA) of albendazole or mebendazole [14,15]. Implementation of PC for school age children (SAC), the group targeted by the current recommendations, has shown to be effective in reducing worm burden and STH prevalence as much as in improving nutritional status and anemia [16,17]. However, critical groups of the population like PSAC and women of reproductive age are not directly reached by this approach and would benefit if another strategy for deworming such as community-based PC were applied [9,18]. The use of a drug active against S. stercoralis (ivermectin) along with albendazole or mebendazole would enhance the PC effectiveness in the control of S. stercoralis and T. trichiura and would carry additional benefits such as the reduction in the prevalence of scabies and impetigo [5,19–21].
Argentina has a heterogeneous prevalence of STH infection, with areas of high prevalence in the north. Previous studies in Salta province, in the Northwest, showed a cumulative prevalence of STH near 50%, with preponderance of S. stercoralis (20–48%) and hookworm (20–45%) [22,23]. Anemia is a public health problem that affects 18% of the Argentinian population, particularly in the northwest, where 38% of the PSAC and 19% of women between 10 and 49 years old are anemic [24,25].
The aim of the present study was to assess the effectiveness of a community based PC program against STH, using a combination of albendazole and ivermectin. For that purpose, we compared the results of fecal, blood and anthropometric surveys carried out before each PC intervention through a longitudinal community based operational research. The effectiveness of the intervention was monitored through the evaluation of STH cumulative prevalence of the participant communities and variation in morbidity indicators such as anemia and nutritional status of children. The study was part of a larger project with the objective to incorporate a program of PC for STH control into the regular activities of the public primary health care system in high prevalence regions of Salta province, Argentina.
Community-based pragmatic non-randomized trial conducted in Tartagal, Salta province, Argentina between August 2012 and May 2015.
All participants selected for surveillance provided written informed consent prior to the study and parents/guardians provided informed consent on behalf of minor participants. The research protocol and the informed consent forms were approved by the Bioethics Committee of the Colegio de Médicos de la Provincia de Salta and by the Bioethics Committee of the Faculty of Health Sciences at the Universidad Nacional de Salta (FWA registered committee). The anthelmintic drugs were used according to currently approved and recommended regimens.
All members of six communities from Tartagal were invited to participate in the study. A community is defined as the group of people with a common ethnic origin that lives in neighboring households and shares a unique community leader; generally including a group of around 100 households. Four of the communities enrolled in the study were peri-urban: Lapacho Alto, Kilometro 6, Las Moras and Lapacho I; and two communities were urban: Pablo Secretario and Tapiete, These communities are served by the Provincial primary health care system; therefore, trained public health personnel (“sanitary agents”) visited each household every three months. The communities were selected by the local public health authorities based on the sanitary risk indicators collected in the census the year before the study started. Communities with report of malnutrition in PSAC, child or maternal mortality of preventable causes or known STH prevalence above 50% were enrolled. All these communities also share similar water and sanitation conditions [23] and are homogeneous in their economic status, most of the families have a low monthly income that comes from informal and transitory jobs of a single breadwinner and from small economic benefits from the government. A sample was recruited for stool and blood surveillance, using stratified random selection with community as the stratification factor and household as the unit for choice. Inhabitants of the selected households, of any age and gender, were asked to provide a stool and blood samples prior to deworming. Since this is a population-based trial, the subjects recruited for surveillance were not necessarily the same individuals at baseline and follow-up. Nevertheless, the random selection strategy and sample size of the survey group were similar throughout the study. The number of stool samples collected within the survey group was 397 at baseline; 130 at first follow up and 181 at second follow up. Blood samples were 409 at baseline; 156 at first follow up and 165 at second follow up.
All the participants were offered anthelmintic therapy, independent of their age (age ranging from 1 to 91 years old) or recruitment to the survey group. Two anthelmintic drugs in single doses were used simultaneously, albendazole 400mg tablets, dosed at 1 tablet (half the dose in children 12 to 24 months of age) and ivermectin 6mg tablets dosed at 200 micrograms/kg; produced by GlaxoSmithKline and ELEA Argentina respectively. The drugs were administered according to the following inclusion and exclusion criteria. Inclusion criteria: 1) Permanent resident of the enrolled community; 2) Willing to take anthelminthic drugs. Exclusion criteria: 1) Age < 12 months (albendazole); 2) Body weight < 15 kg (ivermectin); 3) Confirmed or suspected pregnancy, contraindication for albendazole in the first trimester and for ivermectin in any trimester; 4) Women breast-feeding new born babies, contraindication for ivermectin in the first week of puerperium; 5) Rejection to take anthelminthic drugs; 6) Known allergy to albendazole or ivermectin; 7) Taking an alternative anthelminthic therapy at the moment of the intervention.
PC was carried out by health personnel and members of the research team who distributed the drugs through intensive deworming campaigns in the communities (house by house) and local schools. Each campaign lasted 4 days per community and was carried out sequentially, due to operative limitations. During the deworming campaigns, all the participants who met all the inclusion and none of the exclusion criteria received the anthelmintic therapy. Passive pharmaco-vigilance activities were carried at the local tertiary hospital (Hospital Juan D. Peron) and the respective sanitary posts of ambulatory and emergency services were aware of the interventions. The first intervention took place in August 2012 and the last one in May 2015; during that period, some communities were treated once and others received three rounds of PC with an interval of 9 to 16 months between rounds. Stool and blood surveillance were performed at baseline and follow-up (before each round of PC). The results of the interventions were entered in the data base for each participant as: treated (if the subject was administered one or two drugs); excluded (if both drugs were contraindicated); absent (if the subject was not found at home during the deworming round or the household could not be found); death; migration (if the subject moved from the community before the intervention) and rejection (if the subject did not want to take the anthelminthic drugs when offered). Coverage was calculated through the formula: n people treated/n people eligible for treatment and eligibility was calculated as: total population–(excluded + death + migrants).
The following data were gathered for the study: a) baseline socio-demographic information; b) baseline and follow-up anthropometric data; c) baseline and follow-up stool results; d) baseline and follow-up hematologic results, and e) baseline and follow-up serologic results for S. stercoralis. The methods used for data collection are detailed below:
a) Socio-demographic assessment: public health census forms include data about each household of the community, collected through direct observation by the sanitary agent, along with individual demographic information for each inhabitant of the household. Demographic information was entered in the study database using the sanitation and drinking-water ladders designed by WHO/UNICEF for sanitary census [26].
b) Nutritional survey: during the deworming visits, members of the research team registered weight and height of children from one to fifteen years-old. Weight was measured using a standard electronic scale in kilograms and grams. Height was measured with a steel tape while the child was standing near to a wall and registered in centimeters. In order to make the results comparable, population based sampling was used; the number of baseline height and weight observations was above the minimum sample of 400 suggested by WHO for nutritional surveys. Results were compared to the NCHS/WHO reference population and the expected ranges of standard deviations of the anthropometric indicators were considered to control the accuracy of the measurements. Anthropometric data was analyzed through WHO Anthro and WHO Anthro Plus softwares (Department of Nutrition, WHO) to calculate relevant Z-scores. Weight, height, Weight-for-Age z-score (WAZ), Height-for-Age z-score (HAZ) and Weight-for-Height z-score (WHZ) of each participant child at baseline (before the first PC round) and follow-up (before second and third PC rounds) were registered. Children were classified as stunting (HAZ <-2 SD from the international reference median value); underweight (WAZ <-2 SD from the international reference median value) or wasting (WHZ <-2 SD from the international reference median value) according to the WHO recommendations on z-scores interpretation [27].
c) Parasitological surveillance: a single fresh stool sample without preservatives was collected from each participant of the survey group. Sterile stool containers and instructions were distributed house by house, collected the following morning and analyzed within 24 hours of collection in a reference laboratory. Five parasitological techniques were used: sedimentation/concentration; agar plate culture; Harada-Mori filter-paper culture; Baermann concentration of charcoal-cultured fresh stool, and McMaster egg counting method as described elsewhere (28,29). If the sample volume was insufficient to perform all methods, concentration technique was prioritized due to its overall higher sensitivity in preliminary studies [28]. A high level of certainty in the distinction between hookworm and S. stercoralis larvae in the culture techniques was assured by the different incubation time of the methods (24 hours for Baermann and 7 days for Harada-Mori), adapted to each specie life cycle, and by the long experience of the technicians who observed and supervised the microscopic exams. The findings of the different methods were grouped and entered in the database as positive if at least one method was positive or negative if all the methods were negative, for each STH species. McMaster´s results were recorded as egg per gram (EPG). The stool survey was performed at baseline and before each PC campaign.
d) Hematological surveillance: participants of survey group had 5 mL blood drawn through venipuncture. A complete blood count was performed using a SYSMEX automated hematology analyzer KX 21N. The results of Hemoglobin value (Hgb), white blood cell count (WBC) and eosinophil relative count were registered. Subjects were classified as anemic or not anemic using the Hgb thresholds to define anemia according to sex and age set by WHO/UNICEF [12]. Absolute eosinophil count was calculated and eosinophilia was defined as absolute eosinophil values >500 cells/mm3 [29].
e) Blood samples were centrifuged and an aliquot of serum was preserved frozen at –20°C and analyzed with the in-house enzyme-linked immunosorbent assay (NIE-ELISA) method for the diagnosis of S. stercoralis. NIE-ELISA detects IgG antibodies against recombinant NIE antigen of S. stercoralis L3 larvae, as has been described previously [30,31]. Patient’s sera were tested in duplicate and compared to a standard positive IgG curve run on each plate. The averages of duplicate results were calculated and corrected for background reactivity (no serum added).
The primary outcome was STH prevalence. The cumulative and species specific STH prevalence before and after deworming was compared to monitor the effectiveness of community-based PC. Secondary outcomes were indicators of morbidity potentially due to STH infection such as anemia, eosinophilia and nutritional impairment.
The study was analyzed in two phases: i) baseline assessment that included a description of the study population targeted with PC and an exploration of the survey group at the individual level, searching for associations of STH infection with nutritional and hematological findings; and ii) longitudinal assessment to evaluate the impact of PC on population-related parameters in the communities where repeated PC was carried out.
Sample size was estimated considering a predicted prevalence of 50%, a confidence level of 95%, and a design effect of 2. Sample size calculation was based on observing a specific reduction of 10% or more in STH prevalence; the estimated sample size was n = 190.
Continuous quantitative measures evaluated at different time points were described using proportions with 95% confidence intervals (95% CI); means with standard deviations (SD) and medians with interquartile ranges (IQR). Comparisons between infected and uninfected people at baseline and between pre and post intervention parameters were carried out using T test and Mann-Withney U test. Significant associations between STH infection and morbidity indicators were explored through stratified bivariate analysis and, afterwards, adjusted through multivariate logistic regression models; statistical significance was assessed by Chi-square test with 95% significance. Correlation between continuous quantitative measures was explored through linear regression tests of Pearson´s or Spearman´s (according to the underlying distribution).
All data was entered in Microsoft Access 11.5 (Microsoft, Redmond, WA) with an Epi Info 3.5.4 (CDC, Atlanta, GA) view. Duplicate data entry was performed by trained collaborators. The analysis was performed with EPIDAT 3.1 (PAHO, Washington, DC) and R 3.1.1 (The R Foundation for Statistical Computing, GNU General Public License).
A total of 5070 inhabitants of six communities were enrolled in the study, 400 of them were recruited in the survey group at baseline. Two communities: Kilometro 6 and Lapacho Alto received three deworming interventions, therefore a baseline and two follow up surveillances were carried out. Two other communities: Las Moras and Lapacho I, were studied and treated with anthelminthic only once due to operational issues (geographic isolation, long distance between households and lack of collaboration from public health personnel responsible for the area) that caused a poor coverage in the first intervention and prevented subsequent interventions. In two communities: Tapiete and Pablo Secretario, a first school intervention was made and no successive deworming was applied because the baseline prevalence was considered too low (< 20%) to require regular PC. Baseline assessment was performed taking into consideration the results of the six communities, while for the longitudinal analysis only the two communities (n = 2685) where baseline and follow up surveillance were performed were included. Fig 1 shows the flow diagram of the enrollment of participants in the study.
We found a heterogeneous distribution of STH prevalence between communities varying from 11% to 72%, urban communities (Pablo Secretario and Tapiete) had a significantly lower prevalence than peri-urban communities (Lapacho Alto, Kilometro 6, Lapacho I and Las Moras), p<0.001. Table 1 details the baseline prevalence of STH infections and the coverage reached with the first intervention in each one of the studied communities.
Among the 6 communities, hookworm (34%) and Strongyloides stercoralis (26%) were the most frequent parasitological findings. Regarding hookworm, species identification was done through Harada-Mori in 73 of the 135 positive cases: 63 were A. duodenale; 9 were N. americanus and 1 was a co-infection of both species. The other 62 cases were negative in Harada-Mori. Therefore, A. duodenale accounted for 86% of the hookworm cases identified by species. A total of 104 S. stercoralis cases were diagnosed, 53 of them were detected by parasitological exam, 53 were detected by NIE-ELISA serology and 18 were positive by both methods. Table 2 summarizes a baseline description of the study population and the findings in the survey group.
We found significant differences in mean hemoglobin level, median eosinophil count and mean HAZ according to STH infection status at the individual level (uninfected versus infected subjects). No significant differences were found in mean WAZ for children aged 1 to 15 years old and WHZ of children aged 1 to 4 years old. Table 3 displays the comparison of hematological and nutritional results between infected and uninfected groups.
Hookworm infection was significantly associated with anemia (odds ratio [OR] = 5.5; 95% CI: 2.7–11.4); eosinophilia (OR = 7.48; 95% CI: 3.0–20.8) and stunting (OR = 1.7; 95% CI: 0.8–3.8). S. stercoralis infection was also significantly associated with anemia (OR = 3.5; 95% CI: 1.7–7.1) and eosinophilia (OR = 2.3; 95% CI: 1.2–4.9) but not with stunting (OR = 1.01; 95% CI: 0.4–2.3). Fig 2 displays the adjusted odds ratios after controlling for potential confounders through multivariate logistic regression models.
We found a linear negative correlation between hookworm intensity, calculated as EPG and hemoglobin level. Non-parametric Spearman´s coefficient of correlation (rs) = -0.46; p< 0.001. Fig 3 shows the scatter plot of this correlation. It should be noted that the frequency of heavy and moderate intensity infections was low with most cases having light infections. However, even light egg burden was significantly associated with anemia since 65% (95% CI: 48–81) of the subjects with light hookworm infection were anemic compared to 12% (95% CI: 5–19) of uninfected subjects.
Lapacho Alto and Kilometro 6 inhabitants (n = 2685) received three community deworming interventions. We found significant improvements in the prevalence of STH infection, anemia, eosinophilia and children´s chronic malnutrition (stunting) after community-based PC with albendazole and ivermectin. Table 4 summarizes the indicators measured at baseline (before first PC) and at follow-up (before second and third interventions).
We found a significant decrease in the cumulative prevalence of STH infection and in the species-specific prevalence of hookworm and S. stercoralis after the first intervention. Hookworm infection intensity decreased with deworming, showing fewer moderate and no heavy intensity infections at follow-up. We were not able to calculate hookworm’s Eggs Reduction Rate (ERR) after deworming for several reasons that attempted against the validity of this indicator. The number of stool samples collected at baseline and follow up was below the recommended sample size of 200 for this type of calculation[32]. Follow up infection intensity was assessed much later than the recommended two weeks after PC; therefore, reinfection could have occurred between baseline and follow-up assessments. Not all the samples that tested positive for hookworm through other techniques showed eggs to be counted through McMaster method, suggesting that those cases correspond to low-intensity infections. [32]. S. stercoralis´s seroprevalence showed non-significant changes after the first intervention but diminished significantly after the second intervention. Fig 4 shows the evolution of relative IgG titers across the study.
Anemia and eosinophilia prevalence and severity showed a significant response to the first intervention. When comparing hemoglobin levels between baseline and follow up surveys we found that mean hemoglobin level increased with deworming but also the curve shifted to the right showing higher minimum and maximum levels and eliminating the most extreme cases of anemia. Age and gender distribution of anemia also showed a modification following deworming. At baseline the most affected groups were PSAC and adult women but SAC and adult men were affected as well. In follow-up surveys the prevalence in PSAC, SAC and adult men declined to low levels, while adult women remained anemic; although with lower prevalence (Fig 5).
We found an impact of deworming on stunting prevalence after two deworming campaigns but no changes in underweight and wasting, which were infrequent already at baseline. The severity of the deficit in height showed improvement after one PC intervention: baseline mean HAZ = -0.93 (SD = 1.2) was statistically different than the measurement at first follow up which had a mean HAZ = -0.72 (SD = 1.3) (p = 0.009); although it should be noted that mean HAZ remained negative after PC. When discriminating the evolution of HAZ in PSAC from SAC, we found a significant difference in PSAC between baseline and first follow up and between first and second follow up. While SAC, showed no difference in HAZ after deworming. However, the number of HAZ observations within each group was below the minimum of 400 recommended to perform comparisons. Fig 6 shows the evolution of HAZ values in PSAC and SAC.
The current WHO strategy for the control of STH as a public health problem relies on the use of PC with albendazole or mebendazole for those groups and countries that carry the heaviest burden of disease. The long term and sustainable solution for the STH problem probably remains more closely linked to the provision of water and sanitation. Such an approach will impact several NTDs while addressing broader goals of shared prosperity and equity [33]. While aiming for coverage indicators to achieve public health and morbidity goals, evidence supporting the relationship between coverage and morbidity in STH is still incomplete. With the understanding that WHO recommendations are a minimum set of goals for resource limited settings, this report describes a strategy for STH control in resource-limited communities with significant deficits in sanitation, where a primary care public health system is in place and Strongyloides stercoralis has been described at high prevalence [23,28], through a comprehensive baseline survey and a community based intervention with albendazole-ivermectin, achieving significant impact on morbidity indicators.
The significance of STH as a public health problem in the study region was confirmed in this study both in terms of prevalence and morbidity indicators [22,28], with a baseline prevalence of 48% (IC95% 42–53%). Most of the prevalence was due to hookworm and S. stercoralis. We believe that the scarcity of A. lumbricoides and T. trichiura infections in our study population might be due to the wide distribution of piped water in the households, which might have a protecting effect against orally acquired infections. Similarly, the high prevalence of hookworm and S. stercoralis found might be related to the lack of sanitation facilities in most households, since this could increase the risk of infection by skin-penetrating species. This selectiveness of the risk factors (unimproved sanitation and unimproved water access) to the route of transmission of each STH species has been explored in a previous study of our group [23]. An alternative explanation for the low frequency of A. lumbricoides and T. trichiura infection could be the issue that we used only two methods for the detection of these species, compared to the three or five methods used for the diagnosis of hookworm and S.stercoralis respectively. However, the sensitivity of the two methods that detected eggs (sedimentation/concentration and McMaster) is appropriate (around 90%) as reported in other studies [34,35]. As a matter of fact, the use of Harada-Mori did not add new diagnosis of hookworm to the findings of concentration, this method only contributed with the specie identification; while additional cases of S. stercoralis, that went unnoticed with concentration, were detected through Baermann or agar plate cultures, suggesting an adequate sensitivity of sedimentation/concentration for the detection of eggs but a limited capacity to find larvae.
The prevalence of anemia of 31% puts these communities in the category of “Severe public health significance” according to WHO definitions [12]. Anemia was significantly associated with hookworm and S. stercoralis infection in the baseline cross-sectional analysis. Even though the design of the study is not proper to demonstrate causality, the existence of previous evidence linking STH infections (especially hookworm) with anemia, explained by already well known pathogenic mechanisms [13,36–38]; added to the fact that hematologic parameters improved in the population after the intervention with anthelminthic (and nothing else) make us infer a probably causal relationship between infection and anemia. Guidelines for anemia and STH both recommend deworming as a significant although not sufficient measure to improve the public health situation of communities with high prevalence of anemia[11,12,39].
Impact indicators following PC demonstrated significant reductions in terms of prevalence and morbidity. Statistically significant changes in STH and anemia prevalence were both achieved after a single intervention. Interestingly, the residual anemia in the post intervention assessment was concentrated in adult females, the group that has a significant complementary cause of anemia and depleted iron stores due to menstrual blood loses; still, this group also benefited from the intervention lowering the prevalence of anemia. These results highlight the vulnerability of this group and the need for STH control measures that target women of childbearing age [40,41]. In accordance with other studies showing intervals of up to a year, S. stercoralis prevalence measured through NIE-ELISA only showed significant decreases in prevalence after 2 rounds of MDA [42].
The use of a comprehensive diagnostic panel allowed the identification of S. stercoralis as a significant pathogen and its morbidity, justifying the choice of the selected drug regimen. Should the diagnostic approach have been limited to a quantitative egg detection method, like Kato-Katz, McMaster`s or Mini-FLOTAC, this species would have gone completely unnoticed. Another significant observation from this diagnostic approach is the recognition that low burden hookworm infections carry a significant burden of anemia (Fig 3), which is different than in previous reports [43]. If confirmed, these results should stimulate a reassessment of the current approach that links morbidity solely to moderate and high burden infections. This finding is possibly related to co-existing conditions present in these communities, such as inadequate iron dietary intake and depleted iron stores of the subjects, which increases the likelihood that even relatively small chronic blood losses cause anemia; therefore, our results might not be generalizable.
The impact on growth indicators in children, which are probably multifactorial, showed that height-for-age z-scores (HAZ) were significantly lower in STH infected children at baseline; moreover, as indicated by WHO nutritional guidelines [27], the fact that the mean HAZ is below zero suggests that the whole population is affected or at risk. Stunting, as the nutritional parameter indicating more severe compromise of HAZ, was significantly associated with hookworm infection and underwent significant improvement after the second (but not the first) drug intervention, possibly indicating the interval of time free of STH needed to observe changes in this growth parameter. Even though the mean HAZ significantly increased after two PC interventions it remained negative suggesting that a chronic nutritional impairment persisted despite deworming; these might be related to the relative short time of follow up or to concomitant causes of malnutrition such as deficient dietary intake. We were not able to demonstrate any association of hookworm or S. stercoralis infection with weight for age z-score (WAZ) and weight for height z-score (WHZ), or impact of deworming on these parameters.
The association of STH infection with anemia and nutritional impairment needs to be considered in the context of possible confounding factors such as family income. Poverty is a well-known risk factor for STH infection but may also be a cause or contributing cause of anemia and malnutrition, since poor families tend to have diets insufficient in quantity and quality with low iron content. We did not include poverty in the multivariate analysis to confirm or discard a confounding effect in the association found between hookworm or S. stercoralis infection and anemia or malnutrition because data of family income was unavailable. However, the homogeneity of the socio-economic status of the study population makes us infer that the low family income was similarly distributed in the infected and uninfected groups.
Our study has limitations to be considered: first, the surveillance samples at baseline and follow-up are not matched samples; therefore, we compared the results before and after PC by community and not individually. In a recently published article, we demonstrated in a smaller group in a community project close to the study area, with paired samples, that after a year of follow-up, decreases in NIE-ELISA antibody titers were similar to those observed in this study [44]. Second, some communities that were studied at baseline and treated for the first time were not surveyed at follow up and did not received successive deworming; we included these communities in a cross sectional baseline assessment looking for associations of STH infection but for the prospective analysis only the communities where follow up surveillance was performed were compared. Third, false negative results could occur due to suboptimal sensitivity of the diagnostic methods, which can overestimate the impact of deworming since a reduction in burden might be read and interpreted as a negative result. To deal with this limitation, various diagnostic methods were used. More sensitive methods such as PCR for the detection of STH might overcome this limitation [45]. Fourth, some self-selection bias might have happened related to the fact that some individuals selected for surveillance refused to participate and those among the selected that provided samples for the study might be at higher risk of infection. Fifth, the coverage reached with the interventions was not optimal, which could underestimate the real effectiveness of the drug regimen. Finally, we do not report the impact of deworming on infection intensity; due to the limitations of our study for the estimation of ERR, even at group level. Most of our limitations are typical of population-based trials in a resource-restricted setting. Despite these limitations, the analysis performed, including multivariate analysis, and the strength of the associations and effects, in the context of no other changes in conditions than the PC, provide the foundations for our interpretations regarding the benefits of PC with albendazole and ivermectin.
As previously mentioned, we found difficulty in conducting PC in some communities where the collaboration of the sanitary agents responsible for the areas was meager and were geographically isolated which resulted in low coverage rates and the discontinuation of the study in those communities. For those situations, a school based intervention, as recommended by WHO [46], is a more feasible option, even though with more limited results in terms of morbidity reduction since women and preschool age children are not targeted.
The drug regimen used in these interventions has been used in a few clinical trials for the treatment of STH, mostly aiming at improving efficacy for T. trichiura [20,47], however most of the data regarding its use, safety and even impact on STH, comes from lymphatic filariasis (LF) programmes in areas with onchocerciasis, where millions of individuals have been treated, highlighting the safety of this regimen [48,49]. Since currently most PC programs are dependent on drug donations, the addition of IVM needs to be considered in the context of the critical issue of drug availability.
This is, to our knowledge, the first report of an intervention of PC with albendazole-ivermectin for pure STH control; its positive short-term effects on morbidity are echoed by recent reports proposing this drug combination as a feasible strategy for delaying the emergence of resistance and offering improved efficacy against T. trichiura and S. stercoralis in clinical trials and mathematical models [20,50–52]. In summary, albendazole/ivermectin applied as community based PC in communities with high prevalence of hookworm and S. stercoralis resulted in significant reduction of STH prevalence and improvements in anemia in the general population and nutritional status in children. These results should be further explored in randomized-controlled trials with cost-effectiveness evaluation to overcome the limitations of our study.
|
10.1371/journal.pcbi.1000012 | Representing Where along with What Information in a Model of a Cortical Patch | Behaving in the real world requires flexibly combining and maintaining information about both continuous and discrete variables. In the visual domain, several lines of evidence show that neurons in some cortical networks can simultaneously represent information about the position and identity of objects, and maintain this combined representation when the object is no longer present. The underlying network mechanism for this combined representation is, however, unknown. In this paper, we approach this issue through a theoretical analysis of recurrent networks. We present a model of a cortical network that can retrieve information about the identity of objects from incomplete transient cues, while simultaneously representing their spatial position. Our results show that two factors are important in making this possible: A) a metric organisation of the recurrent connections, and B) a spatially localised change in the linear gain of neurons. Metric connectivity enables a localised retrieval of information about object identity, while gain modulation ensures localisation in the correct position. Importantly, we find that the amount of information that the network can retrieve and retain about identity is strongly affected by the amount of information it maintains about position. This balance can be controlled by global signals that change the neuronal gain. These results show that anatomical and physiological properties, which have long been known to characterise cortical networks, naturally endow them with the ability to maintain a conjunctive representation of the identity and location of objects.
| Forming a coherent picture of our surrounding environment requires combining visual information about the position of objects (where information) with information about their identity (what information). It also requires the ability to maintain this combined information for short periods of time after the stimulus is removed. Here, we propose a theoretical model of how this is accomplished in the brain, particularly when sensory input is incomplete, and missing what information should be supplied from what is stored in memory. The main idea is that local connectivity in cortical networks can allow the formation of localised states of activity. Where information can then be represented by the position of such “bumps”, and what information by the fine structure of the neuronal activity within them. We show that there is a difficulty with implementing this idea: noise and heterogeneity in connectivity cause bumps to drift, thereby losing where information. This problem can be solved by incorporating a localised increase in neuronal gain; this, however, interferes with retrieving what information and maintaining it in working memory. We quantify this interference via theoretical analysis of the model and show that, despite the interference, the proposed mechanism is an efficient one in retrieving what information while representing where information.
| Visual object perception, which is often effortless despite partial occlusion or changes in view, shading, size, etc., has been associated to attractor dynamics in local cortical circuits [1]–[5]. A single pattern of neuronal activity would be associated with an object, and retrieved when an input cue engages the corresponding basin of attraction. This would lead to a distribution of activity over a cortical patch that can be read out by other areas and can persist even after the object is removed. Attractor dynamics can be realised in neuronal networks by Hebbian modifications of synaptic weights on the recurrent connections of a local population of cortical neurons [6]. The experimental observation of persistent activity in monkey prefrontal cortex (PFC) [7]–[9] and inferior temporal cortex (IT) [10]–[12] during memory related tasks supports the idea that attractor dynamics is involved in such tasks.
The above-mentioned paradigm is conceptually very successful in explaining how information about the identity of an object can be retrieved from noisy input and maintained in working memory, even when the input is transient. However, in day to day life, the identity of an object is hardly the only type of information that one needs to retrieve and maintain about it. If you look at a scene for a short time and then turn your head away, you will still remember details about what objects were present in the scene and where they were located. You can even do this if many of the objects in the scene were occluded. These abilities allow us to maintain a coherent representation of our surrounding environment and are crucial for most real world visually guided behaviours. Visually guided behaviour often requires extracting information about identity of objects (what information) from noisy sensory input, and combining this what information with information about the position of objects (where information). It also requires maintaining this combined representation of position and identity of objects in working memory after the visual input is removed. The underlying neural mechanisms for these abilities are, however, unknown. In this paper, we analyse a network model of how this may be accomplished in the brain.
A great deal of experimental work has been focused on understanding this issue [13]–[18]. Single cell recordings from PFC during the delay period of a delay match to sample task show that neurons in this area can maintain information about the conjunction of position and identity [13],[14]. Rao and colleagues [13] also found that some PFC neurons can change their selectivity from conveying what information to conveying where information when the type of information that is required by the task is changed. Selectivity for object-position pairs is further supported by the presence of retinotopically organised maps in PFC regions that are involved in identity working memory tasks [16]. Furthermore, a recent neuroimaging study by Sala and Courtney [17] shows that dorsal and ventral PFC can maintain an integrated representation of position and identity when it is relevant to the task, but represent position or identity when only one of them is task relevant. Although most studies that address the issue of combining what and where information have focused on PFC, similar observations have been reported in IT. While some studies report a considerable position invariance in the response of IT neurons [19]–[21], this view has been challenged by others. More recent studies show that IT neurons can have small receptive fields and can convey detailed information about the position as well as the identity of objects [22],[23]. Furthermore, it has been reported that the receptive fields of IT neurons are much smaller in natural scenes when compared to plain background and are closer to the fovea, thus conveying increased spatial information in their response [24]. Consistent with these properties, Hung et al [25] have shown how, using a simple linear classifier, spatial position can effectively be read off the response of IT neurons. Neuroimaging studies also show that temporal visual areas, just like V4 [26], can be involved in processing the spatial information of objects as well as their identity [15]. Although these neurophysiological studies have not directly assessed the ability of IT neurons to maintain combined what and where information after removing the stimulus, the possibility should be considered that, like PFC, delay activity in IT can also transmit where information in addition to what information. The degree to which the neural code in IT and PFC is committed to one versus the other most likely depends on task requirements, attention or learning [15],[17],[22],[23],[27],[28].
In this paper, we study how a recurrent network can retrieve what information from noisy/transient input, while simultaneously representing where information. In the model that we present here, we consider a recurrent network embedded in a two dimensional tissue, and to each object associate a single discrete pattern of neuronal activity. These patterns do not have any spatial preference and are stored in the synaptic weights of the recurrent connections trough Hebbian learning. We show that, when the connectivity between neurons is metrically organised (that is, nearby neurons are more likely to be connected than those far apart) [29]–[31], the network can retrieve these patterns in a spatially focused way by maintaining localised retrieval states (or “retrieval bumps”), similar to what has been previously studied in one dimensional networks [32],[33]. A localised retrieval state is a stable and localised pattern of activity which has a high correlation with one of the stored patterns, but low correlation with the others. The idea that we elaborate here is to use the position of the bump to represent the position of the object, while the distribution of activity inside the bump represents its identity (In this paper, when we say that a pattern of neuronal activity “represents” a variable, we mean that that variable can be decoded from the pattern of activity). In this way, ideally a continuum of firing patterns would represent the object in different positions. The difficulty in implementing this idea, as we show, is that the retrieval bumps cannot be localised at any target position on the surface of the network, but rather on a limited number of discrete positions. To resolve this problem we need to introduce some additional mechanisms. We show that it takes small modulations of neuronal gain to stabilise the bump on arbitrary positions on the tissue. The gain modulation can be provided or at least initiated by the cue that initiates the retrieval of the pattern, or it can be provided by other areas, e.g. in the form of attentional signals [34],[35]. Importantly, stabilising the bump at a given position through such gain modulation affects the process of retrieving what information from stored representations. We quantify this effect and show that it can be negative, that is a trade-off between the representation of what information and where information, or it can be positive. Whether the effect is positive or negative depends on the average neuronal gain. When neuronal gain is high the effect is negative but it is positive when the gain is low. Moreover, when the cue is incomplete (that is when the stimulus is noisy or occluded) localising the cue in the gain modulated part of the network further helps identity retrieval. We finally discuss the possibility of retrieving multiple patterns, in the form of multiple bumps of activity. The distribution of activity inside each bump again reflects the identity of the corresponding object, and it can in principle be maintained in working memory while a serial attentional mechanism facilitates retrieval of another memory pattern at a different position.
In what follows, we show an example of a retrieval bump in a 2D recurrent network with metrically organised connectivity. We consider a two dimensional network containing N = 4900 neurons in total. The neurons are arranged on a regular lattice with 70 neurons on each side and distance l between neighbouring sites. The connections between neurons have a metric structure: a neuron in position ri is connected to a neuron in position rj with probability(13)
In the simulations reported here the width of the connectivity, σ, is set to 7.5l. Since l is the distance between two adjacent neurons, this means that the probability that two adjacent neurons are connected to each other is ∼0.7. Experimental data estimate this probability to be 0.5–0.8 [30]. The gain of all neurons in the simulations reported in this section is set to a background level g = 0.5.
At the beginning of the simulation a 15×15 square centred on the neuron in position (58,58) is chosen. The activity of neurons inside this square are initialised to their activity in the first stored pattern while the activity of other neurons are set to zero, that is in the beginning of simulation if ri is in the square and νi = 0 if ri is outside it. In this way at the beginning of the simulation the dot product overlap with the first pattern and the others have the following values
Fig. 1A shows the local overlap with the cued pattern (μ = 1) at the beginning of a simulation. The local overlap (Eq. (8)) with the cued pattern after 200 synchronous updates is shown in Fig. 1B and the distribution of activity {νi}, also after 200 time steps, is shown in Fig. 1C. We see that the activity of the network is concentrated on a part of the 2D network, and so is the local overlap. The important point is that this final pattern of activity has a high dot product overlap with the cued pattern but not with other stored patterns, i.e.
Thus by calculating these dot products, or equivalently calculating the sum of the local overlaps miμ over i, in the end of the simulation we can say which pattern was presented, i.e. in this example the first pattern.
The spatial distribution of activity would have been different (Fig. 1D), if instead of the probability distribution in Eq. (13), we had used a uniform distribution
In this case, too, by cueing one of the patterns, as we did for the metrically organised network, after 200 time steps, we have m1(t = 200)≈0.8 and mμ≠1(t = 200)≈0, thus indicating retrieval of the pattern. The difference between the two connectivity models emerges, however, in the final distribution of activity. Whereas in Fig. 1D the activity is distributed uniformly across the network (at a gross spatial scale, since at a fine scale individual units are activated in relation to their selectivity for the cued pattern), in Fig. 1C the activity is localised over a portion of the 2D network. So, metric recurrent connections, as predicted by the mathematical analyses of attractor states and as confirmed by many other simulations, allow activity to stabilise in spatially modulated distributions.
Even though Fig. 1 shows the possibility of localised retrieval in the network with the Gaussian connectivity in Eq. (13), a critical observation is that in Fig. 1B the final local overlap is in a different position than the initial cue (Fig. 1A). The trajectory that the peak of the local overlap follows during the retrieval process is shown in Fig. 2. The green square shows the peak at the beginning of the simulation, before any updates take place (Fig. 1A), and the red circle shows the peak after 200 time steps (Fig. 1B). It is clear that, during retrieval, the “bump” of activity drifts away from its initial position. This raises the question addressed in this paper, of whether where information in the cue can be preserved by spatially modulated attractor states.
Can we code the position of an object by the position of the peak of the retrieval bump? The answer to this question depends on whether the retrieval process can end with the peak of the bump on the intended position. We first examine whether the position of the cue (which can be thought of as the position of an object in the retina) determines the positions of the retrieval bump. If the retrieval bump appears at the same position as (or is uniquely determined by) the centre of the cue, it is possible to read the activity of the network and simply decode both what information, that is, which cue has been presented (the pattern with the highest overlap with final activity), and, in addition, where it has been presented: object position is simply coded by the position of the centre of the bump.
To examine the relation between the position of the initial cue and the final position of the retrieval bump, we ran simulations in which the position of the initial cue was systematically changed across the network and the distance between the position of the retrieval bump and the position of the initial cue was measured. In Fig. 3, we summarise the results from simulating a network of 70×70 neurons with the Gaussian connectivity pattern Eq. (13) with σ = 7.5l, as used in Fig. 1. At the beginning of each trial, the first pattern was cued by initialising the activity of neurons in the following way: , if neuron i was within a 15×15 square, whose centre was varied, across trials, over the entire network; while νi = 0, if neuron i was outside the square. The activity of all neurons was then synchronously updated for 200 time steps and the local overlap with each pattern was monitored. Fig. 3A shows that the position of the bump at the end of each trial (red circles) does not match the peak of the local overlap with pattern 1 at the beginning of the trial (green squares). The bump drifts away from its initial positions, and stabilises on one of, in this particular case, 4 final positions. This small number of final stable positions indicates that one cannot decode from the final position of the retrieval bump where the cue was located, at least not with high accuracy. In fact, by looking at the final position of the bump, one might say whether the initial position of the cue was among the 23 initial positions that converge to the upper left red circle or among the 10 initial positions that converge to the lower right red circle, but nothing more.
The small number of final stable positions of the bump resembles what has been noticed before in recurrent networks with distance dependent weights between neurons but without stored memory patterns. In such models the synaptic weight between two neurons is generally taken to be excitatory at short distances while inhibitory at long distances [51]–[58]. The distance between two neurons in these models can be anatomical distance, or distance, in the feature space, between the features that the neurons are selective for. Models of this type have been used to conceptualise how local networks of orientation selective neurons in visual cortex [52], head-direction neurons [53], location selective neurons in prefrontal cortex [54] and hippocampal neurons [57],[58] can maintain selectivity after the external input has been removed. Studies on rate based models [51]–[53] as well as networks with spiking neurons [54]–[56] show that, under very mild conditions, the stable activity profile of these networks is of the form of a localised “bump”. If the network is strictly homogeneous, the bump can potentially exist anywhere on the network, and it can be smoothly moved from one position to the other. Any small inhomogeneity in the network, however, fractures the continuum of solutions, which therefore represents an ideal limit case, and stable bumps are allowed only at a number of discrete positions [53], [57], [59]–[61]. Coming back to the associative memory network with metric connectivity, it is clear that inhomogeneity is an unavoidable part of its overlaid memory structure. Synaptic weights are required to be different from each other in such a network, to support the retrieval of memory patterns, a situation where a neuron can be active while its nearest neighbour is inactive. As a result, a retrieval bump in our model cannot be maintained at any arbitrary position on the network.
Even though the final position of the bump cannot accurately tell where the cue was initiated, it may still be able to code for a large number of positions in a network with realistic size. This happens if the number of final stable positions increases with the size of the network. To examine this relation, we scaled up the simulated network. The result of such scaling analysis is reported in Fig. 4, which shows the number of final positions resulting with different network sizes, while keeping the number of connections and the width of connectivity constant. One sees a roughly linear increase in the number of stable bump positions.
The approximately linear scaling of the number of final positions with network size indicates that a large number of positions can be represented in realistically large networks, but not any arbitrary position: with our regular 2D lattices and our parameters, the number of stable bump positions is about one thousand times smaller than the number of lattice nodes. Furthermore, the few stable positions of the retrieval bump are different for different patterns (data not shown). This makes the representation of position dependent on object identity and thus hard to decode. We ask, therefore, whether it is indeed possible to stabilise bumps at any arbitrary position. This is discussed in the following sections.
In this section we show that the bump of activity can be stabilised on an arbitrary position on the network if neurons around that position have a slightly higher linear gain than the rest of the neurons. This increase in the linear gain applies to all neurons in that area in the same manner, whether they are selective for the cued pattern or not; that is, it is not pattern selective and solely reflects object position. This local gain modulation can be triggered by an attentional mechanism that modulates the responsiveness of neurons in the part of the network which corresponds to the position of the object. It could also be produced by the pattern itself: when the cue to initialise retrieval is given to the network, the mean activity of the part of the network that receives the cue would be higher than the rest of network. This could trigger changes in the gain of the neurons that may last for several seconds [62]–[64]. We leave discussing the sources of the gain modulation to the Discussion section and first answer the following questions. Can such localised gain modulation stabilise the bump at any desired position and, if so, how strong should it be? How does localised gain modulation affect pattern retrieval?
Suppose that a non-pattern-selective signal changes the gain of those neurons which correspond to the position of the object in the visual scene. The effect of such gain modulation is shown in Fig. 5.
In the simulations of Fig. 5, the activity of ≈4.6% of the neurons, randomly distributed across the network, are initially set to their activity in the first pattern, while the rest are silent (note that the quality of the cue is then the same as what we used in the simulations of Fig. 3, but now the cue is not localised). The localised gain modulation is incorporated into the simulations by first choosing, at each trial, a square box at a different position over the network. The linear gain of neurons inside the square is then increased by a factor of β relative to that of the other neurons in the network. The position of the centre of the high gain square box is in fact chosen in exactly the same way as we chose the centre of the cue in Fig. 3, i.e. at the nodes of a regular lattice, shown as green squares in Fig. 5A and Fig. 5C.
The result of such change in the spatial distribution of the gain is evident for β = 1.5 (Fig. 5A, 5B) and even more for β = 3 (Fig. 5C, 5D). Even though the pattern-selective cue does not contain spatial information, a spatially selective increase in the linear gain of the neurons in a restricted region of the network helps localising the bump in that region. Notably, as shown in Fig. 5D, the distance that the peak in the local overlap drifts from the initial focus is minimal, particularly for successful trials (red circles) (d), whereas averaging across unsuccessful runs (black circles) (d*) substantially increases the drift, as if jumping to the wrong basin of attraction in the space of patterns facilitates similar jumps in physical space.
It should be noted that while in Fig. 3 the localised cue had been removed after initialising the activity, in the results shown in Fig. 5 the change in gain is maintained throughout the simulation. It is true that keeping the localised cue would have helped localising the bump at the right position, without gain modulation, but the essential difference between the two mechanisms should be appreciated: the change in gain is independent of the memory pattern to be retrieved and could thus be produced by a mere spatial signal, with little specific information content besides spatial position itself. The pattern-selective cue, instead, can be thought to commit the informational resources (e.g., the channel capacity [65]) of the ventral visual form processing stream, and it makes sense to hypothesise that it should be removed as soon as possible, to make room for the analysis of other objects by the same pathway.
Even though increasing the gain in a spatially restricted part of the network stabilised the final bump there, there is a disadvantage with this strategy: by using such non-uniform gain, the number of successful runs decreases. Remember that the quality of the cue used in Fig. 5 is the same as the one in Fig. 3, however, there were no unsuccessful runs in Fig. 3 and Fig. 5A, whereas there are 12 unsuccessful runs in Fig. 5C (shown by black circles): better preservation of spatial information (higher gain modulation) is accompanied by, in this example, a higher number of unsuccessful runs. This suggests that preservation of spatial information through gain modulation affects the retrieval of the pattern. In Fig. 5 the effect is negative, an interference, but as we show below it can also be a positive effect. In the following sections, we quantify this interaction using information theory and demonstrate efficient ways to minimise the negative interaction.
In order to quantify the interaction between what and where information, we use Shannon information theory. We estimate the amount of information that the activity of the network, after retrieval, represents about what and where. We do this for different degrees of gain modulation, levels of the average gain, number of stored patterns and also different ways of presenting the cue. This provides us with a quantitative picture of the relation between what and where information.
We denote by Iwhat and Iwhere, the amount of information about what and where, respectively. To compute Iwhat, we look at network activity after 200 times steps and compute its overlap with all stored pattens (Eq. (9)). The pattern with the highest overlap is considered as retrieved and Iwhat measures how much information knowing this retrieved pattern gives us about which pattern was presented. Iwhere, on the other hand, is the information between the position of the bump of activity after retrieval and the centre of the gain modulated area (we put Iwhere = 0 when there is no gain modulation; see section “Continuous attractors are fragmented by superimposed memories”). For exact definitions and details about how we compute Iwhat and Iwhere from the simulations see section “Mutual information measures” in the Materials and Methods.
To start with, we consider a network (with the architecture used before) that has stored p patterns and assume that in the beginning of the simulations a cue similar to one of the patterns is presented (the exact cue presentation is described in the three Conditions below). All neurons have a background gain of g. During recall, either the gain of all neurons is kept equal to g, which is the case of uniform gain, or the localised gain modulation mechanism is turned on. In the latter case the gain of the neurons inside a 15×15 square whose centre is on one of 49 preassigned positions on the network is boosted to βg. Different values of β are chosen in different simulations. In each run, one of the patterns is chosen as a cue and one of the 49 positions is chosen as the centre of the high gain region. As in the previous sections, the centre of the squares surrounding the high gain region is chosen from one of the 49 nodes of a 7×7 regular lattice covering the entire 2D network. Each pattern and each of the 49 positions for the high gain region is used exactly once.
We first calculate Iwhat and Iwhere for a network with the global gain chosen to be g = 0.5. We do this for the case of uniform gain (all neurons have the same gain, thus equal to the background gain g ), three degrees of gain modulation, with β = 1.5,2 and 3 , and three values of p = 5,10 and 15. We consider three alternative ways in which the cue can be presented to the network. These cueing conditions and the resulting Iwhat−Iwhere relation are described below.
In the previous section, the background gain was g = 0.5. Without gain modulation, the network could reach high Iwhat values, sometimes retrieving all stored patterns, even from a very small initial cue. With gain modulation, Iwhere increased but Iwhat decreased. Here, we show that when the background gain is low, the interaction can be reversed, that is, gain modulation can actually increase both Iwhat and Iwhere. We set the background gain to g = 0.25. As shown in Fig. 7, for the case of complete cue (as in Condition 1 above) even without gain modulation Iwhat is very small. When incorporating a gain modulation mechanism, however, the amount of what information maintained by the network increases, together with the amount of where information. In section “Low gain regime versus high gain regime” (see Materials and Methods), we discuss why the relation between Iwhat and Iwhere is different in the low gain and high gain regimes. Intuitively, the reason is as follows. Successful retrieval occurs only when the gain of the neurons that support the retrieved pattern is between a minimum gmin and a maximum gmax. In the low gain regime, the level of background gain is below gmin and the network cannot retrieve the patterns. When the gain is increased in part of the network, then it may enter the range [gmin, gmax], allowing for retrieval to occur. At the same time, since that region has a higher gain, the retrieval bump does not drift away. When the background gain is high, instead, gain modulation stabilises the bump in the gain modulated area. This is accompanied, however, by a decrease in the size of the bump. The reason is that the higher neuronal gain increases the firing rate of neurons inside the bump (the peak of the bump is higher) and, to comply with the constant mean activity condition (Eq. (3)), this increase in the peak activity is accompanied by a decrease in the spatial extent of the bump. Therefore, fewer connections are involved in retrieving the pattern and Iwhat decreases. As expected from this argument, increasing β too much even in the low gain regime should decrease Iwhat. This can be seen in Fig. 7 for β = 5 and p = 10.
When a retrieval bump is localised on a particular position, one can in principle use the rest of the network to retrieve other patterns, in the form of additional bumps of activity. If they can coexist with the first bump, the network would then be able to represent the position and identity of multiple objects simultaneously, without encountering the problem of binding. In random networks with no metric connectivity nor localised retrieval, retrieving multiple patterns is indeed possible, at very low storage loads [1],[45],[66]; in these networks, however, it is not possible to represent the position of the objects, which has to be represented elsewhere. If the what and where of multiple objects are represented in different networks, a binding problem arises. The localised retrieval process described here does not suffer from this problem. It is then important to assess the conditions which make it possible to stabilise (at least) two retrieval bumps simultaneously.
Assume that a pattern is retrieved and, using localised gain modulation, the bump of activity is stabilised on a desired position. A second cue may then be presented to the network at another position. Even though most of the connections to each neuron in the network come from nearby neurons, the second pattern would still affect the first retrieval bump, because of the global inhibition in the simplest version of our model, as inhibition is taken to regulate a common threshold, such that the mean activity of the network is constant (Eq. (3)). This introduces interactions between distal neurons, which are not directly connected by excitatory synapses, and such interactions are generally disruptive. A simple way to reduce such interaction is to assume that when the local mean activity in part of the network exceeds some limit value, the threshold is raised but only locally, regardless of the activity of neurons outside that region. The local threshold may also be regulated downward, to facilitate the emergence of a retrieved pattern in a region which would otherwise be kept at too low a mean activity level. With such additional provisions, multiple bumps can be formed and stabilised, as shown in the example in Fig. 8.
Behaviour requires processing and integrating different types of information, from various sources and modalities, into a coherent picture of the world. Within the visual domain, a specific question is how the brain can analyse the identity of objects, which has to be extracted from raw visual input, while maintaining information about their position, directly present in the input.
Previous theoretical work on the representation of objects in neural circuits has been mainly focused on two issues [2], [5], [67]–[69]: how the hierarchy of visual cortical areas builds representations that are invariant with respect to changes in position, view, etc. of objects, and how this may be accomplished while still preserving information about the relative position of features within objects, to enable object recognition. Among these studies, Olshausen et al [2] and Parga and Rolls [3] also considered how attractor dynamics can be used to retrieve what information from stored invariant representations. However, this body of work did not address how an activity pattern that carries information about both what and where can be produced when what information is retrieved from memory. They also did not consider how this combined what and where representation can be maintained in working memory, after the visual stimulation has subsided. Retrieving information about object identity from memory, as well as maintaining this information in working memory, has been associated to attractor dynamics in local cortical networks. The most straightforward extension of the attractor idea, to store attractors associated to what-where pairs is, however, infeasible due to the extremely large storage capacity that it would require (see the following section “Comparision with storing attractors associated to object-position pairs”).
Our model sheds light onto this issue of how to combine the representation of what and where, by showing that a recurrent network can retrieve stored memories about objects from incomplete transient cues, while maintaining information about their positions. It can account for the what-where delay activity observed in monkey electrophysiology [13],[14] and it can combine what and where information in a flexible manner as has been reported in experimental studies [13],[17],[28]. In our model this flexibility is expressed in the fact that by changing the level of background gain and localised gain modulation, one can control the levels of what and where information that the network retains. When the network is operating in the low gain regime, turning on the localised gain modulation increases both what and where information, whereas in the high gain regime what information decreases and where information increases. Behavioural experiments show a pattern of interaction between what and where information similar to this latter case [70],[71]. It is interesting to note that it has been recently reported [72] that single IT neurons, when they show high selectivity (i.e. they respond vigorously to only a few images in a large sample) also tend to show less position tolerance, suggestive of their ability to convey more where information. This could be interpreted either as the more selective neurons contributing less what information to the population response, or even as implying a different trade-off at the single neuron level from the one we propose to prevail at the population level.
The localised retrieval process described here also offers the possibility of retrieving multiple objects while maintaining their position, without facing a binding problem [73]. The metric excitatory connectivity avoids interference effects mediated by excitatory connections, while inhibitory mechanisms should be such that two activity patterns retrieved at different locations do not destroy each other once they are formed. One such mechanism was briefly described in section “Multiple bumps”. The crucial questions about the coexistence of multiple bumps are of course still open: how does it depend on the parameters of the model, and in particular on its detailed dynamics? how many bumps can simultaneously coexist in a network of a given size? how does the ability to support multiple bumps changes the storage capacity? These questions require further investigations. In the context of networks with spatially dependent weights without stored memory, it has been shown that oscillatory weights can support multiple bumps of activity [74],[75]. The possibility of supporting multiple retrieval bumps using more complicated connectivity schemes remains open in our model.
In our model, independent attractors are set up in a local cortical network only for object identity, as position invariant representations; but they can be accessed in a spatially focused mode, leading to position dependent activity. Associating a single representation to an object, which is then modulated by position, is a particular case of what in cognitive neuroscience parlance is sometimes referred to as type (e.g. table) and token (particular instance of a type: e.g. a table in a particular position) [76],[77]. In the language of our model, the type is the original pattern of activity associated to an object and the token is the bumpy pattern that is localised in a particular position. An alternative mechanism is to store attractors associated to object-position pairs, that is storing a neural activity pattern for each token [78]. In this way, when a particular object is presented in a particular position, the attractor corresponding to the object-position pair would be activated, and could remain activated even after the object has been removed from the scene. The problem is that models which hypothesise distinct, discrete attractors for each possible object-position combination would certainly violate any conceivable storage capacity limit, because of the infinitely large number of possible positions of an object [79]. Furthermore, there is a major difference between the nature of what and where information, which makes attractors associated to object-position pairs unlikely: as opposed to what information, to which the brain can contribute from the information that it has previously stored, the brain does not usually retrieve positional information from memory, but rather has to maintain it as well as it can. Thus, it would seem rather implausible that the brain uses its storage capacity, arguably its most precious resource [29], to store something that it does not have to retrieve.
The difference between these two mechanisms is directly reflected in the storage capacity required for object-position attractors, in order to represent the same amount of information as the mechanisms studied here does through spatial modulation. Representing 6 bits of Iwhere and 2 bits of Iwhat (corresponding to the black diamonds in Fig. 7) would require the storage of 2(6+2) = 256 distinct object-position attractors. This is ca. 2.5 times beyond the number of attractors that a randomly connected network, with the same number of connections per neuron and the same mean activity level as what we used, could store [32]. This exorbitant requirement is due to effectively committing storage space separately to each pair, instead of using the physical arrangement of neurons in the tissue to represent Iwhere. Analytical results valid in the limit of large networks and optimal storage further support this conclusion, as we show in section “Comparision with other models” in the Materials and Methods. There, we also show that the difference in the efficiency of the two models will be even more pronounced for larger networks. There is, of course, a price to pay: the addition of a gain modulation mechanism to stabilise the position of the bump. In what follows, we discuss the possible physiological substrates of this gain modulation.
In our model, localised gain modulation is crucial for maintaining where information as what information is being retrieved, and for maintaining both what and where information after the retrieval process is completed. When an object is presented as a stimulus, a signal should trigger an increase in the gain of neurons in an appropriate part of the network. Such higher gain should then be maintained by the same or a distinct mechanism during retrieval and thereafter, when the object is not present anymore but information about it has to be used (e.g. during the delay period of a delay-match-to-sample task).
What mechanisms can trigger the neuronal gain? In vivo studies show that increasing the activity of a local cortical network increases the gain of its neurons [80]–[82]. Therefore, any mechanism that increases the mean activity of a part of the network could be used for triggering the gain modulation. One such source of increase in the activity is the cue itself. This requires that the pattern selective cue retains some spatial information; a scenario which we have shown to be particularly effective in minimising the trade-off between what and where information (see Fig. 6C). Although this mechanism would be effective in this sense, it is doubtful whether it could be the only source of gain modulation in high level visual cortices. This is because experimental studies show that the position of the peak of the activity in visual cortical areas during visual stimulation is strongly correlated with the categorical properties of the stimulus and exhibits a weaker level of retinotopy [83]–[85] (see also the following section “Storing patterns with spatial prefrence”). The situation may be different in more advanced cortical areas, such as PFC, in which such categorical maps have not been reported.
Another possible source for increasing the gain is attentional signals. In this case the increase in the activity level required for gain modulation is induced by the attentional signal and the position of the bump corresponds to the position of the attentional spotlight. There are several reasons that make attention a likely source of activity localisation through gain modulation. fMRI studies on human subjects show that the retinotopic representation of the position of an attended object in visual cortices show increased activity [86]–[94]. Evidence from monkey neurophysiology also supports the idea that the attentional spotlight increases the gain of neurons inside the spotlight [95]–[100]. Furthermore, many studies in cognitive neuropsychology suggest that spatial, focal attention is critical to allow the binding of what and where information [77],[101], referred to as type and token information respectively [76]. Finally, a recent neuroimaging study shows that attention strongly enhances retinotopic representation in object selective visual areas, thus supporting the idea that attentional gain modulation is important for combined representation of what and where [102]. Although, these studies point to attentional signals as a strong candidate for initiating the gain modulation, a contribution may still be given by a weakly retinotopic initial cue. Further experimental work is required to disentangle the relative effect of the initial cue and attention on triggering the gain modulation.
Once the increase in the gain of neurons in the right part of the network is triggered, it should be maintained during retrieval. Although the same mechanisms that initiated gain modulation can keep the gain high during retrieval, a promising mechanism for maintaining high level of gain, particularly after the stimulus is removed, is single neuron memory. Several studies show that the recent history of spiking increases the responsiveness of neurons, and that this increase can last for several seconds, thus exhibiting a form of single neuron short-term memory [62]–[64]. Assuming that such single neuron short-term memory mechanisms are responsible for the higher gain of neurons inside the bump, global signals that turn them on or off can strongly affect the level of what and where information that the network represents in its activity.
As discussed above, the attentional signal may trigger the increase in neuronal gain and maintain it elevated for some time. After the attentional signal is removed, the increase in neuronal gain can be maintained by single neuron short term memory mechanisms. Attention can then be directed to another object, while what and where information about the first object is still decodable from neuronal activity. How long this information survives depends on how long the short term increase of the gain can be maintained by single neuron mechanisms. Understanding such mechanisms and comparing their time scale with behavioural times for maintaining combined what and where information, as well as pharmacologically interfering with them, one can test whether our model is relevant to real visual perception.
One of the roles of attention is to bias the competition for limited processing resources in favour of the object that it is acting on [103],[104]. Therefore, if the localised gain modulation that is needed in our model for combining what and where is induced by attention, it should be able to do the same. This is verified by computer simulations as shown in Fig. 9. Two localised partial cues, corresponding to two different objects, are simultaneously given to a network. When the neuronal gain is uniform, the object with the larger cue will be retrieved, while the other one will be suppressed. However, if the neuronal gain in the area that receives the smaller cue is sufficiently large, the competition will be biased in favour of it. Interestingly, the level of gain modulation that is required to bias the competition towards the object with the small cue depends on the width of the connectivity, σ. Increasing the width of the neuronal connectivity increases the minimum level of gain modulation that is required for biasing the competition. This emphasises the role of local connectivity.
In the model presented here, the units are taken to be arranged on a retinotopic patch of cortex, corresponding to at least a portion of visual field, but we assumed patterns of activity to be generated from a spatially uniform distribution (see Eq. (4)). A more realistic model, however, should allow for the storage of spatially organised patterns [105]. This is important since, in the case of high level visual cortical areas, the overall position of intense neural activity during visual stimulation is strongly correlated with object identity or category. Regions in the visual cortex have been located that are preferentially selective for faces [106]–[109], pictures of scenes [110],[111] and buildings [112], and complex object features [113]. This strong categorical map may coexist with a retinotopic map. The details of this combined organisation are far from clear, however, particularly insofar as it is expressed in the putative attractor states, after the stimulus is removed (e.g. during delay periods), which is the situation relevant to our study. During visual stimulation, and when attention is not a main factor, some studies suggest that there is a weak retinotopy, with only a peripheral versus central bias and no angular representation [83]–[85]. Others, on the other hand, report the existence of multiple precise retinotopic maps in the same regions [114]–[116], although still much weaker than the level of retinotopy in primary visual areas [117]. As mentioned in the previous section, such retinotopic maps could be enhanced by attention [102].
To include the coexistence of categorical and retinotopic maps in the model presented here, one might consider two limit cases, which roughly correspond to these two views. In the first case, category specificity and weak retinotopy coexist at the same spatial scale; one should then assume, in a refined model, that patterns are generated from multiple distributions, each of them corresponding to one category of objects, and patterns drawn from each have higher activity at a preferred position on the network. In this case, when there is no gain modulation the peaks of the retrieved patterns cluster depending on which distribution they came from. The peaks will also be more weakly correlated with the position of the cue compared to the case of spatially uniform patterns that we have discussed. With attentional gain modulation, one expects to see a clearer retinotopic map. This is in fact consistent with the abovementioned finding that attentional gain modulation enhances the retinotopic representation in advanced visual areas [102]. In the second limit case, retinotopy is expressed in object selective visual areas at a finer scale than category specificity, in which case one should allow for the present model to be simply multiplexed, to include one array on a distinct cortical patch for each object category.
Further work is required, especially in view of many intermediate possibilities, to assess, for example, how much more gain modulation would be needed in order to stabilise a bump of activity away from its preferred position, and how this would affect retrieval.
The ability to represent what and where information in the same network has also been proposed to be crucial to understand the functional significance of the differentiation among cortical layers [118]. Whereas most network models used to study attractor dynamics in associative memory do not consider cortical lamination, the core hypothesis of the proposal is that layer IV units, by virtue of their distinct connectivity, may privilege the representation of position information. Furthermore, through less adaptive spiking activity they may influence the dynamics of pyramidal units in the superficial layers only after these have engaged the attractor basin that leads to retrieve object identity. The differentiation was shown to be advantageous, in the model, through computer simulations, conducted with external inputs maintained active. In this regime no assessment was possible of whether genuine dynamical attractors had indeed been formed during memory storage, that will drive network dynamics in the absence of the cue. While the present work clarifies the conditions allowing a single layer network to represent what and where information, how they could be realized in a network with differentiated cortical layers remains to be explored.
In discussing what and where information, we have made explicit reference, here, to object identity and position in the visual field. Where information could however be any feature that is mapped in the gross topography of the cortical sheet, such as frequency in the auditory system [119], and in relation to which there is no meaning to using attractor dynamics in order to refine the afferent signal with what is stored in memory. In fact, this mapping need not even be topographically organised: the crucial factor is the existence of a map (topographic or not) [120], that is produced as a result of the dependence of in Eq. (5) on i and j, and that is independent of the stored patterns. Where information would ideally be expressed by a continuous attractor and thus maintained e.g. as delay activity, except that continuity at a fine scale is disrupted by the storage of what memories. What information could instead be any feature that could benefit from attractor dynamics, because of its uneven statistical distribution, which makes some interpretation of the afferent signal more likely than others.
If synaptic weights are produced by Eq. (5), the weights of the connections that originate from a given neuron can be both negative and positive. This is against Dale's law and against our assertion that all neurons in the model network are excitatory. In this section, we show how the model described in sections “Firing rate description of the network” and “Stored memory patterns and synaptic weights” (see Model) can be conceptually derived from a more realistic formulation, in which all synaptic weights are positive. Let us first consider a network in which the firing rate of neuron i at time t+1 is determined by(14)in which Thi is the threshold of neuron i, Ii is its inhibitory input, and(15)
The synaptic weights, Wij, in this network take the following form(16)where Jback is the background weight, ϖij = 1 if there is a connection from neuron j to neuron i and ϖij = 0 otherwise, and C is the average number of connections per neurons. For sufficiently large Jback, the resulting synaptic weights in Eq. (16) will be all positive. We can now show that a network with uniform threshold, as assumed in Eq. (2), and synaptic weights of the form Eq. (5), has equivalent dynamics as described by Eqs. (14) and (16), when an additional condition is satisfied.
Combining Eq. (14) with Eqs. (15) and (16), the firing rate of neuron i can be written in terms of the firing rate of the other neurons as(17)in which Jij is the weight of the connection from neuron j to neuron i according to the prescription Eq. (5). The assumption we now make is that the inhibitory feedback reacts in such a way that for each neuron, the last three terms in the parenthesis in Eq. (17) together become equal to a uniform effective threshold, Th. This effective threshold is simply chosen such that Eq. (3) holds. In this way, Eq. (17) reduces to(18)which is the same as Eq. (2).
In this section we briefly describe how the self-consistent equation for the local overlap with the retrieved pattern (Eq. (10)) can be derived. We refer the reader to [32],[33] for more details.
To start with, we assume, without loss of generality, that the first pattern (μ = 1) is retrieved and therefore for . Using Eqs. (1), (5) and (8), we then write the input to neuron i as(19)Denfining zi as(20)and combining Eq. (19) and Eq. (2), the activity of neuron i can be written as(21)
Inserting vi from Eq. (21) into Eq. (8) we arrive at the following self-consistent equation for (22)
Averaging the right hand side of Eq. (22) over the distribution of zj, η, and the connectivity pattern, yields the following equation (which is the same as Eq. (10))(23)where 〈〉η , stands for averaging over the distribution of η, is the probability of connection (Eq. (7)), and F̅j is the gain function, F, averaged over the distribution of zj(24)
We now find the distribution of zj, which we denote by zj Pri(zi). To do this we note that if the first pattern is retrieved, vjs, on the right hand side of Eq. (20) will be independent from each other and from ημ for μ≠1. The assumption of independence is strictly correct when the network is highly diluted, that is when the number of presynaptic neurons shared by any two postsynaptic neurons is small [121],[122]. When the network is not highly diluted, the calculation will be more involved, but yields qualitatively the same results [32],[123]. Thus, for the sake of simplicity, we assume that the assumption of independence holds; for a complete derivation we refer the reader to the aforementioned references. With this independence assumption, the right hand side of Eq. (20) will be a sum of independent random variables, and therefore, Pri(zi) will be a Gaussian distribution. In the following we show that the mean of this Gaussian distribution is zero and also find a self-consistent equation for its variance.
Noting that(25)(26)and using Eq. (21), we get the following equations(27)(28)where indicates averaging over the distribution of η and ϖij, and 〈〉η indicates averaging over η. From Eq. (27), we see that the mean of Pri(zi) is zero. In order to find the variance of Pri(zi), we should average both sides of Eq. (28) over the distribution of zj. This is because, in the limit of large N and large C, this variance is expected not to depend on the exact realisation of any zj in the right hand side of Eq. (28), but only on its statistical distribution. Performing this average yields the following equation for the variance that we denote by ρi2(29)
Equations (23) and (29) form a closed set of equations whose solutions determine the steady states of the system. Finding mi and ρi that satisfy these equations, we can find the activity of neurons in the steady states by plugging them in Eq. (21). In the case of a randomly connected network, that is when is independent of i and j, and gi are also the same for all neurons, the solution of Eqs. (23) and (29) will be of the form mi = m and ρi = ρ. In this case the only spatial dependence of the steady state activities, Eq. (21), will come from the dependence of ηi1 on i and since they are generated identically for each i, the probability that a neuron is active in the steady state will be uniform over the network. Spatially localised retrieval can be observed when depends on the distance between i and j.
In this section we show how we compute what and where information, Iwhat and Iwhere, from simulations. We estimate the amount of what information, Iwhat, from the frequency of successful retrieval runs. To see how, let us assume that we cue pattern μc. Then after some time we look at the pattern of activity of the network, compute its dot product overlap with all stored patterns (Eq. (9)) and find that pattern μr, say, has been “retrieved” in that particular run, i.e., it has the highest overlap with the activity of the network. We denote the probability of retrieving pattern μr given that we have cued pattern μc by Pr(μr|μc). Estimating this probability from the simulations, we can compute the information that the pattern of activity gives us about which pattern was presented as(30)where Pr(μc) is the probability of cueing pattern μc and(31)
In the simulations all patterns are presented an equal number of times, therefore,(32)
We denote the fraction of successful runs (when μc = μr) that we measure from the simulations by f, that is(33)
Since in unsuccessful runs (when μc≠μr), all patterns, except for μc are a priori equally likely to be retrieved, we have(34)
Using Eqs. (32)–(34) in Eq. (30), we can thus write for fixed degree of gain modulation, fixed background gain, and fixed number of patterns,(35)
Note that the above is, strictly speaking, only a measure of the information implicit in the selection among the p patterns operated by attractor dynamics; under certain conditions, however, it can also serve as an indicator of the total information available in the firing pattern itself [124].
Iwhere is the mutual information between the peak of the local overlap after 200 time steps and the centre of the gain modulated area (or the centre of the cue when there is no gain modulation). To estimate where information, Iwhere, we first measure the distance between the peak of the final overlap of the successful runs and the centre of the gain modulation, for each cued pattern. Then we make a histogram of these distances and calculate the fraction of runs which fall in any of the 10 distance bins chosen to be b1 = [0,5],b2 = [5],[10],…,b10 = [45],[50]. In this way we have the conditional probability, Pr(k | x), of having the peak of the activity in the kth distance bin, given that the peak was initially at position x on the lattice. With N neurons on each side of the lattice, we have Pr(x) = 1/N2, and we can write Iwhere as(36)in which we have used the fact that Pr(k | x) does not explicitely depend on x and we can simply denote it by Prk. Similarly to what we do for Iwhat, we have also assumed that for any such ring between the circles of radius 5k and 5(k−1), centred on the gain modulation square, the final bump can be anywhere, with equal probability, on the ring. In this expression the factor 2k−1 accounts for the fact that the area covered by the kth bin is 2k−1 times the area of the first bin, and hence its a priori probability is 2k−1 times higher. The first term in Eq. (36), is the maximum information value, Iwhere≃6 bits, in this approximation, i.e., the logarithm (in base 2) of the ratio between the “area” of the network (4900) and that of the smallest bin (5×5×π), and is achieved when all successful runs end up with a bump at d≤5 from its intended position.
In this section we discuss why in the low gain regime, gain modulation aids retrieval of the patterns whereas in the high gain regime it has a negative effect. We start from the self-consistent equations, Eqs. (23) and (29).
Assume that the steady state of the network is a bump of activity over a part of the network with single neuron gain βg, whereas the rest of the network is silent with gain g. Furthermore, assume that mi and ρi that satisfy Eqs. (23) and (29) are nonzero inside the bump and zero elsewhere. Consider that inside the bump mi = m and ρi = ρ, where m and ρ can be regarded, just for simplicity, to be roughly constant. Then from Eqs. (23) and (29) we have:(37a)(37b)where α = p/C is the storage load and(38)
Eqs. (37) are of the form of mean-field equations of a recurrent network with non-metric connections [79],[125] (assuming uniform values for mi and ρi inside the bump and zero outside is equivalent to assuming that the part of the network, over which the bump is formed, is behaving as an independent network). For each value of α, Eqs. (37) have non-zero solution for m, and thus the network can retrieve the stored patterns, if and only if gmin(α)<βg<gmax(α), where gmax(α) and gmin(α) are functions of α. The effect of background gain g can now be readily seen. When g<gmin(α) retrieval does not happen without gain modulation. With gain modulation, however, the neuronal gain of the part of the network that is gain modulated will be boosted by a factor of β and for large enough β, the neuronal gain will be in the regime that supports retrieval i.e. gmin(α)<βg<gmax(α). When the background gain g is high, βg can exceed gmax(α) , thus retrieval will not be successful.
In this section, we discuss why it is more efficient to spatially modulate attractor states associated to objects, than to store distinct attractors for different positions of each object.
Under optimal conditions, the number of attractors that an associative memory with C connections per neuron, but without metric connectivity, can retrieve is(39)where k is a constant that is primarily determined by the sparsity of the stored patterns [79]. Metric connectivity, which enables localised retrieval, decreases k by a moderate factor γ1≃3−4 [32]. Localised gain modulation, that stabilises the bump at an arbitrary position, decreases k again by another factor, γ2, that for the parameters and network size we used turns out to be γ2≃4. This is actually an overestimation of the decrease in storage capacity due to localised gain modulation, for realistic size networks. This is because when we calculate the mean of the right hand side of Eq. (22) over the distribution of connectivity patterns and η to get Eq. (23), we ignore the fluctuations around this mean, that behave as . These fluctuations are what break the translational symmetry of the self-consistent equation, Eq. (23), and make the bump favour a few positions over the others, and are compensated for by the localised gain modulation. As a result, less gain modulation is required for stabilising the bump when there are more connections per neuron. However, even with this estimate for γ2, the process described here results in a moderate reduction in storage capacity(40)
The spatial modulation described here can represent positional information with a resolution , where l is the lattice spacing and Np is the number of distinct position that can be resolved-in a large network, Np∼O(N) (see Fig. 5).
On the other hand, the naive storage of distinct, unrelated attractors for each object position pair decreases the number of objects, whose identity could be retrieved, to(41)illustrating the wasteful use of memory resources for positional information, which in itself requires no memory.
An alternative arrangement might be to associate attractors to objects, but allow each attractor to be a continuous 2D manifold, different for each object, so that position can be represented by the position of a bump of activity on such attractor manifold, unrelated to the position of the active neurons in the tissue. This arrangement corresponds to the multiple spatial charts model of Samsonovich and McNaughton [58], introduced to account for the ability of rodents to track their own position in multiple spatial environments, by coding it as a group of coactive hippocampal place cells, which comprise a bump on a chart corresponding to each environment. Instead of assigning distinct charts to distinct spatial contexts, such as a square recording box rather than a circular one, one could well assign distinct charts to distinct objects, each of which would then have its own “private” continuous or quasi-continuous attractor, unrelated to the 2D arrangement of neurons in the tissue. The mathematical analysis of the multiple charts model [126] reveals that a network can store a number of charts equivalent to the number of attractors in a standard associative network of the same connectivity, reduced by a factor Nb, which is the number of place cell ensembles, uncorrelated with each other, required to “tile” a chart. In the simplest version of the model, each neuron shows a single place field in each environment (at a different spatial position in each chart) covering a fraction a of the total area of the environment. Then Nb≈(1/a) and, although the number of positions that can be represented accurately can be larger than Nb, still a≪1 for the network to be able to resolve position in space. Therefore, adapting the hippocampal model would also yield a lower capacity(42)because of the cost of creating a separate “virtual” space for each object. Simply utilising the position of neurons in the tissue to represent physical position for all objects, and reserving memory resources for object identity, provides the most efficient solution to combine what and where information. Note instead that in the hippocampus, to the extent that it utilises coactivity patterns to discriminate between different spatial contexts [127], the position of neurons in the tissue cannot be used to code for position in real space, and in fact place field position in the chart is found to be unrelated to cell position in the tissue [128]. It is also worth mentioning that the same problem that we encountered for stabilising the bump at an arbitrary position will also appear in models that associate a distinct chart to each object [57]. Therefore, an extra mechanism will be required in this case, too, and the real pmax will be smaller than pmax in Eq. (42) by a factor similar to γ2 in our model.
|
10.1371/journal.pbio.1001239 | Hedgehog-Regulated Ubiquitination Controls Smoothened Trafficking and Cell Surface Expression in Drosophila | Hedgehog transduces signal by promoting cell surface expression of the seven-transmembrane protein Smoothened (Smo) in Drosophila, but the underlying mechanism remains unknown. Here we demonstrate that Smo is downregulated by ubiquitin-mediated endocytosis and degradation, and that Hh increases Smo cell surface expression by inhibiting its ubiquitination. We find that Smo is ubiquitinated at multiple Lysine residues including those in its autoinhibitory domain (SAID), leading to endocytosis and degradation of Smo by both lysosome- and proteasome-dependent mechanisms. Hh inhibits Smo ubiquitination via PKA/CK1-mediated phosphorylation of SAID, leading to Smo cell surface accumulation. Inactivation of the ubiquitin activating enzyme Uba1 or perturbation of multiple components of the endocytic machinery leads to Smo accumulation and Hh pathway activation. In addition, we find that the non-visual β-arrestin Kurtz (Krz) interacts with Smo and acts in parallel with ubiquitination to downregulate Smo. Finally, we show that Smo ubiquitination is counteracted by the deubiquitinating enzyme UBPY/USP8. Gain and loss of UBPY lead to reciprocal changes in Smo cell surface expression. Taken together, our results suggest that ubiquitination plays a key role in the downregulation of Smo to keep Hh pathway activity off in the absence of the ligand, and that Hh-induced phosphorylation promotes Smo cell surface accumulation by inhibiting its ubiquitination, which contributes to Hh pathway activation.
| The Hedgehog (Hh) family of secreted proteins governs cell growth and patterning in diverse species ranging from Drosophila to human. Hh signals across the cell surface membrane by regulating the subcellular location and conformation of a membrane protein called Smoothened (Smo). In Drosophila, Smo accumulates on the cell surface in response to Hh, whereas in the absence of Hh it is internalized and degraded. The molecular mechanisms that control this intracellular trafficking and degradation of Smo were unknown, but here we show that Smo is modified by attachment of several molecules of a small protein called ubiquitin, which tags it for internalization and degradation within the cell. Hh inhibits this ubiquitination of Smo by inducing another modification, phosphorylation, of its intracellular tail by two types of protein kinase enzymes. This loss of ubiquitination and gain of phosphorylation causes the accumulation of Smo at the cell surface. What's more, we find that another protein called Kurtz interacts with Smo and acts in parallel with the ubiquitination process to promote internalization of Smo, and that the deubiquitinating enzyme UBPY/USP8 counteracts ubiquitination of Smo to promote its cell surface accumulation. Our study demonstrates that reversible ubiquitination plays a key role in regulating Smo trafficking to and from the cell surface and thus it provides novel insights into the mechanism of Hh signaling from the outside to the inside of the cell.
| Hedgehog (Hh) signaling governs cell growth and patterning in species ranging from insects to human [1],[2]. Because of its pivotal role in embryonic development and adult tissue homeostasis, misregulation of Hh signaling activity has been linked to many human disorders including birth defects and cancers [1],[3],[4]. Hh exerts its biological influence through a largely conserved signaling cascade that culminates at the activation of latent transcription factors Cubitus interruptus (Ci)/Gli [1].
The core Hh reception system consists of a 12-transmembrane protein Patched (Ptc) that acts as the Hh receptor and a seven-transmembrane protein Smo that acts as the Hh signal transducer [5],[6]. Hh and Ptc reciprocally regulate the subcellular localization and active state of Smo [7]–[10]. In Drosophila, Hh stimulation or loss of Ptc leads to cell surface accumulation of Smo [7],[11]. Increased cell surface expression and activation of Smo are regulated by Hh-induced and PKA/CK1-mediated phosphorylation of Smo carboxyl intracellular tail (C-tail) [12]–[14].
Several observations suggest that Smo cell surface expression is controlled by endocytic trafficking. A transmission electron microscopic study of Drosophila imaginal discs indicated that Smo is localized primarily in the lysosome of anterior compartment cells but is enriched on the plasma membrane of posterior compartment cells [15]. In Drosophila salivary gland cells, blocking endocytosis promotes Smo cell surface accumulation [11]. Using antibody uptake assay in S2 cells, we have shown that Smo reaches the cell surface but quickly internalizes in the absence of Hh and that Hh stimulation diminishes internalized Smo with a concomitant increase in cell surface Smo [13]. Taken together, these observations suggest that Hh signaling may regulate Smo cell surface expression by blocking its endocytosis and/or promoting its recycling back to the cell surface after internalization.
The mechanisms by which Smo endocytic trafficking and cell surface expression are regulated have remained unknown. Smo intracellular regions lack recognizable endosomal-lysosomal sorting signals such as the NPXY and dileucine-based motifs [16]. However, many membrane receptors are internalized after covalently modified by ubiquitination, as has been demonstrated for receptor tyrosine kinases (RTKs) and G protein coupled receptors (GPCRs) [17],[18]. The close relationship between Smo and GPCRs prompted us to investigate whether Smo cell surface expression is regulated by the ubiquitin pathway. Here we provide both genetic and biochemical evidence that Smo trafficking and degradation are regulated through multi-site ubiquitination of Smo C-tail and that Hh promotes Smo cell surface expression by inhibiting its ubiquitination. We also provide evidence that the non-visual β-arrestin Kurtz (Krz) acts in parallel with Smo ubiquitination to control Smo cell surface expression, and that the deubiquitinating enzyme UBPY promotes Smo cell surface expression by counteracting Smo ubiquitination.
In Drosophila wing discs, Smo cell surface level is low in anterior (A) compartment cells away from the A/P boundary but is elevated in response to Hh in A-compartment cells near the A/P boundary or in posterior (P) compartment cells (Figure 1A) [7]. To determine whether Smo is downregulated by the ubiquitin pathway, we generated mutant clones for Uba1, which encodes the only ubiquitin-activating enzyme (E1) in Drosophila [19],[20]. We employed a temperature-sensitive allele of Uba1, Uba1H33, which behaves like a null allele at the restrictive temperature [19]. Uba1H33 clones were induced at second instar larval stage (48–72 h AEL) by FRT/FLP mediated mitotic recombination. Larva carrying Uba1H33 clones were grown at permissive temperature (18°C) for 3 d and then shifted to non-permissive temperature (30°C) for 24 h before dissection for immunostaining. We found that anteriorly situated Uba1H33 clones accumulated high levels of Smo compared with neighboring wild type cells (Figure 1A–B'), suggesting that Smo is downregulated via the ubiquitin pathway in the absence of Hh. Immunostaining with anti-Smo antibody before membrane permeabilization suggested that Smo was accumulated on the cell surface in anteriorly situated Uba1H33 clones (Figure S1A–A'). A 12-h temperature shift resulted in a less robust Smo accumulation in Uba1H33 clones (Figure S1B–B'), likely due to the perdurance of Uba1 activity. In general, Smo elevation coincided well with Uba1 mutant clones. Intriguingly, Uba1H33 mutant cells situated in the posterior compartment also exhibited slightly higher levels of Smo than neighboring wild type cells (arrowhead in Figure 1B), suggesting that a fraction of Smo still undergoes ubiquitin-mediated degradation in the presence of Hh.
To examine whether Smo is directly ubiquitinated and whether Uba1 is responsible for this activity, we carried out a cell-based ubiquitination assay (see Materials and Methods) [21]. We employed RNAi and/or pharmacological inhibitor to inactivate Uba1. S2 cells stably expressing a Myc-tagged Smo (Myc-Smo) were treated with Uba1 or control double-stranded RNA (dsRNA) in the absence or presence of PYR-41, a cell permeable E1 inhibitor [22]. The efficiency of Uba1 RNAi was confirmed by Western blot analysis of an exogenously expressed tagged Uba1 (Figure 1C). Myc-Smo was ubiquitinated efficiently in the absence of Uba1 inhibition (Figure 1D); however, ubiquitination of Smo was attenuated by Uba1 RNAi and more significantly inhibited by PYR-41 (Figure 1D). The incomplete blockage of Smo ubiquitination by Uba1 RNAi is likely due to partial inactivation of Uba1 by the RNAi approach. Indeed, a combined treatment with Uba1 RNAi and PYR-41 resulted in a more complete inhibition of Smo ubiquitination (Figure 1D).
We next applied a cell-based immunostaining assay to determine whether Uba1 regulates Smo cell surface expression [13]. Myc-Smo expressing cells were treated with control or Uba1 dsRNA in the absence or presence of PYR-41. Cell surface and total Smo were visualized by immunostaining with an anti-SmoN antibody prior to and after cell membrane permeabilization, respectively. As shown in Figure 1E, inhibition of Uba1 either by RNAi or PYR-41 increased the levels of Smo cell surface expression and combined treatment resulted in more dramatic cell surface accumulation of Smo.
Ubiquitinated membrane proteins are internalized through the endocytic pathway and targeted to lysosome for degradation [17]. We therefore examined the effect of inactivation of endocytic components on Smo accumulation in wing imaginal discs. We found that Smo was accumulated in intracellular puncta in mutant clones lacking the Drosophila homolog of HGF-regulated tyrosine kinase substrate (Hrs) (Figure 2A–A'), a protein involved in sorting ubiquitinated membrane proteins into multivesicular bodies (MVBs) [23]. Of note, not all hrs mutant cells exhibited Smo puncta. This could be due to perdurance of Hrs activity and/or disc folding so that Smo puncta are present at different focal planes. RNAi of other endocytic components, including Tsg101 [24], Avalanche (Avl), a Drosophila syntaxin located in early endosomes [25], and Rab5, resulted in Smo accumulation in anterior compartment cells distant from the A/P boundary (arrows in Figure 2B–E), as well as Hh pathway activation as indicated by Ci accumulation and ectopic expression of a Hh target gene decapentaplegic (dpp) (Figure 2B–E). Taken together, these observations suggest that Smo is downregulated via the endocytic pathway in the absence of Hh.
Consistent with Smo being downregulated through the endocytic pathway, treating Myc-Smo expressing S2 cells with a lysosome inhibitor, NH4Cl, stabilized Smo (Figure 3A). Interestingly, treating cells with a proteasome inhibitor, MG132, stabilized Smo more dramatically than treating cells with NH4Cl (Figure 3A). Furthermore, combined treatment of cells with MG132 and NH4Cl had an additive effect on Smo stabilization (Figure 3A), suggesting that Smo is downregulated by both lysosome- and proteasome-dependent mechanisms. However, unlike the case of Hh stimulation or Uba1 inaction where Smo was accumulated on the cell surface, proteasome inhibition stabilized Smo in intracellular vesicles (Figure 3B). Double labeling with endosomal markers YFP-Rab5 (for early endosomes) or YFP-Rab7 (for late endosomes) revealed that Smo was stabilized in Rab7 positive late endosomes after MG132 treatment (Figure 3C). Taken together, these observations suggest that a fraction of internalized Smo was degraded by proteasome in the endocytic pathway before reaching to lysosome.
Hh induces Smo cell surface accumulation both in vitro and in vivo [7],[11],[13]. If Smo ubiquitination is responsible for its internalization, Hh may increase Smo cell surface expression by inhibiting its ubiquitination. Indeed, treating Myc-Smo stably expressing cells with Hh-conditioned medium markedly reduced but did not completely abolish Smo ubiquitination (Figure 4A). Similarly, Ptc RNAi also reduced Smo ubiquitination (Figure 4B).
Our previous study demonstrated that Hh induced Smo cell surface accumulation through PKA/CK1-mediated phosphorylation of Smo C-tail [13]. We therefore determined whether Hh regulates Smo ubiquitination in a manner depending on Smo phosphorylation. We found that Hh stimulation failed to inhibit Smo ubiquitination in the presence of a PKA inhibitor H-89 (Figure 4C). On the other hand, expressing a constitutively active PKA catalytic domain (mC*) inhibited Smo ubiquitination in the absence of Hh (Figure 4D). To further determine whether Smo ubiquitination is regulated by PKA/CK1-mediated phosphorylation of its C-tail, S2 cells were transfected with Myc-tagged wild type Smo, a phosphorylation deficient form of Smo (SmoSA) with three PKA sites (S667, S687, and S740) mutated to Ala, or a phospho-mimetic form of Smo (SmoSD) with three PKA/CK1 clusters mutated to Asp [13], treated without or with Hh conditioned medium, and followed by the ubiquitination assay described above. As shown in Figure 4E, Hh inhibited the ubiquitination of Myc-Smo but did not significantly affect the ubiquitination of Myc-SmoSA. Furthermore, the phospho-mimetic Smo mutant, SmoSD, exhibited diminished ubiquitination and its residual ubiquitination was further reduced by Hh treatment (Figure 4E–F). These results support the notion that Hh-induced phosphorylation by PKA/CK1 inhibits Smo ubiquitination, leading to its cell surface accumulation.
Our previous study revealed that the Smo autoinhibitory domain (SAID) inhibits Smo activity in part by preventing Smo cell surface expression because a Smo variant lacking the SAID domain (SmoΔ661–818 or SmoΔSAID) accumulated on the cell surface in the absence of Hh stimulation [8]. To determine whether the SAID domain regulates Smo ubiquitination, we examined the ubiquitin status of a Myc-tagged SmoΔSAID (Myc-SmoΔSAID). As shown in Figure 4G, deleting the SAID domain diminished Smo ubiquitination, and the residual ubiquitination of Myc-SmoΔSAID was further reduced by Hh treatment.
To determine whether the SAID domain suffices to promote ubiquitination and internalization of a heterologous membrane protein, we fused it to the C-terminus of the Wingless (Wg) receptor Frizzle 2 (Fz2) to construct Fz2-SAID chimeric protein (FS). When expressed in S2 cells, CFP-tagged Fz2 (CFP-Fz2) was largely accumulated on the cell surface with a small fraction internalized and colocalized with the endosomal marker Rab5 (Figure 5A–A”). In contrast, CFP-FS was barely detectable on the cell surface but largely accumulated in Rab5-positive endosomes (Figure 5B–B”), suggesting that SAID can promote endocytosis of Fz2. Introducing the phosphorylation-mimetic mutation to the SAID domain of FS (CFP-FS-SD) reduced its endocytosis (Figure 5C–C”), whereas the chimeric protein carrying a phosphorylation deficient form of SAID (CFP-FS-SA) was internalized as efficiently as CFP-FS (Figure 5D–D”). In addition, we found that adding the phosphorylation-deficient form but not the phospho-mimetic form of SAID to Fz2 promotes the ubiquitination of the corresponding chimeric protein (Figure 5E). Taken together, these observations suggest that the SAID domain suffices to promote ubiquitination and internalization of a membrane protein in a manner inhibited by phosphorylation. Combined with our earlier work [8], it seems that the SAID domain autonomously regulates ubiquitination independent of the C-terminal negatively charged region.
If Smo ubiquitination is responsible for its internalization and degradation, one would expect that ubiquitination-deficient Smo variants should be stabilized and accumulated on the cell surface. We therefore attempted to identify Lys residues responsible for Smo ubiquitination. In general, ubiquitin acceptor sites lack a strict consensus and target proteins can be ubiquitinated at multiple Lys residues. Smo C-tail and intracellular loops contain a total of 49 Lys residues, many of which may serve as ubiquitin acceptor sites, making it difficult to generate Smo variants devoid of ubiquitination. As deleting the SAID domain diminished Smo ubiquitination (Figure 4G), we speculated that this region might contain Lys residues critical for Smo ubiquitination. There are a total of 13 Lys residues between aa 661 and aa 818. We therefore constructed SmoK6R with K665, K695, K700, K702, K710, and K733 mutated to Arg; SmoK7R with K752, K753, K762, K772, K773, K782, and K801 mutated to Arg; and SmoK13R with all the 13 Lys residues mutated to Arg. Using the cell-based ubiquitination assay described above, we found that both Myc-SmoK6R and Myc-SmoK7R exhibited reduced ubiquitination compared with Myc-Smo (Figure 6A). The combined mutations (K13R) resulted in a more dramatic reduction in Smo ubiquitination (Figure 6A), suggesting that Smo is ubiquitinated at multiple Lys residues between aa 661 and aa 818. In addition, the residual ubiquitination of Myc-SmoK13R suggests that Smo is also ubiquitinated at one or more Lys residues outside the SAID domain.
We next determined whether the K13R mutation affects Smo stability and cell surface expression. Myc-Smo and Myc-SmoK13R expression constructs were transfected into S2 cells together with a Myc-CFP expression construct as an internal control. The levels of Myc-Smo and Myc-SmoK13R were monitored at different time points after treatment with the protein synthesis inhibitor, cycloheximide (CHX). As shown in Figure 6B, Myc-SmoK13R exhibited increased half-life compared with Myc-Smo, suggesting that inhibition of Smo ubiquitination leads to its stabilization. We also measured the steady state levels of Myc-Smo and Myc-SmoK13R in the absence or presence of MG132 and/or NH4Cl. While Myc-Smo was stabilized by both MG132 and NH4Cl, Myc-SmoK13R was stabilized by NH4Cl but insensitive to MG132 treatment (Figure 6C), suggesting that inhibition of Smo ubiquitination blocks its degradation by proteasome.
To determine whether inhibition of Smo ubiquitination leads to its cell surface accumulation, S2 cells were transfected with Myc-Smo or Myc-SmoK13R expression construct, followed by treatment with or without Hh-conditioned medium. Cell surface and total Smo were monitored by immunostaining with the anti-SmoN antibody before and after cell permeabilization, respectively. As shown in Figure 6D, Myc-SmoK13R exhibited higher basal level of cell surface expression than Myc-Smo; however, the level of cell surface Myc-SmoK13R in the absence of Hh was still lower than that of Myc-Smo or Myc-SmoK13R in the presence of Hh (Figure 6D). Thus, although SmoK13R exhibits increased stability and cell surface expression, it is still internalized and degraded by lysosome and can be further stabilized by Hh.
To determine whether the K13R mutation affects Smo stability in vivo, we generated transgenic flies expressing either UAS-Myc-Smo or UAS-Myc-SmoK13R from the same genetic locus using the phiC31 integration system to ensure similar expression level from different constructs [26]. We used the wing specific Gal4 driver MS1096 coupled with tub-Gal80ts to drive a pulse of UAS-Myc-Smo or UAS-Myc-SmoK13R expression by shifting late third instar larvae to the non-permissive temperature for 12 h. After chasing for different periods of time, wing discs were immunostained with an anti-Myc antibody. As shown in Figure S2, after a 10 h chase, Myc-Smo was barely detectable in A-compartment cells distant from the A/P boundary, whereas Myc-SmoK13R persisted in these cells, suggesting that Myc-SmoK13R has a longer half-life than Myc-Smo.
Internalization of SmoK13R is likely due to its residual ubiquitination at a Lys residue(s) outside the SAID domain. In addition, SmoK13R could also be internalized by Smo interacting proteins, as have been shown for other receptors [27],[28]. It has been shown that the non-visual arrestin, β-arrestin 2, can bind and internalize mammalian Smo [29]. The Drosophila non-visual arrestin is encoded by krz [30]. We therefore carried out both gain- and loss-of-function studies to determine whether Krz regulates Smo cell surface expression. We found that overexpression of Krz in wing imaginal discs using a dorsal compartment specific Gal4 driver, ap-Gal4, blocked Smo accumulation in posterior-dorsal compartment cells (compare Figure 7B with Figure 7A). However, we found that Smo was not accumulated in krz mutant clones located in the anterior compartment of wing discs (Figure 7C). Similar observations were obtained by a recent study [31].
Using a coimmunoprecipitation assay, we found that Smo interacted with Krz through its C-tail as both Myc-Smo and Myc-SmoCT (a Smo variant only containing its C-tail) but not Myc-SmoΔCT (a Smo variant with its C-tail deleted) pulled down a C-terminally YFP-tagged Krz (Krz-YFP) when expressed in S2 cells (Figure 7D). Furthermore, Krz-YFP could internalize SmoSD but not SmoΔCT in S2 cells (Figure 7F), suggesting that Krz internalizes Smo by binding to its C-tail. The association between Smo and Krz was attenuated by Hh stimulation because Myc-Smo pulled down less Krz-YFP in the presence of Hh conditioned medium (Figure 7E). In addition, Myc-SmoSD pulled down less Krz-YFP than Myc-SmoSA (Figure 7E), suggesting that Smo/Krz interaction is inhibited by Hh and PKA/CK1-mediated phosphorylation.
The observations that overexpression of Krz promoted Smo internalization but its loss of function did not lead to Smo cell surface accumulation suggest that a redundant mechanism(s) may act in parallel with Krz to internalize Smo. For example, in the absence of Krz, ubiquitination of Smo might be sufficient to promote its internalization and degradation. On the other hand, Krz could internalize Smo when Smo ubiquitination is compromised. This may explain, at least in part, why SmoK13R is still internalized and degraded by lysosome. To test this model, we examined the effect of Krz inactivation on the cell surface expression of Myc-Smo and Myc-SmoK13R in S2 cells. Consistent with the finding that loss-of-Krz has no effect on the cell surface expression of endogenous Smo in wing discs (Figure 7C), Krz RNAi did not significantly affect the cell surface expression of Myc-Smo in S2 cells (Figure 7G). In contrast, Krz RNAi increased the cell surface expression of Myc-SmoK13R (Figure 7G), suggesting that SmoK13R is, at least in part, internalized by Krz. Similarly, Krz RNAi enhanced the cell surface accumulation of Myc-Smo induced by Uba1 RNAi or PYR41 (Figure S3), suggesting that Krz acts in parallel with ubiquitination to internalize Smo. On the other hand, overexpression of Krz-YFP blocked the cell surface accumulation of Myc-SmoK13R and this blockage was alleviated by Hh treatment (Figure 7I), suggesting that Hh inhibits Krz-mediated Smo internalization.
Ubiquitination is a reversible process and ubiquitin attached to target proteins can be removed by deubiquitinating enzymes/DUBs [32]. Compared with the large number of E3 ubiquitin ligases that catalyze ubiquitination of targeted proteins, each genome encodes a much smaller number of DUBs. For example, the Drosophila genome encodes over 200 annotated E3s but less than 30 annotated DUBs (Flybase; Table S1). To determine whether Smo ubiquitination is regulated by DUBs, we systematically knocked down individual DUBs by RNAi and examined the effect on Smo ubiquitination in S2 cells stably expressing Myc-Smo. From this screen, we found that RNAi of the Drosophila UBPY/USP8 significantly increased the basal levels of Smo ubiquitination (Figure S4). The effect of UBPY RNAi on Smo ubiquitination was confirmed by an independent dsRNA for UBPY (Figure 8A). We also found that inactivation of UBPY by RNAi increased Smo ubiquitination in the presence of Hh (Figure 8A), suggesting that UBPY counteracts Smo ubiquitination in both Hh signaling “off” and “on” states. Consistent with UBPY being able to counteract Smo ubiquitination independent of Hh signaling states, overexpression of UBPY reduced Smo ubiquitination in S2 cells both in the absence and presence of Hh (Figure 8B).
We then carried out coimmunoprecipitation assays to determine whether UBPY physically interacts with Smo. As shown in Figure 8C, Myc-Smo and Myc-SmoCT but not Myc-SmoΔCT pulled down a flag-tagged UBPY (Fg-UBPY) when expressed in S2 cells, suggesting that UBPY interacts with Smo through its C-tail. The association between UBPY and Myc-Smo was not significantly affected by Hh stimulation (Figure 8D). Furthermore, UBPY appears to interact equally well with Myc-Smo, Myc-SmoSA, and Myc-SmoSD, suggesting that the bulk of Smo/UBPY association is not regulated by Hh signaling.
We next examined the effect of loss- or gain-of-UBPY on Smo cell surface expression. In wing discs carrying UBPY mutant clones, Smo cell surface accumulation was attenuated in P-compartment situated UBPY mutant cells (Figure 8E–E”). On the contrary, expression of UAS-UBPY using the wing specific Gal4 driver MS1096 resulted in Smo accumulation in anterior compartment cells away from the A/P boundary (Figure 8G,J). Similarly, overexpression of UBPY in S2 cells markedly increased the cell surface expression of Myc-Smo (Figure 8K). Overexpression of UBPY in wing discs stabilized full-length Ci (Figure 8G',J') and induced ectopic expression of dpp-lacZ in anterior dorsal compartment cells where MS1096 was expressed at high levels (Figure 8G”). Smo RNAi suppressed the ectopic dpp-lacZ expression induced by UBPY overexpression as well as the endogenous dpp-lacZ expression near the A/P boundary (Figure 8H–H”). However, overexpression of UBPY induced little if any ectopic expression of ptc-lacZ (Figure 8J”), which is normally induced by higher levels of Hh signaling than dpp-lacZ. Taken together, these results suggest that UBPY can reverse Smo ubiquitination to promote its cell surface accumulation and induce low but not high levels of Hh pathway activation. This is in line with our previous finding that overexpression of wild type Smo only induced low levels of Hh pathway activation and full activation of Smo requires additional steps, including a phosphorylation-mediated conformational switch in Smo C-tail [7]–[10],[13].
It is generally thought that monoubiquitination or multiubiquitination (monoubiquitination at multiple sites) is responsible for receptor internalization and degradation by lysosome, whereas Lys 48-linked polyubiquitination targets proteins for proteasome-mediated degradation. The observation that Smo is degraded by both lysosome and proteasome dependent mechanisms implied that Smo might undergo both types of modification. To determine if Smo could be monoubiquitinated, Myc-Smo or its KR variants was coexpressed with a HA-tagged mutant form of Ub with all Lys residues mutated to Arg (HA-UbK0) in S2 cells. In this case, addition of HA-UbK0 prevents the formation of polyubiquitination chains, generating modified proteins with one or more sites monoubiquitinated. We found that Myc-Smo was effectively modified by HA-UbK0 (Figure 9A). HA-UbK0 was also incorporated into Myc-SmoK6R, Myc-SmoK7R, and Myc-SmoK13R, albeit with reduced efficiency compared with Myc-Smo (Figure 9A), suggesting that Smo can be monoubiquitinated at multiple sites.
In the absence of proteasome inhibitor, HA-UbK0 and wild type HA-Ub were incorporated into Myc-Smo at similar levels (Figure 9B), suggesting that the ubiquitinated Smo species modified by HA-UbK0 or HA-Ub detected under these conditions were mostly mono- or multi-ubiquitinated. Furthermore, Hh stimulation inhibited Smo ubiquitination under these conditions (Figure 9B). However, after MG132 treatment, more HA-Ub conjugated Smo was detected than HA-UbK0 modified Smo (Figure 9B), suggesting that a fraction of Myc-Smo underwent polyubiquitination that was normally degraded by proteasome. The proteasome inhibitor also increased the level of HA-UbK0 conjugated Smo (Figure 9B), suggesting that a fraction of HA-UbK0 conjugated Smo might undergo polyubiquitination via endogenous Ub.
To confirm that Smo could be modified by Lys 48-linked polyubiquitination, we probed Smo immunopurified from S2 cells stably expressing Myc-Smo with a Lys 48-linkage specific polyubiquitin antibody (K48, Cell Signaling). As shown in Figure 9C, immunoprecipitated Myc-Smo was recognized by the K48 antibody and the signal was markedly increased by MG132 treatment, suggesting that Smo can also be modified by Lys 48-linked polyubiquitination that targets it for proteasome-mediated degradation.
Regulation of Smo cell surface expression is a key step in Hh signal transduction [7],[11],[13], but the underlying mechanism has remained unknown. In this study, we provide the first evidence that Smo is ubiquitinated in a manner regulated by Hh signaling and PKA/CK1-mediated Smo phosphorylation. We provide both genetic and biochemical evidence that Smo ubiquitination regulates its endocytic trafficking and cell surface expression. In addition, we provide evidence that the non-visual β-arrestin Krz acts in parallel with Smo ubiquitination to promote its internalization and that Smo ubiquitination is antagonized by the deubiquitinating enzyme UBPY.
Several lines of evidence suggest that the ubiquitin pathway regulates Smo endocytic trafficking and degradation: (1) Smo was accumulated in mutant clones lacking the ubiquitin-activating enzyme Uba1 in wing imaginal discs, and inactivation of Uba1 in S2 cells inhibited Smo ubiquitination and promoted its cell surface accumulation; (2) Smo was accumulated when the activity of several endocytic components or lysosome was inhibited; (3) Hh and PKA/CK1-mediated Smo phosphorylation inhibited Smo ubiquitination and increased Smo cell surface expression; (4) the Smo autoinhibitory domain (SAID) promoted receptor ubiquitination and internalization; (5) Smo was ubiquitinated at multiple sites both inside and outside the SAID domain and mutating the ubiquitin acceptor sites in SAID increased Smo half-life and cell surface expression; and (6) Smo cell surface expression was promoted by the deubiquitinating enzyme UBPY that binds Smo and counteracts Smo ubiquitination.
Early studies with yeast membrane receptors provided evidence that monoubiquitination of GPCRs mediates their agonist-induced internalization [33],[34]. Later studies with mammalian GPCRs and other receptors suggested that both mono- and polyubiquitination could be involved in receptor endocytosis and degradation [18]. However, it has been shown that “polyubiquitination” of some receptors is due to monoubiquitination at multiple sites (multiubiquitination) instead of forming a polyubiquitination chain at a single site [35],[36]. Here we provide evidence that Smo is both mono- and polyubiquitinated. It is possible that mono- or multiubiquitination may lead to Smo internalization and that internalized Smo could be further ubiquitinated in the endocytic pathway, leading to the formation of Lys 48-linked polyubiquitin chain that targets Smo for proteasome-mediated degradation (Figure 10). Thus, multiple ubiquitination events provide a robust mechanism for Smo downregulation to prevent aberrant Smo activity in the absence of Hh.
Regulation of Smo trafficking and cell surface expression provides a new paradigm for how the ubiquitin pathway controls the activity of a membrane receptor. Unlike all the other cases whereby receptor ubiquitination is triggered by ligand or agonist stimulation and serves as a mechanism to control the duration of cell signaling, Smo ubiquitination occurs in the absence of ligand stimulation and serves as a mechanism to keep the basal pathway activity in check. Smo ubiquitination is inhibited upon ligand stimulation; as a consequence, Smo is accumulated on the cell surface where it becomes activated. Thus, the regulation of Smo ubiquitination by the upstream signal is in the opposite direction compared with other receptors.
How does Hh block Smo ubiquitination? Smo intracellular regions such as SAID could recruit one or more E3 ubiquitin ligases to catalyze Smo ubiquitination and E3 recruitment could be inhibited by Hh stimulation and PKA/CK1-mediated Smo phosphorylation. An alternative but not mutually exclusive mechanism is that Hh and Smo phosphorylation could promote Smo deubiquitination by regulating the binding and/or activity of one or more DUBs. In a systematic RNAi-based screen, we identified UBPY as a Smo DUB. UBPY binds Smo C-tail and antagonizes Smo ubiquitination. UBPY may modulate Smo cell surface expression by attenuating Smo endocytosis and/or promoting Smo recycling (Figure 10). However, we found that UBPY decreases Smo ubiquitination regardless of the Hh signaling states and that the association between UBPY and Smo is not significantly affected by either Hh stimulation or Smo phosphorylation, suggesting that Smo deubiquitination by UBPY is unlikely to be a major mechanism by which Hh inhibits Smo ubiquitination, although we cannot rule out the possibility that Hh regulates UBPY binding to Smo in a subtle way that escaped the detection by our coimmunoprecipitation assay. The mechanism underlying the regulation of Smo ubiquitination might be analogous to those regulating the phosphorylation of many proteins in which kinases instead of phosphatases are usually regulated by upstream signals. Thus, identifying the E3 ligase(s) involved in Smo ubiquitination may shed important light on the mechanism by which Smo ubiquitination is regulated.
We have also obtained evidence that the non-visual β-arrestin Krz can promote Smo internalization by binding to its C-tail and this activity is inhibited by Hh. However, while Krz overexpression effectively internalized Smo, loss-of-Krz-function did not lead to a significant change in Smo cell surface expression (Figure 7C,G) [31]. Our results suggest that Smo ubiquitination can act independently of Krz to internalize Smo, leading to its degradation by both proteasome and lysosome so that the requirement of Krz in internalizing Smo can only be revealed when Smo ubiquitination is compromised (Figure 7G–I). It is possible that Smo ubiquitination plays a major role while Krz only plays a minor role in the regulation of Smo trafficking and cell surface expression.
The mechanisms that regulate Smo trafficking and cell surface expression exhibit interesting similarities to as well as important differences from those regulating GPCRs. For example, it has been shown that agonist-induced downregulation of β2-Adrenergic Receptor (β2AR) is mediated by both β-arrestin and receptor ubiquitination [27]. In addition, β2AR internalization and degradation is regulated by both proteasome- and lysosome-dependent mechanisms [27],[37]. However, β2AR ubiquitination is induced by agonist and serves as a mechanism for desensitization [27],[37], whereas Smo ubiquitination is inhibited by Hh and serves as a mechanism for keeping pathway activity off in the absence of the ligand. β-arrestin binding to β2AR is induced by agonists and requires GRK2-mediated phosphorylation of the activated receptor [27], whereas Krz binding to Smo is attenuated by Hh and Smo phosphorylation (Figure 7). Although GPRK2/GRK2 also regulates Smo in Drosophila, its function appears to be uncoupled from that of Krz because loss of GPRK2 exhibits a phenotype distinct from that exhibited by loss of Krz [31],[38]–[40]. Furthermore, Krz can internalize Smo in the absence of GPRK2 [31]. β-arrestin is required for β2AR ubiquitination [27],[37], whereas Krz inactivation does not significantly affect Smo ubiquitination (unpublished observations). Finally, while the proteasome inhibitor MG132 blocks agonist-induced β2AR internalization [27], it does not prevent Smo internalization but instead inhibits Smo degradation after internalization (Figure 3).
It is also interesting to note that β-arrestin has been implicated in the regulation of Smo trafficking and Shh signaling in vertebrates [29],[41],[42]. Furthermore, β-arrestin binds to mammalian Smo (mSmo) in a manner promoted by Shh and GRK2-mediated phosphorylation of mSmo C-tail [42],[43], which is analogous to agonist-induced β-arrestin binding to GPCRs. However, instead of internalizing mSmo for degradation, β-arrestin appears to promote mSmo ciliary accumulation [42], which correlates with its positive role in Shh signaling. Both Drosophila and vertebrate Smo proteins can activate trimeric G-proteins [44]–[46], suggesting that they are not only structurally but also functionally related to GPCRs. It is conceivable that Smo proteins may employ multiple mechanisms utilized by GPCRs to control their intracellular trafficking and activity. Thus, it will be interesting to determine whether vertebrate Smo is also regulated by the ubiquitin pathway.
Mutations used in this study are Uba1H33 [19], l(2)23AdD28/hrs [23], krz1 [47], and UBPYKO [48]. Mutant clones were generated by FLP/FRT-mediated mitotic recombination as previously described [49]. The genotypes for making clones are as follows: Uba1 clones: yw 122; FRT42 Uba1H33 /FRT42 hs-Myc-GFP; hrs clones: yw 122; l(2)23AdD28 FRT40/ hs-Myc-GFP FRT40; krz or UBPY clones: yw 122; FRT82 krz1 or UBPYKO /FRT82 hs-Myc-GFP. Transgenic RNAi lines used are UAS-Tsg101-RNAi (VDRC# 23944), UAS-Avl-RNAi (VDRC# 5413), and UAS-Rab5-RNAi (VDRC# 34096). UAS-Krz and UAS-UBPY are previously described [47],[48]. Constructs for various tagged forms of wild type Smo, SmoΔCT, SmoCT, SmoΔ661–818, SmoSA, and SmoSD are previously described [8],[13],[50]. CFP-tagged Fz2 is described [8]. To construct Fz2/Smo chimeric proteins, the coding sequence for the wild type and mutant forms of SAID (aa 661–818) was amplified by PCR and inserted at a Kpn I site between the coding sequence for Fz2 and CFP. To construct Krz-YFP, the coding sequence of Krz was amplified by PCR and inserted between Not I/ Kpn I digestion sites of pUAST vector, and YFP was inserted in frame to the C-terminus of Krz between Kpn I/ XbaI digestion sites. SmoK6R, SmoK7R, and SmoK13R were generated using PCR-based site-directed mutagenesis to introduce K to R mutations in corresponding Lys residues.
Drosophila S2 cells were cultured in Drosophila SFM (Invitrogen) with 10% fetal bovine serum, 100 U/ml of penicillin, and 100 mg/ml of streptomycin at 23°C. Transfection was carried out by Calcium Phosphate Transfection Kit (Specialty Media) according to the manufacturer's instructions. Hh-conditioned medium treatment was carried out as described [51]. Cells were treated with 50 µM MG132 (Calbiochem) for 4 h to inhibit proteasome or 20 mM NH4Cl (Sigma) for 18 h to inhibit lysosome. Immunoprecipitation and Western blot analysis were carried out using standard protocols as previously described [52]. For Smo cell surface staining assay, S2 cells were harvested and washed with PBS, fixed with 4% formaldehyde at room temperature for 20 min, and incubated with the mouse anti-SmoN antibody in PBS at room temperature for 90 min. Cells were washed 3 times by PBS followed by secondary antibody staining. Immunostaining of imaginal discs was carried out as described [13],[49]. Quantification of immunostaining and autoradiography densitometric analysis was performed using ImageJ software. Antibodies used in this study were: mouse anti-SmoN (DSHB), rat anti-Ci 2A1 [53], rabbit and mouse anti-Flag (Sigma), mouse anti-Myc (Santa Cruz), mouse anti-HA (Santa Cruz), mouse anti-GFP (Millipore), rabbit anti-GFP (Santa Cruz), rabbit anti-LacZ (ICN Pharmaceuticals, Inc.), anti-Ub (P4D1) (Santa Cruz), and anti-Poly-UbK48 (Cell signaling).
Ubiquitination assays were carried out based on the protocol described previously [21]. Briefly, Myc-Smo stably expressing S2 cells or S2 cells transfected with Smo variants with or without HA-Ub (wild type or mutants) were treated with MG132 or NH4Cl before harvesting. Cells were lysed in 100 µl of denaturing buffer (1% SDS/50 mM Tris, pH 7.5/0.5 mM EDTA/1 mM DTT). After incubation for 5 min at 100°C, the lysates were diluted 10-fold with lysis buffer and then subjected to immunoprecipitation and Western blot analysis.
dsRNA was generated by MEGAscript High Yield Transcription Kit (Ambion: #AM1334) according to the manufacturer's instruction. DNA templates targeting Uba1(aa 1–172), Krz(aa 191–365), UBPY(aa 25–191 and aa 124–290) or other DBUs (Table S1) were generated by PCR and used for generating dsRNA. Ptc RNAi was carried out as previously described [51]. dsRNA targeting the Fire Fly Luciferase coding sequence was used as a control. For RNAi knockdown experiments, S2 cells were cultured in serum free medium containing indicated dsRNA at 23°C for 8 h. After adding fetal bovine serum to a final concentration of 10%, dsRNA treated cells were cultured overnight before transfection. 48 h after transfection, cells were harvested for further analysis.
|
10.1371/journal.ppat.1000293 | Detailed Analysis of Sequence Changes Occurring during vlsE Antigenic Variation in the Mouse Model of Borrelia burgdorferi Infection | Lyme disease Borrelia can infect humans and animals for months to years, despite the presence of an active host immune response. The vls antigenic variation system, which expresses the surface-exposed lipoprotein VlsE, plays a major role in B. burgdorferi immune evasion. Gene conversion between vls silent cassettes and the vlsE expression site occurs at high frequency during mammalian infection, resulting in sequence variation in the VlsE product. In this study, we examined vlsE sequence variation in B. burgdorferi B31 during mouse infection by analyzing 1,399 clones isolated from bladder, heart, joint, ear, and skin tissues of mice infected for 4 to 365 days. The median number of codon changes increased progressively in C3H/HeN mice from 4 to 28 days post infection, and no clones retained the parental vlsE sequence at 28 days. In contrast, the decrease in the number of clones with the parental vlsE sequence and the increase in the number of sequence changes occurred more gradually in severe combined immunodeficiency (SCID) mice. Clones containing a stop codon were isolated, indicating that continuous expression of full-length VlsE is not required for survival in vivo; also, these clones continued to undergo vlsE recombination. Analysis of clones with apparent single recombination events indicated that recombinations into vlsE are nonselective with regard to the silent cassette utilized, as well as the length and location of the recombination event. Sequence changes as small as one base pair were common. Fifteen percent of recovered vlsE variants contained “template-independent” sequence changes, which clustered in the variable regions of vlsE. We hypothesize that the increased frequency and complexity of vlsE sequence changes observed in clones recovered from immunocompetent mice (as compared with SCID mice) is due to rapid clearance of relatively invariant clones by variable region-specific anti-VlsE antibody responses.
| Lyme borreliosis is the most common vector-transmitted infection in Europe and North America, and is caused by the spirochete Borrelia burgdorferi and other closely related Borrelia species. Lyme disease Borrelia have an elaborate mechanism for varying the sequence of VlsE, a surface-localized, immunogenic lipoprotein. This antigenic variation is thought to be important in immune evasion and thus in the ability of Lyme disease Borrelia to cause long-term infection. In this study, we examined 1,399 B. burgdorferi clones isolated from infected immunocompetent and immunodeficient mice to gain a better understanding of the rate and variety of VlsE sequence changes that occur during infection. We determined that clones with few or no VlsE sequence changes are rapidly cleared in mice with active immune responses, whereas clones with many VlsE changes persist. The vls antigenic variation system can utilize any of the 15 silent cassette sequences as sequence “donors,” and does not exhibit obvious preferences in the location of changes within the vlsE cassette region or the types of VlsE sequence variations found in different tissues, such as in joints or in the heart. Our findings provide further evidence that the vls locus represents a remarkably robust recombination system and immune evasion mechanism.
| Lyme borreliosis is caused by Borrelia burgdorferi and other members of the genus Borrelia, and is the most prevalent vector-borne disease in the United States [1]. Spirochetes are transmitted to mammalian hosts by Ixodes ticks, causing a local skin infection, usually accompanied by a lesion called erythema migrans. As the infection advances, Borrelia disseminate into deeper tissues despite a strong immune response against the pathogen [2]–[6]. However, Lyme disease Borrelia are able to escape clearance and cause disease manifestations (including neurologic, arthritic, cardiovascular, and dermatologic symptoms) for months to years after the initial infection.
Antigenic variation results from changes in surface antigen genes that occur during the course of infection at rates higher than the expected mutation frequency [7]. This mechanism is particularly important for organisms that cause long-term or repeated infections. Pathogens with antigenic variation systems are able to evade the immune response, thus gaining a selective advantage over their more antigenically stable counterparts and posing a challenge in the development of vaccines. Influenza virus [8] , HIV [9], Neisseria gonorrhoeae and N. meningitidis [10], Mycoplasma synoviae [11], Mycoplasma pulmonis [12], Anaplasma marginale [13], Borrelia burgdorferi [14], Borrelia hermsii [15],[16], Treponema pallidum [17], Campylobacter jejuni [18], Candida species [19], Plasmodium falciparum [20] and Trypanosoma brucei [21] are some examples of viruses, bacteria, fungi and parasites that avoid immune clearance through antigenic variation.
A surface-exposed lipoprotein, VlsE, contributes to immune evasion and persistence of Lyme borreliosis organisms in infected mammalian hosts through an elaborate antigenic variation mechanism [14], [22]–[25] . The Vmp-like sequence (vls) locus of B. burgdorferi B31 is located on the linear plasmid lp28-1. The vls locus is composed of an expression site (vlsE) encoding the 35 kDa lipoprotein VlsE and a contiguous array of 15 unexpressed (silent) vls cassettes. The silent cassettes have high homology to the central cassette region of vlsE (90.0% to 96.1% nucleotide sequence identity and 76.9% to 91.4% predicted amino acid sequence identity), and most of the sequence differences are concentrated in six variable regions within each cassette [22]. Clones lacking lp28-1 exhibit an intermediate infectivity phenotype, characterized by decreased persistence and aberrant tissue distribution in immunocompetent mice but no change in virulence in SCID mice [24],[26],[27]. Recent studies by Bankhead and Chaconas [23] demonstrated that removal of the vls locus by telomere-mediated truncation resulted in the same phenotype as the loss of lp28-1, whereas truncation of the other end of the plasmid had no detectable effect on mouse infection by needle inoculation. These results support the role of the vls locus in immune evasion.
Previous analysis of a limited number of clones recovered from experimentally infected mice or rabbits indicated that vlsE sequence variation occurs within 4 days and continues throughout the course of infection [25],[28]. Only the cassette region of the vlsE is subject to sequence variation during these recombination events. Segments of the vls silent cassette sequences replace portions of the vlsE cassette region through a gene conversion process, such that the sequence and organization of the silent vls cassettes remain unaltered [14]. vlsE antigenic variation has not been detected during in vitro culture or during tick infection, but occurs during mammalian infection in both immunocompetent and severe combined immunodeficiency disease (SCID) animals [14],[22],[25],[29],[30]. Attempts to induce vlsE recombination ex vivo have been unsuccessful. Therefore, the induction of vlsE recombination occurs through an as yet unidentified signaling mechanism.
Most sequence changes that occur during vlsE recombination events are localized within the six variable regions. The six invariable regions within the cassette region [22] contain relatively few variable codons and are likely to be important in preserving overall protein structure and biological function [31]. The variable regions form random coil structures on the membrane distal surface of the protein where antibody interactions are most likely [31]. Immunoglobulins specific for these regions are generated during the course of infection [32]. Also, the resulting variants exhibit decreased reactivity to antisera raised against a recombinant form of the vlsE cassette region from the parental clone; indicating that the sequence changes result in real antigenic variation [22]. The mechanisms that promote the selectivity and unidirectionality of gene conversion in the vls locus have not been identified.
In the current study, B. burgdorferi clones acquired 4 to 365 days following infection of immunocompetent or SCID mice were examined to gain a better understanding of the vlsE recombination process. The results provide further evidence of the remarkable randomness of recombination events occurring within the vlsE cassette region.
We analyzed the vlsE cassette region sequences of 1399 clones recovered during the time course of infection of immunocompetent C3H/HeN and immunocompromised C3H/HeN SCID and CB-17 SCID mice (Table 1). These results comprised 85 previously reported clones [14],[22] and 1320 clones derived during this study. The earlier studies were performed with B. burgdorferi B31clone 5A3 and utilized CB-17 SCID mice, whereas our recent analyses used clone B31 5A4 and C3H/HeN SCID mice. Although clone 5A3 is lacking plasmids lp28-2 and lp56 and CB-17 mice have a different genetic background than C3H/HeN, the results obtained were comparable (data not shown); therefore, the results obtained with the two B. burgdorferi B31 clones and the two SCID mouse strains were combined to increase the number of isolates and time points analyzed without necessitating additional animal experiments. Bladder, heart, joint, ear and back skin biopsy isolates were obtained to examine the rate and nature of vlsE recombination occurring in different tissues. In the current analysis, we focused on days 4, 7, 10, 14 and 28 post-inoculation because individual recombination events can be discerned more commonly at these earlier time points.
As previously observed [25], we found that clones had already undergone vlsE recombination within 4 days post infection in both immunocompetent and SCID mouse models (Figure 1). In immunocompetent mice, only 50% of the retrieved spirochetes retained the parental vlsE sequence after 4 days of infection, meaning that the remaining 50% of the population had already incurred one or more vlsE recombination events. By 14 days post infection, clones with the parental vlsE sequence were few in number (3% of all examined) and were not detected at 28 days after inoculation (Figure 1). In SCID mice, 87% of the recovered bacteria retained the parental vlsE sequence at 4 days post infection. The proportion of parental bacteria decreased more slowly than in immunocompetent mice, such that parental clones represented 18.7% and 15% of the populations recovered in SCID mice at 14 and 28 days post infection, respectively.
The parental bacteria thus persisted longer in the absence of an adaptive immune response. The rapid clearance of the parental genotype in immunocompetent mice actually preceded the detection of anti-VlsE antibodies by ELISA 8 days post infection in C3H/HeN mice [33]; this result suggests that the anti-variable region immune responses are present in small quantities within a few days of infection and are extremely effective in eliminating clones expressing the corresponding variable region epitopes. The more gradual decrease in the proportion of parental clones in SCID mice most likely represents the simple dilution of the initial genotype by variant clones.
At 4 days post infection, only the back skin biopsies (taken at a site distant from the inoculation site) and blood samples (not shown) exhibited positive culture results in C3H/HeN and SCID mice, suggesting that the spirochete had not colonized the other tissues examined to a detectable extent at this early time point (Table 1, Figure 2). At 7 days post infection in both mouse models, samples from the ear pinnae did not yield positive cultures, while all other sites were culture positive; this result indicates that the colonization of the external ear takes more time than the other tissues tested (Table 1, Figure 2). In comparing the different tissues, the proportion of clones with the parental vlsE sequence was not significantly different in either mouse model at day 7 post infection. Thereafter in C3H/HeN mice (but not in SCID mice), the proportion of parental clones dropped drastically in bladder, heart, and skin samples between 7 to 10 days post infection and in joint and ear samples between 10 and 14 days post infection (Figure 2A). These results indicate that the parental bacteria are cleared more quickly (or, alternatively, undergo more rapid vlsE recombination) in bladder, heart and skin than in joint and ear tissues in immunocompetent mice.
The more rapid clearance of B. burgdorferi with the parental vlsE sequence in heart, bladder and skin may indicate a higher accessibility of the bacteria to the adaptive immune system in these sites. Alternatively, bacteria in joint and ear tissues may localize in immunoprotective niches (e.g. in relatively avascular or highly collagenous regions) that allow those expressing the parental vlsE sequence to survive longer. In previous studies, it has been demonstrated that, in immunocompetent mice, B. burgdorferi clones lacking lp28-1 [24],[26],[33] or the vls locus [23] persist for longer periods in joint tissue than in other tissues. In contrast, organisms with these genotypes are able to infect and disseminate to all tested tissues in SCID mice. These results support the concept that immune evasion mechanisms provided by VlsE expression and sequence variation promote the survival of B. burgdorferi, but that bacteria that either do not express VlsE or have not undergone sequence variation are relatively protected in some tissues, such as those present in the tibiotarsal joint.
We analyzed more specifically the population of variants (n = 1,073) by excluding all clones with the parental vlsE sequence (n = 326). The group of variants included 921 ‘unique’ variant sequences and 158 additional sibling clones (i.e. variants with the same sequence in the same tissue specimen). The number of codon changes observed was paralleled closely by the number of amino acid changes (Figure 3), in concordance with the high proportion of nonsynonymous codon differences in the silent cassettes that serve as templates for these sequence changes.
In immunocompetent mice, the median number of codon or amino acid changes in the vlsE variant clones did not increase significantly between 4 to 10 days post infection, but at 14, 21 and 28 days post infection the number of changes increase rapidly and significantly (Figure 3A; P<0.001 for differences in the median number of changes on days 10 and 14, days 14 and 21, and days 21 and 28). There was no significant difference in the median number of changes at 28 days and 365 days post infection. The process of recombination in vlsE is still functional after 28 days, but the number of changes relative to the parental strains becomes asymptotic [25], as addressed further below. In immunocompromised mice, the number of codon or amino acid changes in vlsE was not significantly different when comparing 4 days and 14 days post infection (P>0.05). On days 14 and day 28, the number of changes was significantly lower in SCID mice than in immunocompetent C3H/HeN mice (P<0.001, Figure 3A and B). Thus, the immune pressure provided by the adaptive immune system not only results in the more rapid elimination of parental clones (Figure 1), but also selects for clones with more sequence changes and hence antigenic differences. These findings are again consistent with the observation that the presence of lp28-1 or, more specifically, an intact, functional vls locus [23] is required for long-term survival of B. burgdorferi in immunocompetent mice, but not in SCID mice.
Each of the 1,073 clones that had undergone vlsE sequence variation was examined individually to provide a global view of the length, location, and most likely silent cassette sources of the recombination events. As in previous analyses of vlsE sequence variation, segmental gene conversion events were observed; in no instance was the entire cassette region of vlsE replaced by a silent cassette. In most cases, the sequence changes could be attributed to a particular silent cassette sequence or set of potential donor sequences. However, in many instances, the donor sequence could not be identified unequivocally due to the high degree of sequence redundancy among the silent cassettes. Tentative identifications of recombination events and the corresponding donor sequences were thus based on those sequence alignments that incorporated the longest stretch of sequence changes (minimal recombination event) flanked by regions that were shared between the parental vlsE and vls silent cassette sequences (constituting the maximal possible recombination event).
We developed a method for visual, semi-automated analysis of the recombination events using Visual Basic macros in an Excel spreadsheet. An example is shown in Figs. 4A and 4B, in which the sequence of clone D10M8H8 (a variant isolated from a C3H/HeN mouse heart 10 days post infection) was aligned with the parental vlsE sequence and each of the silent cassette sequences. Each silent cassette (vls2 through vls16) is represented sequentially by a different colored bar. Solid color regions represent individual codons that have undergone a sequence change and match the corresponding silent cassette in the aligned sequences. Hatched color regions are contiguous codons that match both the parental and silent cassette sequences in that part of the alignment. As shown in Figure 4B, silent cassette vls13 (arrow) represents the most likely donor sequence due to the uninterrupted region of sequence identity spanning VR2 and VR3. Many silent cassettes match at one or more codons in this region, due to the high degree of redundancy among the vls at the individual codon level. However, vls13 is the only silent cassette that provides a contiguous match over this entire region. It is interesting to note that vls13 is the most likely donor sequence in two regions of the D10M8H8 sequence, separated by a short sequence that matches the parental vlsE sequence but not the vls13 sequence (Figure 4A). Thus this variant may represent an example of intermittent recombination (see below).
The lengths of predicted minimum recombinations varied widely, from a single nucleotide change (e.g. variant D7M5J12, Figure 4C) to a region of at least 372 nucleotides (e.g. variant D14M2B01, Figure 4C). In some cases, especially at early time points, some variant sequences exhibited several distinct regions of gene conversion using apparently the same silent cassette source, separated by regions of unchanged parental sequence (e.g. variants D10M8H8 [Figure 4A] and D7M6H09, [Figure 4C]). This ‘skipping’ appears to be due to alignment of a silent cassette sequence with the vlsE sequence over a long distance, followed by intermittent strand invasion and replacement of the vlsE sequence or intermittent cleavage of a single invaded strand. These observations indicate the occurrence of so-called “intermittent recombination events” in vlsE. An apparent intermittent recombination event in B. burgdorferi had been reported previously by Knight et al. [34]; in this case, a sequence containing the putative Shine Delgarno sequence had been ‘skipped’ during targeted allelic exchange of the gyrase A C-terminus (gac) gene.
Most of the sequence changes in vlsE could be explained as straight-forward genetic recombinations from silent cassette sequences to vlsE. However, genetic changes that could not be attributed to simple gene conversion events with silent cassette sequences were found in 167 clones (Table S1 and Figure S1). These ‘template-independent’ changes encompassed a variety of genetic events ranging from single nucleotide changes, apparent illegitimate recombination events, triplet repeat expansions/contractions, or other insertions or deletions of up to 9 base pairs; they also tended to cluster in the variable regions (Figure S2). Certain codons had particularly high rates of template-independent sequence changes; for example, template-independent sequence changes of codons 73–75 were identified in 19 variant sequences isolated from 10 different animals (11% of template-independent clones). Similarly, sequence variants containing template-independent changes of codons 194–199 were isolated 23 times from 13 different animals (14% of template-independent clones). It is unclear whether these areas represent mutation ‘hotspots’ or whether mutations arising in the variable regions are more likely to be maintained due to their location. The crystal structure of VlsE [31] reveals that the variable regions (VRs) form loop structures at the membrane distal surface of VlsE while the invariant regions (IRs) form structured alpha helical bundles. Mutations arising in the IRs might destabilize the proteins. Conversely, mutations in the VRs may aid the organism in antigenic variation and be maintained preferentially.
We did recover a large number of clones showing deletions or mutations in one IR region. Codons 10–15 in IR1 contained deletions or template independent changes in 24 variants from 12 different animals (14% of template-independent clones). Sequence changes in this region included a group of 7 variants (the last clones listed in Table S1) in which related sequences differed at between 9 and 12 of 18 nt in the vlsE1 sequence. These clones originated from skin and joint tissues of one mouse and the skin from another mouse 14 days post infection in the same experiment. These 18 nt sequences were not found elsewhere in the B. burgdorferi B31 genome sequence, so their source is unknown. (They are not cloning artifacts, because the PCR products were sequenced directly without cloning.) The amino acid sequence in this segment of VlsE1 is LLDKLV, whereas the corresponding variant sequences are SAVRKE, SAVQQK, SAVRQE and SADQKE. This region of IR1 is part of the α3 alpha helix in the VlsE structure [31]. Interestingly, the variant sequences preserve the alpha helical structure according to protein structure prediction programs (data not shown); thus these replacements most likely would not disrupt VlsE secondary structure.
Overall, sequences containing template-independent changes represented 15% (169 of 1,073) of vlsE variants, reinforcing the conclusion that they occur at a rate much higher than found in the rest of the B. burgdorferi genome [35]. These genetic mechanism(s) therefore may play an important role in antigenic variation of vlsE. Remarkably, only two of the 169 template-independent changes (a frame shift in D10M8H7 and a stop codon in SD14M4E1) represented interruptions in the open reading frame, indicating that the genetic mechanism(s) and/or selective pressure favor preservation of the full-length vlsE gene.
Many vlsE clones appear to have undergone multiple recombination events. No direct lineage of recombinations could be identified in most cases due to the high degree of sequence variation. In rare instances, we were able to identify clones that were likely in the same ‘recombination lineage’, i.e. represented a sequential series of recombinations. Figure 5 shows an example of three clones, recovered from a single day 14 mouse bladder specimen, that have apparently undergone sequential recombination events. In panel A, the first clone, D10M10B3 was predicted to be the result of an intermittent recombination with vls5 (the only silent cassette that contained all the sequence changes observed in both regions). The other two clones in the panel, D10M10B1 and D10M10B21, had the same sequence changes as D10M10B3 but also contain significant differences in other variable regions. Both clones contained identical sequence changes in the VR1 region that suggest a recombination event involving silent cassette 12 took place in a daughter of D10M10B3. However, D10M10B1 and D10M10B21 differ in VR6, consistent with these clones having undergone additional recombination events. Both clones contained sequence changes that are consistent with recombination with silent cassette 10 in VR6, but clone D10M10B21 exhibited additional changes in VR6 that suggest another recombination event with silent cassette 8 at some time following the recombination with vls10. Panel B summarizes the postulated sequence of recombination events: an initial recombination with vls5, followed by recombinations with vls12 and vls10 (in either order) and a final recombination with vls8 in clone D10M10B21. Thus we propose that these clones represent a series of sequential recombinations.
The median number of putative recombination events per clone was tabulated for each time point post infection (Figure 6). This value was found to increase significantly during the infection of immunocompetent C3H/HeN mice between day 7 and day 28 post infection (P<0.0001, Figure 6). The higher number of recombination events identified at 214 days and 365 days post infection provides further evidence that recombination continues to occur throughout the course of infection. On day 4 post infection, the number of recombination events is probably over-estimated, because several variant sequences contained intermittent recombination events (see below); by default, we considered an intermittent recombination as multiple recombination events. In contrast to the results obtained with immunocompetent mice, the number of deduced recombination events did not increase significantly between day 4 and day 28 post infection in SCID mice. These results again support a role of the adaptive immune system in the selection of clones with a higher number of putative recombination events. In a previous study, Anguita, et al. [36] examined a small number of B. burgdorferi clones and reported that the recombination at the vls locus is impaired in the absence of interferon (IFN)-γ-mediated signals. The proportion of clones that initiate vlsE gene conversion and the average numbers of changes per clone were lower in samples from IFN-γ receptor α-deficient mice than in wild-type mice [36]. In our study, we cannot exclude the possibility that the observed difference in the accumulation of vlsE variants in immunocompetent and SCID hosts is due in part to alterations in IFN-γ expression or other cytokine-mediated pathways. Other infectious agents (including Escherichia coli, Mycobacterium tuberculosis and Trypanosoma cruzi [37],[38],[39]) have developed diverse ways to subvert the immune system through the alteration of cytokine responses, so it is not outside the realm of possibility that vlsE antigenic variation is influenced by host cytokine production.
The silent cassette vls11 sequence contains a stop codon within invariable region 4 (IR4); recombination of this codon into vlsE would result in translational termination and a truncated polypeptide representing 63% of the full length VlsE [22],[25]. In the current study, 16 independent clones containing this stop codon in the vlsE sequence were isolated in the population of 1,399 clones analyzed. To determine whether a clone containing this stop codon in the vlsE sequence can colonize a mammalian host, two clones, 1379A and D7M5H5, were inoculated into immunocompetent C3H/HeN mice. The colonization of the mice was successful, as demonstrated by the detection of organisms in the skin at day 7 and in all tissues cultured at 28 days post infection (data not shown). We cloned sequences from the vlsE expression cassette to examine the ability of B. burgdorferi defective in full-length VlsE expression to undergo vlsE recombination in mice. Seven days after inoculation, bacteria recovered from back skin biopsies from 5 mice were analyzed for vlsE recombination. Interestingly, all 10 of the sequences analyzed still possessed the stop codon, although 50% of the clones showed changes in other parts of the vlsE sequence as compared to the sequence of the parental clone 1379A (data not shown) . At day 28 post-infection, vlsE sequences lacking the stop codon were recovered, indicating that sequences derived from the vls11 silent cassette are capable of undergoing recombination to generate full length VlsE. Taken together, these results indicate that continuous expression of a full length VlsE protein is not required for either the successful colonization of mice or the occurrence of vlsE recombination. This phenomenon could be considered a form of phase variation, as occurs in the pilin expression system in Neisseria species [10]. These data are also consistent with several previous studies indicating that B. burgdorferi clones lacking either lp28-1 or the vls locus can disseminate and survive for short periods (<18 days) in immunocompetent mice, yet can apparently survive indefinitely in SCID mice [23],[24],[33],[40],[41].
To investigate the length of individual vlsE recombination events, we performed a detailed examination of clones with only one apparent event of recombination. These results comprised 126 independent clones recovered from all tested tissues from both immunocompetent and SCID mice during the time course of infection (Table 2). Variant sequences with a single event of recombination encompassed a broad range of one (e.g. D7M5J12) to 22 (e.g. D14M2B01) codon changes (Figure 4C). The recombination observed in clone D7M5J12 represented only a GGG→AAG conversion at codon 84 in the aligned sequences, and could have arisen from any of the vls silent cassettes containing the AAG codon at this position (vls3, vls8, vls11, vls13, vls15 and vls16). In this case, the minimal recombination event comprised only two nucleotides, whereas possible maximal recombination events (the range in which the variant sequence matches both the ‘recipient’ and ‘donor’ sequences on either side of the sequence change) ranged from 2 to 17 nt upstream and 21 to 37 nt downstream, depending on the silent cassette involved. Overall, the recombination event in this example involved a maximum of 25 to 48 nt of DNA, indicating that vlsE recombination can take place in a very small region. At the other end of the spectrum, the putative recombination event with vls4 in clone D14M2B01 (Figure 4C) encompassed a minimum of 349 nt and a maximum of 423 nt of donor sequence, with 64 nt and 10 nt of sequence identity flanking the region of sequence change on the upstream and downstream ends, respectively. Thus the vlsE recombination system appears to promote both minuscule and long recombination events within the cassette region.
A subset of 126 vlsE variants was identified that appeared to represent ‘templated’ single recombination events (Table S2). All of the sequence changes in this group had corresponding template sequences in one or more silent cassettes, and most had regions of homology with both the ‘donor’ and ‘recipient’ sequences both upstream and downstream from the sequence change (e.g. D7M5J12, Figure 4C). The majority of these clones (70 of 126, or 55%) exhibited a minimum region of recombination of 1 to 5 nucleotides (Figure 7A) Amazingly, 33 of 126 (26%) had only a single nucleotide change (Figure 7A). These most likely represent templated gene conversion events, because they occur at a much higher frequency than that of template-independent single nucleotide sequence changes (76 of 1,073 sequences examined, or 7%) (Table S1).
An additional 28 clones (22%) exhibited minimal recombination events of 6 to 14 nt, whereas only 28 clones (21%) had minimal recombinations ≥15 nt. In contrast, the lengths of the predicted maximum recombination events were more widely distributed between 7 to 419 nucleotides (Figure 7B). (In this analysis, the maximal recombination event was based on the longest region of sequence identity if more than one silent cassette could have served as the ‘donor’ for the recombination event.) These results indicate that the length of DNA that is utilized during the recombination process is highly variable, but tends to include a short region of nonhomologous DNA. The observed median minimal recombination (±S.E.) was 5±0.27 nt, whereas the median maximal recombination (±S.E.) was 89±0.37 nt. The difference between minimal and maximal recombinations reflects the high homology between the silent cassette sequences and the vlsE cassette region. Thus, each round of vlsE gene conversion (i.e. each vlsE recombination event) often introduces only one or two amino acid changes in each variant protein sequence, although a much larger region may be involved in each recombination event. To examine this mechanism further, the base changes occurring in 33 well-defined ‘templated’ single nucleotide changes (proposed gene conversion events) were compared with 76 ‘template-independent’ changes (Table S3). While the proportion of nucleotide conversions were similar overall, C→A transversions were favored in the templated group, and C→G transversions were predominant in the template-independent group; this result implies that different mechanisms are operative in the two groups. While we cannot determine conclusively whether single nucleotide changes were introduced by genetic recombination or hypermutation as previously proposed by Sung et al. [35], our study indicates that most sequence changes in vlsE result from gene conversion events between the vls silent cassettes and the vlse1 expression cassette.
The putative silent cassette usage was determined for the sequence of clones showing a single, non-ambiguous recombination event (Figure 7C). We observed that some silent cassettes, including vls6, vls7, vls9, appeared to be used more often than the others. Although it had been proposed previously that the 17 bp direct repeat sequences present at the 5′ and 3′ ends of each silent cassette are involved in vlsE recombination [22], those silent cassettes with poorly conserved direct repeats (e.g. cassettes 10 and 11) are used during vlsE variation (Figure 7C). In the population of clones with a single recombination event, no clones were identified in which silent cassettes vls2, vls14 and vls16 were used as template for recombination. To extend this analysis, we also identified well-separated, unambiguous recombination events in all 1,073 clones with vlsE variations, including those with multiple recombination events. In this extended group, examples where vls2, vls14 or vls16 were unambiguously used as template were observed (data not shown). These results indicates that any region of any silent cassette can be use as template, although the silent cassettes present in the central part of the silent cassette locus tend to be used more frequently.
In our study, we analyzed the location of sequence changes within the vlsE cassette region during the time course of infection. No evidence of a recombination ‘program’ in which recombinations involved certain variable regions earlier than others was observed (data not shown). We also checked each individual variant amino acid sequence to determine if a specific VlsE protein sequence can be linked with a specific tissue tropism. In the relapsing fever organism Borrelia turicatae, the variable large protein/variable small protein (Vlp/Vsp) antigenic variation system influences tissue tropism as well as immune evasion [42],[43]. For example, B. turicatae expressing VspA are neurotropic while those expressing VspB achieve higher concentrations in the blood in a mouse model [42],[43]. In our study, there were no obvious differences in the amino acid sequence changes observed in different tissues (Figure 8). These results suggest that the vlsE gene conversion system is primarily involved in immune evasion rather than tissue tropism. Interestingly, we were also able to identify 5 pairs of clones presenting the exact same variant sequence in different tissues of the same mouse (e.g. D28M2BX1 from bladder and D28M2HX6 from heart); an additional 33 pairs of identical variants in different tissues were identified in other mice. These findings verify the occurrence of dissemination of variant clones in individual mice.
The vls antigenic variation system is an example of segmental gene conversion, which is also found in the N. gonorrhoeae pilE system [44], the A. marginale msp2 system [13], and the Babesia bovis ves1α system [45]. In each of these systems, a set of silent gene segments serves as the source of the ‘donor’ sequence, but the donor site remains unaltered in progeny following the recombination. Also, the recombination events occur at a high rate within the target gene, indicating that special mechanisms facilitate unidirectional genetic change in the target site (but not the donor sites). Another unusual property shared by segmental gene conversion mechanisms is that the recombination is ‘unanchored’ within the target gene, i.e. it does not start or stop at a certain sequence. The relapsing fever antigenic variation system is different in this aspect, in that most gene conversion recombination events occur at specific upstream and downstream homology sequences [16]. Our data also indicated that very short recombinations occur in the vls system, and that long flanking regions of sequence identity between the donor and recipient sequences are not required. Indeed, in our analysis of probable single recombination events, there were examples where there was no sequence identity on one end or the other of the recombination (e.g. clones D7M3B12, D10M9H6, and D10M7J4). The day 7 joint isolate D7M2J05 (data not shown) exemplified clones with short segments of sequence identity at both ends of the recombination, with regions of identity of only 2 nt and 6 nt upstream and downstream of an 11 nt region of sequence replacement (from cassette 10). These results indicate that the vls recombination system requires very little sequence identity to initiate the recombination event. In this regard, the vls system appears to be similar to the N. gonorrhoeae pilE system, in which sequence changes ranging in size from as short as 1 nt to as long as 200 nt have been observed; in addition, over 50% of the recombinations are 15 nt or less [44],[46],[47]. In no case, however, has illegitimate recombination into nonhomologous sites been observed in vlsE (or pilE), demonstrating that some extent of sequence complementarity and alignment is needed to ‘nucleate’ the recombination event.
The mechanisms involved in segmental gene conversion during antigenic variation are not well understood. pilE sequence variation is RecA-dependent [48], and appears to involve circular intermediates that are derived from pilS silent cassette sequences [49]. We propose that a vls silent cassette sequence (in the form of lp28-1, a separate episomal DNA copy, or possibly even an RNA copy) undergoes strand invasion, displacing the parental strain. This process requires very short regions of sequence identity and could be facilitated by a DNA-binding protein or endonuclease to nick the recipient and/or donor DNA, although the lack of specificity in terms of the site of sequence change suggests that these activities would not be very site-specific. The strand invasion also appears to be terminated in a non-specific manner, in that the lengths of recombination were variable (although predominantly short). Termination may not require a region of sequence identity, in that there were examples where no region of sequence identity was found at one end of the vlsE sequence change. We believe that as yet unidentified mediators of vlsE recombination are induced or activated during mammalian infection, as evidenced by rapid occurrence of vlsE sequence changes during mammalian infection and the lack of detectable sequence variation during in vitro culture or tick infection. (Alternatively, an inhibitor of vlsE recombination could be repressed or inactivated during mouse infection; however, this scenario seems unlikely in that vlsE recombination has not been observed in E. coli transformed with constructs containing vlsE and an adjacent region of the silent cassettes [S. J. N. and J. K. Howell, unpublished data].) Study of this phenomenon and its cis- and trans-acting mediators would be aided greatly by the identification of conditions that activate vlsE recombination in vitro, or vls shuttle constructs that undergo recombination in B. burgdorferi and can be genetically manipulated to permit the identification of cis-acting elements.
Antigenic variation and phase variation in bacterial surface proteins are common and have been shown to contribute to avoidance of adaptive immune responses, to tissue tropism, or to the pathogenesis process (e.g. altered adherence properties). Our studies provide direct in vivo evidence of the function of gene conversion of the Borrelial VlsE lipoprotein. In wild-type mice (in comparison with SCID mice), clones having the parental vlsE sequence persist for a shorter period; in addition, vlsE codon changes and recombination events accumulate more rapidly. These data indicate that variants are selected in immunocompetent mice, most likely due to antibodies specific for the variable regions of VlsE [32]. Similar results have been observed in the Vlp/Vsp antigenic variation system of relapsing fever Borrelia [50] and the Vsa phase variation system in Mycoplasma pulmonis [12]. The adaptive immune system thus acts as a selective force, killing clones with less variation but not eliminating clones with more extensive variation (and hence antigenic differences). In this study and as previously observed by Zhang and Norris [25], the presence or absence of the adaptive immune system is not required to induce the vlsE gene conversion mechanism. However, we cannot rule out the possibility that the adaptive immune system can directly affect the kinetics of the ongoing process of vlsE recombination, i.e. that the bacteria exhibit increased recombination under the influence of immune pressure (e.g. production of specific antibody or cytokines) [36]. Indeed, vlsE expression is increased under the influence of the immune pressure, specifically when functional B cells are present [51]. An interesting experiment would be to follow the rate of vlsE variant accumulation during the time course of infection of immunocompetent or SCID mice previously immunized with recombinant VlsE protein. An additional finding was that the production of a stable VlsE protein is not required for the vlsE gene conversion process to occur. Any silent cassette (and any region thereof) can be involved in a recombination event, and a variety of apparent template-independent genetic changes contributed to sequence variation. The recombination events are evenly distributed throughout the vlsE cassette region and exhibit no apparent bias for particular regions. Furthermore, no VlsE motif was associated with infection of a specific tissue site. The degree of variation observed indicates that the vlsE recombination system is one of the most robust antigenic variation systems found in pathogens.
The high-infectivity B. burgdorferi B31 clones 5A3 (B31-5A3, lacking plasmids lp28-2 and lp56) and 5A4 (B31-5A4, containing all plasmids) were isolated previously from low-passage strain B31 [27]. Small quantities were removed from the surface of frozen stocks by scraping with sterile 1 ml pipet tips and were inoculated into 6 ml of BSKII medium [52]. Cultures used in this study had undergone no more than two passages since clone isolation, thus minimizing the likelihood of plasmid loss.
All research involving animals was approved by the Animal Welfare Committee of the University of Texas Health Science Center at Houston. Eight-week-old, female C3H/HeN mice (Harlan, Indianapolis, IN), C3H/HeN severe combined immunodeficiency (SCID) mice (Harlan) and CB-17 SCID mice (Charles River Laboratories, Wilmington, MA) were housed in microisolator cages and provided with antibiotic-free food and water. For mouse inoculation, frozen stocks of low passage B. burgdorferi strains were cultured in BSK II medium [52] at 37°C in 3% CO2 until the mid-log phase of growth. The cultures were diluted in BSK II medium to a concentration of 106 bacteria/ml as determined by dark-field microscopy, and 0.1 ml (105 organisms) was injected subcutaneously at the base of the tail. Groups of 4 to 6 mice were sacrificed on 4, 7, 10, 14 and 28 days post infection, and samples from tissue sites (bladder, heart, joint, ear and skin) were acquired under aseptic conditions and cultured in 6 ml of BSK II broth with an antibiotic mixture to reduce the occurrence of microbial contamination (Sigma Aldrich; 50 µg/ml rifampin, 20 µg/ml phosphomicin and 2.5 µg/ml amphotericin B). After 7 days, the cultures were checked for growth, diluted, and subsurface plated in BSKII-agarose medium to obtain individual clones as described previously [53].
Well-isolated colonies from BSKII-agarose plates were inoculated in BSK II medium with antibiotics and cultured for 4 days prior to use as PCR templates (≈104 cells per reaction). Alternatively, agarose plugs containing individual colonies were added directly to the PCR reaction. The vlsE cassette region of each clone was amplified by PCR using the Phusion High-Fidelity DNA Polymerase (New England BioLabs, Ipswich, MA) and vlsE primers 4066 and 4120 as described previously [25]. The PCR products were purified and sequenced on both strands at the High-Throughput Genomics Unit (Department of Genome Sciences, University of Washington , Seattle), utilizing the same primers used for the amplification. The PCR products were sequenced directly (without cloning the products) to minimize the effects of sequence errors due to PCR infidelity. The chromatographs corresponding to each DNA sequence were examined individually to verify the quality of the sequence data, and each sequence difference (in comparison to the parental vlsE1 sequence) was checked for sequence accuracy. Variant clones that originated from the same tissue specimen and had identical sequences were considered siblings but were treated separately in this analysis.
The B31 parental vlsE (allele vlsE1), vls silent cassettes, and all of the vlsE variants sequences presented in this study are contained in GenBank entries U76406, U76405, EU484573–EU485396, EU485400–EU485403, EU485405–EU485714, EU485716–EU485724, EU485726–EU48748, and EU485750–EU485984; a list of the clone numbers and the corresponding GenBank accession numbers is at http://www.uth.tmc.edu/pathology/borrelia/. Most clone numbers are in the following format: S = SCID mouse infection; D4 = 4 days post infection; M3 = mouse 3; B, E, H, J, S = bladder, ear, heart, tibiotarsal joint, and skin, respectively; number and/or letter designations = individual clone from that animal and tissue. An ‘X’ indicates that a colony PCR product was sequenced, and no B. burgdorferi culture was retained. Infecting clone refers to either B31 5A3 (lacking plasmids lp28-2 and lp56) or B31 5A4 (containing all plasmids).
The DNA sequences of the parental vlsE cassette region (vls1) and the silent cassettes (vls2 to vls16) were aligned manually to match the arrangement in Figure 3 of Zhang et al. [22], and their sequences were formatted into codons (corresponding to vlsE codons). Indels were recorded using the letters “OOO” as a place marker. The aligned sequences were inserted into a Microsoft® Excel spreadsheet (one codon per cell), creating the template used to analyze vlsE variant sequences. Each vlsE variant sequence was codon-formatted, trimmed, and optimally aligned with vls1 (using the ClustalW multiple alignment algorithm embedded in the Bioedit software [http://www.mbio.ncsu.edu/BioEdit/BioEdit.html] (followed by manual adjustments) prior to analysis. The sequences were then compared to the parental vls1 sequence and the silent cassette sequences using a set of macros written using Microsoft® Visual Basic for Applications. The Excel template and macros may be obtained by contacting the authors.
The nucleotide and deduced amino acid sequences of each variant were compared computationally to both vls1 and the silent cassette nucleotide and predicted amino acid sequences. We first analyzed the overall number of codon differences between vls1 and the variant sequence. The codon sequence for each individual observed difference was compared to the sequences present among the silent cassettes at the same position, thus determining the putative silent cassette source(s) of any non-parental codon found in a given variant sequence. By combining the location and the possible silent cassette sources for each change in a variant sequence, we were able to identify regions of sequence variation and to propose putative events of recombination as well as the silent cassettes potentially used as templates. For each region of sequence variation, the minimal deduced regions of recombination were defined as groups of contiguous codons differing from the parental sequence and matching a silent cassette sequence, whereas the maximal deduced regions of recombination included all contiguous codons in either direction in which the variant sequence matched both the parental and silent cassette sequences. The possible recombination events were thus determined computationally and portrayed graphically by the Excel™ spreadsheet, as exemplified in Figure 4. In cases where more than one silent cassette could serve as the template for a recombination event, the silent cassette showing the longest maximal recombination pattern was selected as the possible template.
By comparing the silent cassette sequences and the vls1 sequence, we determined the probability of change at each codon in vls1. Each vlsE variant sequence was then compared to vls1 to determine the number and position of codon and amino acid differences in that variant. The results obtained for variant sequences at a given time post infection were analyzed together, and the total number of differences at each position was calculated, normalized for the number of variants, and compared with the probability of change at each position in vls1. The deduced amino acid changes occurring at each position were also compared to probability data obtained from vlsE1/silent cassette comparisons and displayed using the program WebLogo [54].
Statistics were performed in Microsoft® Excel using the unpaired Student's t test.
|
10.1371/journal.pgen.1007555 | Multilevel effects of light on ribosome dynamics in chloroplasts program genome-wide and psbA-specific changes in translation | Plants and algae adapt to fluctuating light conditions to optimize photosynthesis, minimize photodamage, and prioritize energy investments. Changes in the translation of chloroplast mRNAs are known to contribute to these adaptations, but the scope and magnitude of these responses are unclear. To clarify the phenomenology, we used ribosome profiling to analyze chloroplast translation in maize seedlings following dark-to-light and light-to-dark shifts. The results resolved several layers of regulation. (i) The psbA mRNA exhibits a dramatic gain of ribosomes within minutes after shifting plants to the light and reverts to low ribosome occupancy within one hour in the dark, correlating with the need to replace damaged PsbA in Photosystem II. (ii) Ribosome occupancy on all other chloroplast mRNAs remains similar to that at midday even after 12 hours in the dark. (iii) Analysis of ribosome dynamics in the presence of lincomycin revealed a global decrease in the translation elongation rate shortly after shifting plants to the dark. The pausing of chloroplast ribosomes at specific sites changed very little during these light-shift regimes. A similar but less comprehensive analysis in Arabidopsis gave similar results excepting a trend toward reduced ribosome occupancy at the end of the night. Our results show that all chloroplast mRNAs except psbA maintain similar ribosome occupancy following short-term light shifts, but are nonetheless translated at higher rates in the light due to a plastome-wide increase in elongation rate. A light-induced recruitment of ribosomes to psbA mRNA is superimposed on this global response, producing a rapid and massive increase in PsbA synthesis. These findings highlight the unique translational response of psbA in mature chloroplasts, clarify which steps in psbA translation are light-regulated in the context of Photosystem II repair, and provide a foundation on which to explore mechanisms underlying the psbA-specific and global effects of light on chloroplast translation.
| Our experiments address the effects of light on protein synthesis within chloroplasts, whose ~80 genes are essential for photosynthesis and account for a large fraction of the protein synthesis in leaf tissue. Light is necessary for photosynthesis but it also triggers photo-oxidative damage. It is known that light-induced changes in chloroplast translation aid photosystem repair and prioritize energy consumption in the dark. However, prior studies have been limited in scope and offer conflicting views of the nature of these responses. In this study, we used new methods to generate a comprehensive description of chloroplast translation in maize at various time scales after shifting plants from dark to light and vice versa. We discovered that psbA mRNA, which encodes a protein that is particularly prone to photodamage, exhibits dynamic changes in ribosome occupancy in response to light and that it is unique in this regard. Ribosome occupancy on other chloroplast mRNAs is static even after many hours in the dark; however, these mRNAs are nonetheless translated at reduced rates in the dark due to a reduced rate of ribosome movement. Resolution of these superimposed effects clarifies the phenomenology of light-regulated chloroplast translation and provides a basis for exploring underlying mechanisms.
| Energy from sunlight fuels life on earth through the process of photosynthesis. Light is both an essential resource and a source of stress for photosynthetic organisms, as it damages cellular structures through photo-oxidative processes and the production of reactive oxygen species. Photosynthesis is compromised when excess light damages the Photosystem II (PSII) reaction center or when excitation of Photosystem I (PSI) and PSII is unbalanced [1, 2]. A key target of photodamage is the D1 reaction center protein of PSII, which is encoded by the chloroplast psbA gene. In an elaborate repair cycle, damaged D1 is degraded and then replaced with newly synthesized D1 [3, 4]. To compensate for its high rate of turnover, D1 is the most rapidly synthesized protein in photosynthesizing cells. In plants and algae, light exerts rapid changes in D1 synthesis at the level of translation [5]. The phenomenon of light-regulated psbA translation has been intensively studied due to its central role in maintaining photosynthetic homeostasis and the ease of monitoring D1 synthesis. However, light-regulated translation in chloroplasts is not limited to the psbA mRNA [6]. This has been documented most thoroughly for the rbcL mRNA, whose translation initiation and elongation rates have been shown to change in response to light [7–12].
Despite a large body of literature on light-regulated chloroplast translation, major gaps remain in the characterization of the basic phenomenology: i.e. which genes respond to light at the translational level, with what kinetics, and at which step in translation. Studies have been limited to the few most rapidly synthesized proteins, so no information is available for the majority of the chloroplast’s ~80 protein-coding genes. In addition, many studies examined light responses in plants that had been grown for extended periods in the absence of light, at which point effects on chlorophyll synthesis and energy supply confound data interpretation. For example, an influential series of reports employed a de-etiolation regime in which barley seedlings were germinated and grown in the absence of light for many days. Illumination of these plants triggered the incorporation of radiolabeled amino acids into chlorophyll binding proteins in isolated chloroplasts, a result that was initially interpreted as light-induced translation [13]. However, subsequent experiments showed that much of this effect was due to stabilization of nascent apoproteins by chlorophyll, whose synthesis is induced by light [14, 15]. The degree to which regulated translation contributed to that phenomenon remains unclear. Further ambiguities arise from the fact that many studies assayed translation in isolated chloroplasts or etioplasts [5], whose energy and redox status may differ from those in vivo.
In this study, we revisit the phenomenon of light-regulated chloroplast translation using ribosome profiling [16], a method that was not available at the time of the work summarized above. Ribosome profiling provides a genome-wide, quantitative and high-resolution snapshot of ribosome occupancy in an intact organism at the time of harvest. To minimize effects of light on energy supply, we used young maize seedlings grown in diurnal cycles prior to depletion of their seed reserves, and we monitored ribosome occupancy shortly after shifting plants from the light to the dark, or vice versa. We distinguished effects of light on translation initiation and elongation rates by following the kinetics of ribosome occupancy after introducing lincomycin, which specifically inhibits ribosomes at the first few codons in an open reading frame (ORF). Our results show that transition from dark to light is accompanied by a global increase in the rate of translation elongation in chloroplasts, and that the rapid recruitment of ribosomes to the psbA mRNA is superimposed on this global response. Surprisingly, the abundance and positions of ribosomes on all chloroplast mRNAs other than psbA are maintained largely unchanged even after 12 hours in the dark, implying a balanced decrease in rates of initiation and elongation. This comprehensive analysis clarifies the phenomenology of light-regulated chloroplast translation and provides a basis for mechanistic hypotheses to be tested in future studies.
We grew maize seedlings for 8 days in light-dark cycles. Plants at this stage are photosynthetically competent but have not exhausted seed reserves. We subjected these plants to one of three light-shift regimes (Fig 1A): (i) Plants that experienced 15 minutes of light at dawn were compared with plants that were maintained in the dark and harvested at the same time; (ii) Plants that experienced one hour of dark at midday were compared with plants that were maintained in the light and harvested at the same time; (iii) Plants that were reilluminated for 15 minutes after one hour of dark at midday were compared with plants harvested prior to reillumination. We used a moderate light intensity (~300 μmol m-2s-1) to minimize photodamage. Three biological replicates were performed for each comparison.
Leaf tissue was processed for ribosome profiling (Ribo-seq) and RNA-seq analysis as described previously [17]. Ribo-seq reads mapping to the nuclear and chloroplast genomes exhibited the expected size distributions and three-nucleotide periodicity, and mapped primarily to protein-coding sequences (S1A–S1D Fig), demonstrating that they derive primarily from ribosome footprints. Read counts from chloroplast genes were normalized to the length of the ORF and sequencing depth by expressing them as reads per kilobase (in the ORF) per million reads mapped to nuclear protein coding sequences (RPKM). The Ribo-seq and RNA-seq RPKM values were highly reproducible across the replicates (Pearson Correlation ~ 0.99, S1E Fig). Every chloroplast gene was represented by at least 47 Ribo-seq reads and 1000 RNA-seq reads in each replicate, and the majority were represented much more deeply than this (S1F Fig).
The results are presented in Fig 1B as the ratio of values in light to dark for each of the three comparisons described above. The abundance of ribosome footprints on psbA RNA is highly dynamic in response to light, increasing approximately 6-fold after 15 minutes of light at dawn, decreasing approximately 8-fold after 1 hour in the dark at midday, and increasing approximately 8-fold after 15 minutes of reillumination at midday (see Ribo-seq, top panel). Strikingly, the psbA mRNA was the only chloroplast mRNA to behave in this manner: ribosome footprint abundance on all other chloroplast ORFs changed less than two-fold in each comparison. The abundance of psbA mRNA did not vary (see RNA-seq, middle panel), indicating that the changes in ribosome footprint abundance are due solely to effects on translation. Several other mRNAs showed small changes in ribosome density (ratio of Ribo-seq to RNA-seq reads) in response to light, but the change exceeded 2-fold only for ndhJ at midday (Fig 1B bottom). To assess whether additional differences become apparent at longer times after shifting from one condition to the other, ribosome density after 7 hours in the light (midday) is compared to that after 12 hours in the dark (dawn) in Fig 1C. This comparison shows that ribosome density on all chloroplast ORFs excepting psbA changes very little even after extended times in light or dark.
Normalized read counts for each condition are plotted separately in Fig 2 and S2 Fig. This view of the data shows that mRNA abundance across all five conditions exhibited only small variations, although mRNAs from several genes were approximately 2-fold more abundant at dawn than at midday regardless of light condition. Furthermore, all chloroplast mRNAs other than psbA maintained similar ribosome density (Ribo-seq/RNA-seq) at all time points. The Ribo-seq RPKM from psbA after 1 or 12 hours in the dark was similar to that from other genes encoding photosystem subunits. Illuminating plants for 15 minutes restored psbA ribosome occupancy to the same level as that in plants that had been maintained in the light for 7 hours.
These findings were corroborated by results from traditional polysome assays in which the association of mRNAs with ribosomes was assessed by their rate of sedimentation through a sucrose gradient. Consistent with the ribosome profiling data, the distribution of psbA mRNA shifted toward lower molecular weight fractions shortly after shifting to the dark (and vice versa), whereas the sedimentation profiles of the rbcL and atpB/E mRNAs were indistinguishable even when comparing plants that had been in the dark or light for many hours (Fig 3 and S3 Fig). Although rapid sedimentation could result from the association of mRNAs with large non-ribosomal ribonucleoprotein particles, the ribosome profiling data argue against this possibility: the ribosome footprints exhibited similar size, 3-nucleotide periodicity and confinement to ORFs among all assayed conditions (S1 Fig). Together, these findings strongly support the interpretation that light has minimal impact on the ribosome-association of the vast majority of chloroplast mRNAs, whether comparing plants that have been shifted from one condition to the other for short (15 minutes) or long (many hours) periods of time. By contrast, cytosolic ribosomes shift toward smaller polysomes after 1-hour in the dark (Fig 3 bottom, 25S and 18S rRNAs), as reported previously for Arabidopsis [18].
Pioneering studies detected paused ribosomes at several positions in the psbA RNA in isolated barley etioplasts, and showed that these pauses increase in magnitude after many hours of illumination [19, 20]. To assess the effects of light on ribosome pausing in developed chloroplasts, we examined the distribution of ribosome footprints along specific chloroplast ORFs; peaks in the distribution can be used to infer positions at which ribosomes dwell for an unusually long time [21]. Ribosome distribution along psbA was remarkably similar in material harvested at each assayed time point following a shift from one condition to the other (Fig 4), arguing against a role for regulated ribosome pausing in light-regulated psbA expression. Analogous plots for six other chloroplast ORFs are shown in Fig 4, and likewise showed very similar ribosome distributions in all conditions; that said, a site near the petA start codon appeared to be more highly occupied in the dark than in the light. A quantitative, plastome-wide comparison of ribosome distributions detected several locations at which ribosome dwell-time may differ to a small degree in dark and light (S4 Fig). Although several of these were reproducible in our experiments, their statistical significance is unclear. These possible exceptions aside, our results demonstrate that light does not cause wide-spread or high-magnitude changes in the pausing of ribosomes at specific locations on chloroplast mRNAs.
To examine the kinetics with which ribosomes are lost from psbA mRNA upon a shift to dark, plants were processed for ribosome profiling over a time course after shifting plants to the dark at midday (Fig 5A). A small reduction in ribosome footprints was apparent after 10 minutes in the dark, whereas 30 minutes was sufficient (or nearly so) to reduce ribosome occupancy to that after 1 hour in the dark. These data, together with those shown in Fig 2, show that new steady-state ribosome occupancies on psbA RNA are established within approximately 15 minutes after shifting plants to the light and 30 minutes after shifting plants to the dark.
Although ribosome occupancy is typically a good proxy for relative rates of protein synthesis within a cell (or organelle) under any single condition [22, 23], the relationship between ribosome occupancy and protein synthesis becomes unpredictable when comparing different conditions due to possible differences in rates of translation elongation. In fact, classic studies reported a decrease in the translation elongation rate on the psbA and rbcL mRNAs after shifting isolated chloroplasts (psbA) or intact plants (rbcL) to the dark [8, 24–27]. To address this possibility in maize seedlings, we used in vivo pulse-labelling assays to examine how the rates of RbcL and PsbA synthesis change over time after shifting plants to the dark or light (Fig 5B). Pulse-labeling was performed during four consecutive 15-minute windows following shifts to dark and back to light at midday. The results show that PsbA synthesis drops rapidly after the shift to the dark and increases rapidly after reillumination, correlating in a general sense with the ribosome profiling data. Notably, however, the decrease in PsbA synthesis was apparent in the first 15 minutes after the shift to the dark (lane “0” in Fig 5B), preceding the clearance of ribosomes from its mRNA. Although labeled RbcL was poorly resolved in these experiments, the results suggest that RbcL synthesis decreased after the shift to dark and increased again after the shift to light, despite the unchanged association of its mRNA with ribosomes. These results suggest that the rate of ribosome movement along the psbA and rbcL mRNAs slows shortly after the shift to dark, and is restored shortly after the shift back to the light.
Results above show that ribosome occupancy on all chloroplast ORFs other than psbA changes very little following shifts from light-to-dark or dark-to-light, even after many hours in the new condition. This is consistent with two scenarios: either light has minimal effect on their translation or it triggers concerted changes in rates of initiation and elongation such that the average number of ribosomes associated with each mRNA is maintained. The data shown in Fig 5 together with the prior studies discussed above support the view that the rate of ribosome movement along the psbA and rbcL mRNAs does decrease soon after a shift to the dark. However, it is not known whether other chloroplast mRNAs are similarly affected.
To provide a plastome-wide view of the effects of light on translation elongation, we performed ribosome profiling over a time course following treatment of seedlings with the peptidyl-transferase inhibitor lincomycin (Fig 6). Lincomycin does not inhibit ribosomes harboring nascent peptides longer than approximately five amino acids, so ribosomes that have passed the first few codons continue to elongate in its presence [28]. Thus, changes in the translation elongation rate will be reflected by changes in the rate of ribosome clearance from ORF bodies after treatment with lincomycin (see Fig 6A). Lincomycin does not inhibit cytosolic ribosomes, so we normalized chloroplast read counts to cytosolic read counts.
These experiments required that lincomycin be introduced into the chloroplasts of intact seedlings as rapidly as possible. Of the approaches we explored (see Materials and Methods), we found the introduction of lincomycin via thread wicks sewn through the stem to be most effective. Pilot experiments demonstrated that chloroplast protein synthesis in the leaf is inhibited starting approximately 10 minutes after initiating this treatment. Therefore, we harvested leaf tissue for ribosome profiling immediately prior to lincomycin treatment, and 12 and 30 min after initiating treatment (Fig 6A, bottom). For analysis of elongation rate in the dark, plants were dark-adapted for 30 minutes prior to lincomycin treatment, and were maintained in the dark throughout the treatment. During the 30-minutes of dark-adaptation, ribosome occupancy on psbA mRNA is reduced to its dark steady-state level (see Fig 5A); therefore, this experiment monitored the elongation rate only of those ribosomes that remained bound to psbA RNA after that time. We observed that the abundance and size distribution of ribosome footprints mapping to chloroplast start codons changed over time following lincomycin treatment (S5 Fig). The change in footprint size likely results from the fact that lincomycin traps ribosomes in a “rotated” conformation [29]. This effect was similar in the light and dark (S5 Fig), demonstrating that lincomycin inhibited chloroplast ribosomes in both conditions.
The distribution of ribosome footprints along the rbcL and psbA mRNAs at each time point following lincomycin treatment in the dark or light is displayed in Fig 6B (top). The rate of ribosome clearance is much slower in the dark than in the light, consistent with the prior evidence that light increases the rate of translation elongation on the psbA and rbcL mRNAs [8, 24–27]. Analogous plots for other genes show similar effects (Fig 6B bottom, S6A Fig), indicating that a reduction in translation elongation rate upon a shift to the dark is not specific for psbA and rbcL. We did not observe an obvious shift in ribosome footprints toward the 3’-end of ORFs over the lincomycin time course. This may be due to the fact that treatment of intact plants results in highly asynchronous exposure of chloroplasts to the antibiotic.
To provide a plastome-wide accounting of rates of translation elongation rate in light versus dark, ribosome footprint abundance on each ORF at each time point following lincomycin treatment is plotted as a percentage of the value just prior to lincomycin treatment in Fig 6C. The results show that the rate of ribosome clearance is, in general, considerably slower in the dark than in the light. A correlation plot of the ratio of ribosome footprints on each gene in light versus dark after 30 minutes of lincomycin treatment in the two replicates (S6C Fig) supports this conclusion, and suggests further that ORFs may differ in the degree to which the translation elongation rate drops following a shift to the dark. However, systematic differences between replicates suggest that varying transport of lincomycin to the apical region of the leaf accounts for some or all of this variation (see S6C and S6D Fig). Thus, additional experiments will be necessary to make firm conclusions about gene-specific differences in translation elongation rates.
We attempted to monitor the effects of light on translation initiation rates by following the buildup of ribosomes at start codons following lincomycin treatment. Ribosomes build up at the psbA start codon much more rapidly in the light than in the dark following lincomycin treatment (Fig 6B), consistent with the ribosome profiling and polysome data indicating that light stimulates the recruitment of ribosomes to the psbA mRNA. Furthermore, a metagene analysis showed that, on average, ribosomes build up at start codons more rapidly in the light than in the dark (S5 Fig). However, our ability to make inferences about initiation rates was complicated by several considerations. First, many chloroplast start codons did not accumulate ribosomes following lincomycin treatment either in the light or in the dark. Furthermore, the rate of ribosome build-up at start codons will be influenced by the reservoir of mRNAs with a vacant translation initiation region, a parameter that we cannot assess. That said, results above showing that (i) ribosome occupancy and distribution on most chloroplast ORFs is similar in the light and dark, (ii) the rate of ribosome movement along these ORFs is slower in the dark, and (iii) the average rate of ribosome buildup at start codons is greater in the light than the dark in lincomycin-treated plants, imply that the rate of translation initiation for most chloroplast mRNAs changes roughly in concert with changes in the elongation rate following light shifts.
Unexpectedly, we observed that ribosome footprints accumulated to high levels at a number of sites outside of translation initiation regions over the lincomycin time course, even as bulk ribosomes cleared these genes as expected (see Fig 7A and S6B Fig for examples). The positions and magnitude of this feature were reproducible between replicate experiments (Fig 7B). The size-distribution of the footprints that build up at these sites was similar to that of lincomycin-bound ribosomes at start codons and distinct from that of elongating ribosomes (S5 Fig bottom), suggesting that these are footprints of ribosomes in the initiation mode. To examine that possibility, we compared the sequences around these “ribosome build-up” sites to those at sites from which ribosomes cleared following lincomycin treatment (Fig 7C). The sites of ribosome build-up (Fig 7C top) show a strong enrichment for sequences resembling ribosome binding sites: a start codon preceded by a predicted Shine-Dalgarno element. Several observations argue against the possibility that these are stalled elongating ribosomes: (i) the size distribution of their footprints resembles that of initiating ribosomes and not elongating ribosomes; (ii) many of them are out of frame with the ORF in which they reside and several are in UTRs; (iii) the abundance of ribosomes that accumulate at many of these sites exceeds that of the ribosomes found upstream prior to lincomycin treatment (see e.g. psbL in S6B Fig).
These observations strongly suggest that translation initiation complexes formed anew at ectopic sites following lincomycin treatment. This may be a consequence of the clearing of ribosomes from ORFs following lincomycin treatment, which will increase the accessibility of sequences resembling translation initiation regions while also increasing the concentration of free ribosomes available for initiation. Regardless of the mechanism, these footprints obscured calculations of ribosome run-off rates, and were therefore excluded from calculations of ribosome occupancy for the purpose of comparing elongation rates in light and dark (Fig 6C).
We repeated a subset of experiments with Arabidopsis to determine whether light affects chloroplast translation similarly in a C4 monocot (maize) and a C3 dicot (Arabidopsis) (Fig 8). Similar to maize, the psbA ORF in Arabidopsis showed a roughly 7-fold decrease in ribosome footprint abundance (Ribo-seq) and ribosome density (Ribo-seq/RNA-seq) following one hour in the dark at midday, and restoration to the original level after 15 minutes of reillumination (Fig 8A). Also as in maize, this highly dynamic response was unique to psbA and was not accompanied by a change in ribosome distribution (Fig 8B), implying that light did not have a substantive effect on the dwell time of ribosomes at specific sites. Several other genes showed a similar pattern of ribosome loss and gain but over a much smaller dynamic range (e.g. ndhJ and ndhK). In general, RNA levels were more dynamic in Arabidopsis than in maize over the conditions sampled. In particular, RNAs encoding many ribosomal proteins and NDH subunits increased after one hour in the dark and even more at the end of the night. For most ribosomal genes, this increase in RNA was reflected by an increase in ribosome footprint abundance such that ribosome density (Ribo-seq/RNA-seq) remained fairly constant. However, for most ndh genes, ribosome footprint abundance did not increase proportionally to RNA abundance in the dark, resulting in reduced ribosome density (see ndhG, ndhI, ndhJ, ndhK).
The only striking difference between the data for Arabidopsis and maize involved ribosome density at the end of the night (dark blue bars in Fig 8A). Whereas in maize, ribosome density was maintained or even increased through the night for most chloroplast ORFs (Fig 1C, Fig 2), ribosome density was considerably lower at the end of the night than at midday for many genes in Arabidosis (e.g. psbB, psbH, psbK, ndhC, rps11). Thus, although ribosomes remained associated with all RNAs through the night in both species, there was generally more clearing of ribosomes over the night in Arabidopsis.
To complement the ribosome profiling data, we used in vivo pulse-labeling to assay rates of chloroplast protein synthesis in Arabidopsis seedlings after shifting to the dark and following reillumination at midday (Fig 8C). These experiments were performed in the presence of the cytosolic translation inhibitor cycloheximide to facilitate the detection of chloroplast-encoded proteins. D1 synthesis decreased dramatically after ~40 minutes in the dark and was restored within ~15 minutes of reillumination, correlating with the change in ribosome footprint abundance on psbA RNA. The synthesis of all of the other identifiable proteins (RbcL, PsbB, PsbC, PsaA/B, AtpA, AtpB) also decreased following the dark shift and increased after reillumination, albeit with less dynamic range than D1 synthesis. Decreased synthesis of these proteins in the dark was not reflected by reduced ribosome footprint abundance (Fig 8A), implying that it results, at least in part, from a reduced rate of translation elongation.
Together, these results suggest that many of the themes established with the more comprehensive analyses in maize hold true also in Arabidopsis. That said, the data also suggest some differences that will be interesting to explore in the future. It remains to be determined whether these are meaningful differences or simply reflect differences in physiological status: Arabidopsis plants were grown on synthetic sucrose-containing medium, whereas maize seedlings were grown in soil and had not yet exhausted seed reserves.
Results presented here provide a plastome-wide accounting of ribosome occupancy, distribution, and elongation rate at various time scales following the transfer of maize seedlings with developed chloroplasts from dark to light, and vice versa. Our experiments resolved two layers of regulation: (i) ribosomes are gained and lost on the psbA mRNA shortly after shifting plants to the light or dark, respectively; (ii) the rate of translation elongation decreases globally following a shift to the dark. The superposition of a gene-specific recruitment of ribosomes to psbA on the global increase in translation elongation upon shifting from dark to light programs a massive increase in D1 synthesis. Ribosome occupancy and distribution on all other chloroplast ORFs changes very little, whether comparing short or long times (e.g. 15 minutes versus 7 hours) after the shift from one condition to another. That ribosome occupancy remains constant despite the global change in elongation rate suggests that initiation rates change roughly in concert with elongation rates on all chloroplast ORFs except psbA.
Most prior studies of the effects of light on chloroplast translation involved assays in isolated chloroplasts, and/or illumination of plants that had been germinated and grown in the absence of light and therefore lacked chlorophyll. These approaches provided important insights into the role of chlorophyll in stabilizing nascent chlorophyll binding proteins [30–32] and the impact of ATP and redox poise on psbA translation [26, 27, 33–35]. Prior studies showed further that translation elongation rates on several chloroplast ORFs decrease in the dark [8, 25, 26, 36, 37]. However, a coherent view of the effects of light on chloroplast translation has been lacking because (i) few genes (often only psbA) had been assayed, (ii) psbA-specific effects were rarely resolved from plastome-wide effects, (iii) in organello translation assays generally monitored only the elongation phase and employed ATP and reducing agents, which may have obscured light-induced effects that rely on fluctuations in these signals. By providing a comprehensive view of which chloroplast ORFs respond to light in plants harboring developed chloroplasts, over what time scales, and at what step in translation, our results provide a broad and relatively physiological context in which to interpret this large body of prior work. Our findings validate some prior views, expand on others, and suggest that some commonly held assumptions are incorrect.
It is well known that light stimulates psbA translation in chloroplasts [5, 38, 39], so our finding that psbA ribosome occupancy changes rapidly when green plants are shifted between dark and light may appear to be nothing more than confirmatory. However, prior experiments involved illumination of undeveloped chloroplasts (de-etiolation) and/or failed to address whether light increases the rate of psbA translation initiation over and above any global stimulation. To our knowledge, the only prior studies to examine the effect of light on psbA ribosome association in plants involved the illumination of etiolated barley seedlings [9, 36, 40, 41]; in these experiments the psbA and rbcL mRNAs were recruited to polysomes following illumination, but this was mirrored by a global increase in chloroplast polysome content. A second set of experiments cited as evidence for light-induced psbA translation initiation employed a reporter gene fused to the psbA 5’UTR in tobacco plastids [42, 43]. However, these experiments compared reporter expression in etiolated seedlings to that after 24 hours (or more) of illumination, during which time the activation of photomorphogenetic programs might impact chloroplast gene expression. Furthermore, the possibility that increased reporter expression reflected a global increase in initiation and/or elongation rate was not addressed.
The regulation of psbA translation by light has been intensively studied in the single-celled alga Chlamydomonas reinhardtii. The prevailing view from this body of work is that psbA translation is regulated at different steps in the contexts of PSII biogenesis and repair; the former is believed to involve regulated initiation and the latter regulated elongation [5, 38, 44]. In our experiments, we exposed leaves with assembled photosystems to moderate light intensities. It is clear that the large and specific induction of psbA translation initiation we observed is a homeostatic repair mechanism: the response takes place in the context of developed chloroplasts, it is triggered within minutes of the shift to light, and it results in a substantial over-production of D1 with respect to other PSII subunits. Thus, our results add an important piece to this understanding by showing that light induces a rapid increase in the rate of translation initiation on the psbA RNA over and above any global increase in developed plant chloroplasts, and that this does not require excessive light intensities. The physiological and environmental contexts of light-regulated psbA translation are entirely different in land plants and Chlamydomonas: the former are sessile, multicellular, and land-dwelling whereas the latter are motile, single-celled, and aquatic. It would not be surprising if distinct mechanisms have evolved to maintain PSII homeostasis in these different contexts. That said, psbA mRNA in Chlamyomonas is lost from polysomes after one hour in the dark and regained within ~15 minutes of illumination at a moderate light intensity [45], much as we observed in maize and Arabidopsis. Although those experiments did not address whether this response was specific to psbA, they do suggest that light may regulate psbA translation at the initiation step in the context of PSII repair (and not just de novo biogenesis) in Chlamydomonas, as in plants.
Classic studies demonstrated that the rate of translation elongation on several chloroplast mRNAs in plants changes in response to light. The two primary examples focused on (i) psbA, whose translation elongation rate decreased and increased when isolated chloroplasts were shifted to the dark and light, respectively [24–27, 46], and (ii) rbcL, which maintained polysome association after shifting Amaranth seedlings to the dark despite reduced RbcL synthesis [8]. By using ribosome profiling to follow the rate of ribosome run-off after treating maize seedlings with lincomycin, we were able to extend these conclusions in several ways. First, we showed that the effects on psbA translation elongation previously detected in isolated chloroplasts occurs also in vivo. Second, we showed that the effects on both rbcL and psbA are not gene-specific, but rather reflect a plastome-wide change in translation elongation rate. Our results leave open the possibility that the magnitude of this effect varies among ORFs, but additional experiments would be required to make firm conclusions in this regard.
Our results also clarified the effects of light on ribosome pausing. Pioneering ribosome toe-printing assays had revealed that ribosomes pause at specific sites on the psbA RNA in barley chloroplasts [19], leading to speculation that light may regulate psbA translation by altering pausing at specific sites. A subsequent study showed that toe-print patterns change many hours after illuminating etiolated barley seedlings [20]. By contrast, our data showed that light had no apparent impact on the distribution of ribosomes along the psbA mRNA in mature maize and Arabidopsis chloroplasts over the time scales we examined. Similar results were obtained for all other chloroplast ORFs. Thus, our results provide strong evidence that light does not have any major effects on ribosome pausing on chloroplast mRNAs. Although ribosome dwell time can be influenced by RNA structure, the sequence of the nascent peptide, and other features [21, 47], these behaviors do not appear to be modulated in the context of light-regulated protein synthesis in chloroplasts.
Our finding that psbA was the only ORF to experience a substantial increase in ribosome occupancy following a transfer to light was unexpected because light has been reported to activate translation of several other chloroplast mRNAs via effects on proteins that bind their 5’-UTRs [38]. That ribosome occupancy on maize chloroplast mRNAs other than psbA was maintained almost unchanged even after twelve hours in the dark was also unexpected. Although there is prior evidence that ribosomes remain bound to several chloroplast mRNAs in the dark [8, 36], it has commonly been assumed that many chloroplast ORFs lose ribosome association over the course of the night; this is illustrated by proposals that special mechanisms are required to stabilize chloroplast RNAs at night due to their lack of ribosome association [e.g. 48]. Our results show that this is not the case in maize (excepting psbA), and that it applies to no more than a handful of RNAs in Arabidopsis. Although we did not directly measure initiation rates, our measurements of ribosome occupancy and elongation rates suggest that ribosome occupancy is maintained through a concerted reduction in the translation elongation and initiation rates. The maintenance of ribosome association even after many hours in the dark may have evolved as a means to protect mRNAs from degradation, and to allow a rapid resumption of translation upon exposure to light.
Our results provide a foundation on which to delve into the mechanisms underlying each layer of light regulation, including the nature of the light-induced signal(s), the proteins whose activities are modulated by those signals, and the mechanisms by which these stimulate or inhibit translation in either a global or psbA-specific manner. The light-induced signals that trigger psbA-specific ribosome recruitment, the plastome-wide increase in elongation rate, and the plastome-wide increase in initiation rate may be shared or distinct. Studies involving isolated chloroplasts treated with inhibitors of photosynthetic electron transport [26, 34, 35, 37] have shown that light impacts chloroplast translation via its effect on photosynthesis. Photosynthesis-dependent changes in ATP, the ATP/ADP ratio, NADPH, reduced thioredoxin, reduced plastoquinone, stromal pH, and a trans-thylakoid proton gradient have all been invoked as potential triggers of light-induced effects on chloroplast translation [5]. Experiments that resolve the psbA-specific response from the global responses, and that assess the effects of disrupting specific steps in photosynthesis in intact plants will be required to pinpoint the photosynthesis-dependent signals that regulate chloroplast translation.
Changes in translation can be observed within minutes of shifting plants from light to dark (and vice versa), strongly suggesting that the effects are mediated by post-translational modifications of pre-existing proteins. Candidates for proteins that regulate psbA-specific ribosome recruitment include HCF173 and HCF244, which are required specifically for psbA translation initiation in Arabidopsis [49, 50]. Plastome-wide effects may involve modifications of general elongation and initiation factors or other global translation modulators. For example, PSRP1, the chloroplast ortholog of a bacterial ribosome hibernation factor, has been proposed to repress chloroplast translation in the dark [51]. Future experiments can be directed toward testing these hypotheses, as well as assessing whether changes in light intensity and quality trigger behaviors similar to those described here. Given that the majority of chloroplast-encoded proteins reside in complexes that include nucleus-encoded subunits, our findings also raise the question of whether fluctuations in rates of chloroplast protein synthesis in response to light are reflected by corresponding changes in the synthesis of the nucleus-encoded proteins with which they partner.
Zea mays inbred line B73 was used for all experiments involving maize. Plants were grown in soil under light/dark cycles of 12 h/12 h (for experiments in Fig 1) or 16 h/8 h (for other experiments) at a temperature of 28° and 26°C for the light and dark periods, respectively. Plants were illuminated using a light intensity of 200–300 μmolm-2s-1, which is on the low end of typical “growth light” intensities used for maize and much lower than high-light treatments which typically exceed 1000 μmol m-2s-1. [e.g. 52, 53]. Light-shift experiments were performed on the ninth day after planting, at which time the third leaf was starting to emerge. The apical half of leaves two and three were used for ribosome profiling experiments. This tissue was flash-frozen in liquid N2 immediately after the indicated light treatment and stored at -80°C. Each Ribo-seq replicate pooled tissue from three seedlings. A single seedling was used for each pulse-labeling assay.
Arabidopsis (Arabidopsis thaliana Col-0) seedlings were grown on agar plates (1x Murashige and Skoog Basal Medium (Sigma), 0.3%(w/v) Phytogel, 1% (w/v) sucrose, pH 5.8). Seeds were planted with 1-cm spacing to minimize shading. Plants were grown at 22°C under light/dark cycles of 16 h/8 h for 14 days for ribosome profiling, or for 10 days for pulse labeling. Light intensity was approximately 100 μmol m-2s-1, consistent with typical Arabidopsis growth light intensities and much lower than high-light treatments [54]. For sequencing, aerial parts were harvested from approximately 25 seedlings per replicate, flash-frozen in liquid N2 and stored at -80°C. The treatment and harvest of maize and Arabidopsis in the dark were performed under a dim green light.
We explored multiple methods to introduce lincomycin into maize. Watering intact seedlings with a lincomycin solution resulted in an unacceptably slow response time. Vacuum infiltration of detached leaves had secondary effects on chloroplast ribosome behavior, as revealed by Ribo-seq analysis of a mock-treated sample. Addition of lincomycin to maize protoplasts completely inhibited D1 synthesis within 10–15 min, but protoplast preparation is time consuming and is unsuitable for studies of dark to light transitions. We ultimately chose to introduce lincomycin into maize seedlings through thread wicks, because chloroplast protein synthesis in leaves was inhibited in 10–15 min (as assayed by pulse-labeling) and seedlings remained intact. Cotton threads (DMC crochet thread size 5) soaked in 1 mg/ml lincomycin (Sigma) were sewn four times through the stem of each seedling beneath the first (lowest) leaf. With 2 threads in each sewing, each wick consisted of 8 threads. The threads were cut at ~5 cm and the ends were placed in a 1.5 ml tube containing lincomycin. The apical half of leaves two and three were processed for ribosome profiling.
For maize, the labeling was performed as in [55]. In brief, an emery board was used to create two parallel scratches (1 cm apart) across the upper surface of leaf two, approximately 3-cm from the leaf tip. About 50 μCi of EasyTag Express 35S Protein Labeling Mix (PerkinElmer: 35S-methionine and cysteine; >1000 Ci/mmol, 11 mCi/mL) in 10 μL was added to the wounds. After 15-min of labeling, a 3-cm tissue section spanning the wounds was harvested and frozen in liquid N2. Plants were illuminated with a light intensity of 250 μmol m-2s-1 or maintained in the dark, as indicated. The tissue was homogenized in protein homogenization buffer (10 mM Tris-Cl pH 7.5, 10% glycerol, 5 mM EDTA, 2 mM EGTA, 40 mM β-mercaptoethanol, 2 μg/mL pepstatin, 2 μg/mL leupeptin, 2 mM phenylmethylsulfonyl fluoride). Lysates were fractionated by SDS-PAGE and transferred to nitrocellulose. Radiolabeled proteins were then detected with a Storm phosphorimager. Because the labeling efficiency of the leaf scratch method is variable, we used radiolabeled cytosolic proteins to normalize sample loading. This precluded the use of cycloheximide, which limited the number of chloroplast gene products we could resolve.
For each Arabidopsis sample, the first two rosette leaves from three seedlings were pooled and placed in a clear 24-well plastic plate, with care taken to avoid overlap of leaves. Leaves were pre-incubated for 30 min in 135 μl of labeling buffer containing cycloheximide (20 μg/mL cycloheximide, 1x Murashige and Skoog Basal Medium (Sigma), 1% (w/v) sucrose, pH 5.8). Labeling was then initiated by the addition of 15 μl of EasyTag Express 35S Protein Labeling Mix. Labeling was performed for 20 min under a light intensity of approximately 100 μmol m-2s-1 (or in the dark, as indicated). After labeling, leaves were washed once in the labeling buffer (lacking 35S) and then frozen in liquid N2. The tissue was homogenized as for maize. The membrane fraction was collected by centrifugation at 13,000 x g for 5 min and washed once with the homogenization buffer. Samples were fractionated in SDS-PAGE gels containing 6M urea (11% polyacrylamide), with loading normalized on the basis of equal chlorophyll (assayed as in [56]). Proteins were transferred to nitrocellulose and imaged with a Storm phosphorimager.
With the exception of the first replicate of the experiment in Fig 1, we prepared ribosome footprints and total RNA according to the small-scale protocol described in [57], and purified footprints between approximately 20 and 40 nucleotides. For the first replicate of the experiment in Fig 1, we used the protocol described in [17], and purified footprints between approximately 20 and 35 nucleotides.
Ribo-seq libraries were prepared using the NEXTflex Small RNA Sequencing Kit v2 or v3 (Bioo Scientific) with additional steps described previously [17]. rDNA was depleted after first strand synthesis using biotinylated DNA oligonucleotides together with Dynabeads M-270 Streptavidin or MyOne Streptavidin C1 (ThermoFisher). Replicate 1 of the experiments in Fig 1 used the set of 47 biotinylated DNA oligonucleotides described in [17]. All other experiments used the oligonucleotides described in [57]. For RNA-seq, total RNA samples extracted from aliquots of the same lysates used for Ribo-seq were treated with TURBO DNase (ThermoFisher) followed by treatment with the Ribo-Zero rRNA Removal Kit (Plant Leaf) (Illumina). One hundred ng of the rRNA-depleted RNA was used for library construction using the NEXTflex Rapid Directional qRNA-Seq Kit (Bioo Scientific). The libraries were sequenced on a HiSeq 4000, HiSeq 2500 or NextSeq 500 instrument (Illumina), with read lengths of 50 to 100 nucleotides for Ribo-seq and 100 nucleotides for RNA-seq.
Read processing, alignment and analysis were performed according to the procedures described previously [17]. In brief, adapter sequences were trimmed using cutadapt [58]. Ribo-seq analyses used reads with lengths between 18 and 40 nucleotides. Read alignments were performed using Bowtie 2 with default parameters [59]. Reads were aligned sequentially to the following gene sets, with unaligned reads from each step used as input for the next alignment: (i) chloroplast tRNA and rRNA; (ii) chloroplast genome; (iii) mitochondrial tRNA and rRNA; (iv) mitochondrial genome; (v) nuclear tRNA and rRNA; (vi) nuclear genome. For metagene analysis, all protein coding sequence (CDS) coordinates from all transcript variants were combined to make a union CDS coordinate. Custom Perl scripts extracted mapping information using SAMtools [60]. The distribution of ribosome footprint lengths and RPKM for both the Ribo-seq and RNA-seq data were calculated based only on reads mapping to CDS regions. For chloroplast RPKM calculations, the reads mapping to the first 10 and the last 30 nucleotides of the CDS, which arise from initiating and terminating ribosomes, respectively, were excluded, and we defined the total number of mapped reads as the number mapping to nuclear CDS. For intron-containing chloroplast genes, Ribo-seq RPKM was calculated only from the CDS in the last exon, with the exception of rps12, where exon 2 was used. The rationale for these choices was discussed previously [17].
For the lincomycin assays, the ribosome run-off time course was determined from the RPKM values in the ORF body (from codon 8 to the stop codon). Because maize chloroplast ribosome footprints translating the same codon share a similar 3’-end position regardless of footprint size [17], the normalized abundance of footprint 3’-ends was used to determine sites at which ribosomes accumulate during lincomycin treatment. The 3’-end positions were extracted using SAMtools and normalized to million reads mapped to nuclear protein coding sequences. Sites at which the normalized 3’-end coverage increased more than 5-fold after 30-min of lincomycin treatment in the light (see e.g. S6B Fig) were removed from calculations used for the ribosome run-off analysis shown in Fig 6C and S6C and S6D Fig. Evidence discussed in the text strongly suggests that these result from internal binding of ribosomes in the initiation mode rather than stalled elongating ribosomes. Ribosome build-up sites located outside annotated translation initiation regions that had >50 reads per million after 30 min of lincomycin treatment in the light were reported in Fig 7.
Read mapping statistics and chloroplast RPKM values from all the experiments are provided in S1 Table.
Polysome analyses were performed as described previously [55]. The psbA, atpB/E, rbcL and ndhJ probes used for RNA gel blots correspond to maize chloroplast genome nucleotide positions 295–1074, 54590–55790, 57036–57607 and 50535–51014, respectively.
Illumina read sequences were deposited at the NCBI Sequence Read Archive with accession number SRP133508. Alignments of reads to the maize chloroplast genome used Genbank accession X86563. B73 RefGen_v3 assembly (maizegdb.org) was used for other genomes. The gene set from maize genome annotation 6a (phytozome.jgi.doe.gov) was reduced to the gene set annotated in 5b+ (60,211 transcripts) (gramene.org). For Arabidopsis, we used TAIR10 genome and annotation (arabidopsis.org).
|
10.1371/journal.pcbi.1002629 | The Role of Local Backrub Motions in Evolved and Designed Mutations | Amino acid substitutions in protein structures often require subtle backbone adjustments that are difficult to model in atomic detail. An improved ability to predict realistic backbone changes in response to engineered mutations would be of great utility for the blossoming field of rational protein design. One model that has recently grown in acceptance is the backrub motion, a low-energy dipeptide rotation with single-peptide counter-rotations, that is coupled to dynamic two-state sidechain rotamer jumps, as evidenced by alternate conformations in very high-resolution crystal structures. It has been speculated that backrubs may facilitate sequence changes equally well as rotamer changes. However, backrub-induced shifts and experimental uncertainty are of similar magnitude for backbone atoms in even high-resolution structures, so comparison of wildtype-vs.-mutant crystal structure pairs is not sufficient to directly link backrubs to mutations. In this study, we use two alternative approaches that bypass this limitation. First, we use a quality-filtered structure database to aggregate many examples for precisely defined motifs with single amino acid differences, and find that the effectively amplified backbone differences closely resemble backrubs. Second, we directly apply a provably-accurate, backrub-enabled protein design algorithm to idealized versions of these motifs, and discover that the lowest-energy computed models match the average-coordinate experimental structures. These results support the hypothesis that backrubs participate in natural protein evolution and validate their continued use for design of synthetic proteins.
| Protein design has the potential to generate useful molecules for medicine and chemistry, including sensors, drugs, and catalysts for arbitrary reactions. When protein design is carried out starting from an experimentally determined structure, as is often the case, one important aspect to consider is backbone flexibility, because in response to a mutation the backbone often must shift slightly to reconcile the new sidechain with its environment. In principle, one may model the backbone in many ways, but not all are physically realistic or experimentally validated. Here we study the "backrub" motion, which has been previously documented in atomic detail, but only for sidechain movements within single structures. By a twopronged approach involving both structural bioinformatics and computation with a principled design algorithm, we demonstrate that backrubs are sufficient to explain the backbone differences between mutation-related sets of very precisely defined motifs from the protein structure database. Our findings illustrate that backrubs are useful for describing evolutionary sequence change and, by extension, suggest that they are also appropriate for rational protein design calculations.
| Proteins routinely incorporate amino acid changes over evolutionary time by adapting their conformation to the new sidechain. However, it remains a difficult task to predict such a conformational response, especially when subtle backbone adjustments are involved. This issue is of central importance to the burgeoning field of computational protein design, which has recently enjoyed a string of exciting developments [1]–[4].
A number of descriptions of backbone motion have been implemented for the purposes of protein design in the past, each with its own set of advantages and disadvantages. Anticorrelated “crankshaft” adjustments of the ψ(i−1) and φ(i) torsions [5] are evident from order parameters derived from molecular dynamics (MD) simulations, but unrealistically distort the ends of the peptide if employed in isolation. Helical parameters [6] and normal mode analysis [7] enable efficient exploration of conformational space near the starting model, but are only useful for a small subset of protein architectures: respectively, coiled-coils and structures for which a small number of motional modes dominates conformational diversity. Peptide fragments [8] implicitly reflect local protein energetics because they are extracted from experimental structures, but can be computationally inefficient because most random fragment insertion attempts are incompatible with a given local structural context and will therefore be rejected. This being the case, it may be prudent to let nature inform our notion of backbone motion by using a move set based on empirical observations, which may encode aspects of protein energetics and sidechain/backbone coupling that are difficult to handle explicitly.
One such model is the backrub (Figure 1), a highly localized backbone motion tightly coupled to sidechain rotamer jumps, initially characterized by examining alternate conformations in ultra-high-resolution crystal structures [9]. A simple geometrical model of the backrub consists of a small (<15°) rotation of a dipeptide about the axis between the first and third Cα atoms. Resulting strain in the N-Cα-C bond angle τ of all three residues may be partially alleviated and backbone H-bonding maintained with small counter-rotations of the two individual peptides. Note that this Cα formulation is a simplified but very close approximation of the real molecular mechanism, which probably involves a computationally unwieldy set of small shifts in 6–10 backbone torsion angles, as discussed in [9]. Backrubs were seen for 3% of the total residues in that previous study, and for 2/3 of the alternate conformations with a change in Cβ position – far exceeding the next most common shifts, which are either peptide flips or local shear in a turn of helix.
Several studies have successfully used the backrub approach to expand the search space of protein design efforts and improve agreement between computed sidechain dynamics and nuclear magnetic resonance (NMR) measurements [10]–[14]. Recent work has shown that computational design of backbone structures generated by backrub sampling can recapitulate much of the sequence diversity found in the natural ubiquitin protein subfamily [15] and by phage display experiments [16]. However, the backrub has only been empirically demonstrated to accompany dynamic rotamer changes, not actual changes in amino acid identity. Importantly, no direct experimental evidence has been presented to support the assumption implicit in these studies that a dynamic, low-energy motion on the pico-to-nanosecond timescale is relevant on an evolutionary timescale. The contribution of this current study is to address in atomic detail the specific mechanisms by which backrubs accommodate amino acid changes during processes like subfamily evolution. We use a data set of 5200 high-resolution, high-quality crystal structures to examine differences in local backbone conformation between well-defined motif populations related by a single amino acid difference, and find that the backrub motion explains the majority of the mainchain movement. Furthermore, we demonstrate that a provably-accurate flexible-backbone design algorithm allowing backrubs at those positions, in conjunction with a common molecular mechanics force field, accurately recapitulates such mutation-coupled backbone changes. These findings validate inclusion of the empirically observed backrub motion as part of the repertoire of “moves” for protein design and other modeling efforts.
The N-cap or C-cap position of a helix is defined as the residue half-in and half-out of the helix: the peptide on one side of the cap makes standard helical backbone interactions, while the peptide on the other side has quite non-helical position and interactions [17]. α-helix N-cap residues can make several types of interactions that stabilize or specify the structural transition from loop into α-helix, the most common and dominant of which is a sidechain-mainchain hydrogen-bond to the i+3 amide [17]–[19]. The N-cap H-bond enhances protein stability by compensating for the loss of a mainchain H-bond at the helix start relative to the middle of a helix. Note that the sidechain cannot reach this H-bonding position if the residue has helical φ,ψ, so this interaction also specifies the exact helix start position and the direction from which the backbone can enter [20]. Asn, Asp, Ser, and Thr are especially favored at N-caps because their sidechains have the proper chemical character and shape to mimic the helical backbone interactions (which Gln and Glu are too long to do).
Notably, Asn/Asp sidechains are longer than Ser/Thr sidechains by one covalent bond, yet their H-bond distances (N-cap sidechain O to i+3 amide H) are only slightly shorter (2.01±0.18 vs. 2.17±0.18 Å) based on a survey of all N-caps with i+3 H-bonds in the Top5200 database (described below and in Methods). This means the backbone must slightly adjust to maintain similar H-bond geometry in both cases.
With this motivation, we wished to confirm the appropriateness of the backrub model for this case of mutational rather than rotamer change. However, backbone coordinate shifts due to backrubs are very small – on the order of the coordinate differences between crystal structures of the same protein [21], [22], thus obscuring differences between genuine shifts and experimental noise. The initial description of the backrub bypassed this problem by comparing alternate conformations within single structures [9]. Our approach here, in contrast, was to use the collective weight of many examples to ensure that observed local conformational differences were in fact genuine.
Aromatic residues often pair with glycine in antiparallel β-sheet by adopting rotamers with χ1≈+60°, which places the aromatic ring directly over a Gly on the adjacent strand across a narrow pair of backbone H-bonds [29]. Aromatic-glycine pairings in antiparallel β-sheet have been demonstrated to yield a synergistic thermodynamic benefit [30]. If the opposite residue is changed to anything other than Gly, a sidechain including at least a Cβ atom is now present, which would sterically clash with the aromatic in its original conformation. However, the “plus χ1” aromatic rotamer will still be compatible with some rotamers of the opposite sidechain, provided that the aromatic may shift slightly to re-optimize packing of its ring against the opposite residue's Cβ hydrogens. Here we investigate whether backrubs enable this relaxation by excursions in both directions from a “neutral” β-sheet conformation. The leverage provided by such backbone motions could lean the aromatic residue forward/backward to maintain close inter-strand contact when the identity of the opposite residue is changed to/from Gly.
It is known that backrubs relate conformations that interchange dynamically [9]. In this study we further show that, at least for certain specific motifs, backrubs relate conformations that “interchange” based on point mutations. For the β aromatic motif, local restraints on steric packing influence the aromatic residue's backbone indirectly via the other altered sidechain. This is in contrast to helix N-caps where the sequence change and the backrub occur at the same residue, as seen previously for rotamer changes [9]. Taken together, these findings support the intriguing idea that backrubs may “foster” mutations, easing them into the structure and promoting their survival into future generations. In this paradigm, backrubs enable individual mutations that provide the raw material for natural selection.
The two specific motifs analyzed here represent only about 0.5% of the protein residues in our Top5200 data set, and thus in one sense the scope of this study is relatively narrow. However, a tight focus was necessary to substantiate the idea of mutation-coupled backrubs with sufficient certainty, due to the coordinate error problem in the alternative approach of comparing individual wildtype and mutant crystal structures directly. These two cases were chosen as common, well-defined motifs where the primary interaction environment of the changing sidechain is provided by local secondary structure and is therefore consistent across hundreds of examples. For the general case of an individual mutation, the potential interaction environment is also the same before and after; however, it is seldom simple enough to be closely repeated in numerous proteins. Furthermore, as shown in previous work at ultra-high resolution [9], 2/3 of alternate conformations that move Cβ demonstrate backrubs between rotamers of the same amino acid. All in all, therefore, it is reasonable to assume that the general prevalence of backrub accommodation at sites of mutation is significantly higher than the “lower bound” provided in this study.
The backbone shift considered on its own is continuous and low-energy, without a barrier, while the two-state behavior is contributed by the sidechain switch between rotamers, between H-bond partners, or between amino acids. In the dynamics case of jumps between distinct sidechain rotamers or H-bonds [9], both conformations are quite favorable, but backbone and sidechain must change together. In the evolutionary case, such as the single amino acid changes between stable motifs illustrated in this paper, the two amino acid types cannot coexist in the same molecule. The sidechain-backbone coupling shifts the energy landscape for the backbone [31], stabilizing a different choice within a shallow energy well. From a modeling perspective, examination of existing fast-timescale structural dynamism may illuminate other possibilities for mutations on an evolutionary timescale [32].
Mutation-coupled backrubs are small local changes, which presumably mediate neutral drift much more often than they aid large-scale structural rearrangements or changes in function. However, the accumulation of changes via neutral drift over time may in fact enable future large-scale changes by subtly altering the native state energy landscape such that eventually a tipping point is reached. Recent analysis of the evolution of an ancient protein confirms that some function-altering mutations required structural pre-stabilization by earlier “permissive” mutations [33]; backrubs may facilitate such preemptive sequence changes by shifting the backbone such that the functionally neutral amino acids can fit. Backrub-related sequence changes could also sometimes enable functional change based on a purely local adaptation when they occur in active sites, either directly or by first enhancing functional promiscuity.
Note that we do not directly address true evolutionary relationships between proteins in this study. Rather, we substantiate the idea that backrubs enable single amino acid changes at specific motifs, which could aid actual evolution within a protein family [15].
It is only natural to segue from the role of backrubs in protein evolution to their utility for protein design – essentially a computational analog of molecular evolution. Our results indicate that, despite the relative simplicity of their functional form, molecular-mechanics-based force fields like Amber plus EEF1 that are commonly used for protein design can in fact accurately recapitulate empirically-observed backbone conformation for multiple specific structural motifs, given the chance to access them via a backrub. (Note that the cases presented here were dominated by single interactions such as H-bonds or steric packing; a higher-cost energy function might be needed to maintain similar accuracy if different interactions are competing and need to be compared quantitatively.)
Thus, predicting the conformational consequences of a sequence change in computational protein design is in large part a search problem: if the appropriate regions of protein conformational space are searched efficiently, in many cases low-cost energy functions can do the rest. Unfortunately, that space is vast indeed even for a single sequence, as we know from Levinthal's famous thought experiment [34]. The additional consideration of combinatorial mutations (even when conformational changes are restricted to simple sidechain rotamer alternatives) creates a space that is even more difficult to search, as shown by the proof that protein design is NP-hard [35]. Of course, backbone flexibility further enlarges the search space.
However, flexible-backbone design algorithms like BRDEE are excellent candidates for this task in many cases because (1) they are based on empirically demonstrated types of flexibility and (2) they come with mathematical guarantees of their accuracy with respect to the input parameters. Other algorithms that search over amino acid and rotamer identities, then minimize over backrub degrees of freedom post facto are not guaranteed to identify the global minimum energy conformation (GMEC) given the input model (starting structure, rotamer library, energy function). An advantage of BRDEE is that it incorporates backrub minimization awareness directly into the amino acid and rotamer comparison stages of dead-end elimination, and thus is guaranteed to identify the GMEC given the input model. Because BRDEE avoids becoming trapped in local minima, it effectively decouples the often intertwined issues of conformational search and scoring. Therefore, as a result of using BRDEE, this paper gives the limit of how well any algorithm can perform given our input model. In the future we plan to implement additional empirically-validated small backbone motions, such as peptide flips and tripeptide shears [9], [14], to improve coverage of conformational space.
Overall, we have demonstrated that the backrub, a model of local backbone motion previously only documented for dynamic rotamer changes, also applies to local sequence changes. This finding is an important direct validation for the application of the backrub to the study of natural protein evolution and to continuing efforts in computational protein design.
To identify numerous examples of the desired motifs, we used a “Top5200” database of high-quality protein structures. The rapid growth of the Protein Data Bank (PDB) [36] in recent years enabled the creation of a high-quality database with an order of magnitude increase in size relative to the previously described Top500 [23] while maintaining similar standards of resolution and structure quality. However, due to sheer logistics it also necessitated a more automated selection protocol.
We included at most one protein chain per PDB 70% sequence-similarity cluster as of April 5, 2007. We chose the representative for each such cluster as the chain with the best average of resolution and MolProbity score [37] where resolution is <2 Å. MolProbity score is an estimate of the resolution at which a structure's steric clashes, rotamer quality, and Ramachandran quality would be average; thus the average of resolution and MolProbity score is a combined experimental and statistical indicator of structural quality [37]. The homology filter prevents redundancy and thus over-representation of certain motifs or substructures.
To calculate the MolProbity score for each chain, first hydrogens were added with the program Reduce [38]. The -flip flag was used in order to allow Asn/Gln/His flips throughout the structure, including at interfaces where multimer partners may participate in hydrogen-bonding networks. All protein chains with at least 37 residues were then extracted, along with any “het” atoms or waters with the same chain identifier, and MolProbity score was calculated for each chain.
Two “post-processing” steps were required. First, we removed four chains whose PDB structures had been obsoleted and replaced them with updated structures where possible (1sheA→2pk8A, 1wt4A→2v1tA, 2eubA→2pl1A, 2f4dA→no replacement). Finally, we removed two chains with incomplete or unclear PDB files (1c53A had only Cα atoms, 3ctsA had only “UNK” unknown residue types). The resulting 5199 protein chains make up about a million residues.
For all BRDEE calculations, we used the same input parameters as described previously [10] for energy function, rotamer library, τ filter, etc. Briefly, the energy function consists of the Amber electrostatic and van der Waals terms [24] plus the EEF1 pairwise solvation energy term [25]. The τ filter prevents large strain at the only bond angle allowed to change in the calculations [10].
The KiNG graphics program [39] was used both to study the superposition results and to produce the figures. Dataset S3 provides raw PDB coordinate files for N-cap and aromatic crystal structure examples, average crystal structures, and BRDEE lowest-energy models.
|
10.1371/journal.pcbi.1006348 | A mechanism for bistability in glycosylation | Glycosyltransferases are a class of enzymes that catalyse the posttranslational modification of proteins to produce a large number of glycoconjugate acceptors from a limited number of nucleotide-sugar donors. The products of one glycosyltransferase can be the substrates of several other enzymes, causing a combinatorial explosion in the number of possible glycan products. The kinetic behaviour of systems where multiple acceptor substrates compete for a single enzyme is presented, and the case in which high concentrations of an acceptor substrate are inhibitory as a result of abortive complex formation, is shown to result in non-Michaelian kinetics that can lead to bistability in an open system. A kinetic mechanism is proposed that is consistent with the available experimental evidence and provides a possible explanation for conflicting observations on the β-1,4-galactosyltransferases. Abrupt switching between steady states in networks of glycosyltransferase-catalysed reactions may account for the observed changes in glycosyl-epitopes in cancer cells.
| While enzymes tend to have a narrow substrate specificity, there are a number of enzymes that are promiscuous, acting on a wide range of substrates. In this article we derive expressions for general multi-substrate competitive inhibition for the class of transferases, with particular emphasis on glycosylation. By extending the enzyme reaction mechanism to include inhibition by high substrate concentrations, we show that switching behaviour (bistability) is possible within a thermodynamically open systems of glycosylation enzymes. The biological implication of this finding is that small changes to a predictor variable may induce abrupt changes in the secreted products.
| With the ready availability both of computing power and software tools for numerical simulation, the mathematical modelling of metabolic systems has become a core component of cell biology. Models of classical metabolic pathways, such as glycolysis [1–3], the citric-acid cycle [4], the urea cycle [5] and biosynthetic pathways such as N-linked and O-linked glycosylation [6, 7], have been developed as a way to understand how such processes are regulated. Online repositories of such models, such as the BioModels database [8], allow many of these models to be examined without the need for programming ability on the part of the user. Software such as E-Cell [9] have enabled more complex models to be constructed at the cellular or organelle level.
This paper examines a particular class of metabolic model, in which one or more enzymes can act on multiple substrates. To this class belong the cytochrome P450 enzymes that are involved in detoxifying multiple xenobiotics [10], ribonuclease P [11] and also the enzymes of N-linked glycosylation [12–15]. Such enzymes recognise multiple substrates, and the products of the reactions can themselves become substrates, thus introducing a form of competitive inhibition with catalysis. It is known that, in the case of two substrates acted upon by the same enzyme the Michaelis constant of the kinetic rate law will be modified to include the effects of competing substrates upon one another [16, 17]. In the first part of this paper, a general form of the Michaelis-Menten equation for n competing substrates is derived, and extended to an ordered-sequential mechanism involving a donor molecule held in common by all reactions. In the second part, we model galactosyltransferase acting on an initial acceptor glycoprotein to form two products, each of which are substrates for the same enzyme. Here we propose a possible mechanism for such behaviour and apply it to the glycosyltransferase model, demonstrating the switching between stable steady states over a range of parameter values.
Consider the case of a general two-substrate enzyme mechanism, in which a donor molecule, Ax, transfers the x moiety to an acceptor, B,
Ax + B → A + Bx ,
a reaction type that is common to the transferases. We consider the situation in which there are n acceptor substrates, B1 … Bn. For random-order binding of donor and acceptor (Fig 1A), an expression for the initial rate of appearance of the jth acceptor product, Bxj, is
v j = V j [ Ax ] [ B j ] K s Ax K m B j ( 1 + s B j ′ ) + K m B j [ Ax ] ( 1 + s B j ) + K m j Ax [ B j ] + [ Ax ] [ B j ] (1)
where Vj = kj[E0] is the maximal velocity obtained at saturating levels of Ax and Bj, K m j Ax = K s Ax K m B j / K s B j, s B j = ∑ i ≠ j [ B i ] / K m B i and s B j ′ = ∑ i ≠ j [ B i ] / K s B i. The derivation of this equation under rapid-equilibrium conditions is given in the S1 Appendix. In this model the K s Ax and K s B i are, respectively, the individual dissociation constants of Ax and Bi from the E⋅Ax and E⋅Bi enzyme-substrate complexes, while the Michaelis constants of these species, K m j Ax and K m B i are the corresponding dissociation constants of the E⋅Ax⋅Bi complex. The s B j and s B j ′ terms are sums of dimensionless acceptor substrate concentrations representing the degree to which the enzyme is competitively inhibited by substrates other than Bj itself. In the absence of substrate competition, s B j = 0, and Eq (1) reduces to the standard form of a bisubstrate enzyme mechanism. In the limit, as [Ax] → ∞, (1) becomes
v j = V j [ B j ] K m B j ( 1 + s B j ) + [ B j ] , (2)
an equation that is similar in form to that obtained in other studies [14, 18, 19].
Although the s B j symbolism is a convenience in order to show which terms of the rate law are affected by competitor concentrations, a representation that is more useful in computer simulations is the sum of concentrations of all its substrates, each weighted by its K m B j or K s B j:
A E = ∑ i = 1 n [ B i ] K m B i , A E ′ = ∑ i = 1 n [ B i ] K s B i . (3)
Substituting into Eq (1),
v j = V j [ Ax ] [ B j ] K s Ax K m B j ( 1 + A E ′ ) + K m B j [ Ax ] ( 1 + A E ) . (4)
Whereas a rapid-equilibrium random-order mechanism is a feature of polypeptide N-acetylgalactosaminyltransferase [20], sulfotransferases [21], fucosyltransferases [22] and sialyltransferases [23], with other glycosyltransferases, such as those of the N-acetylglucosaminyltransferase and galactosyltransferase families, the enzyme must bind the donor first, before catalysis can occur [24]. Under quasi-steady-state conditions (Fig 1B), the rate law for the compulsory order binding is (Eq S3 in S1 Appendix):
v j = V j [ Ax ] [ B j ] K s Ax K m B j ( 1 + s B j ) + K m B j [ Ax ] + K m j Ax [ B j ] + [ Ax ] [ B j ] (5)
In such a case, the inhibitory effect of multi-substrate competition will lessen as the concentration of the donor is increased towards saturating levels.
Thus far, the possibility of abortive (dead-end) ternary enzyme complexes has not been considered, which in random-order mechanisms are likely to occur [25]. Experimental evidence for the existence such complexes can be the appearance of inhibition at high substrate concentrations; in the case of glycosyltransferases, the inhibition is usually that of the acceptor [26–29], but can also be that of the donor [30]. If we consider only the acceptor, an examination of the mechanism (Fig 1A) reveals that four additional binding events can occur, with the E⋅Ax, E⋅Bj, E⋅A and E⋅Bxj complexes. We consider binding of Bj to the second of these complexes, E⋅Bj, to provide a possible explanation for substrate inhibition with increasing acceptor concentration. Not only will Bj bind, but so will any competitive acceptor-substrate Bi, i = 1, …, n.
The oligosaccharides attached to glycoproteins (glycans) can be multivalent, meaning that the same acceptor has more than one recognition domain. By way of illustration, the enzyme β-N-acetylglucosaminylglycopeptide β-1,4-galactosyltransferase (GalT; EC 2.4.1.38), catalyses the transfer of d-galactose (Gal) residue to a terminal N-acetylglucosaminyl (GlcNAc) residue on a glycoprotein, glycopeptide or polysaccharide, with the general reaction:
UDP-α -D-Gal + β -D-GlcNAc-R → UDP + Gal-β 1 , 4 -D-β-D-GlcNAc-R
A theoretical system, similar to that studied experimentally by Paquêt and co-workers [31], is shown in Fig 2, in which galactose is incorporated into glycopeptide in four steps, starting with the initial acceptor B1, to form the final product with two terminal galactoses (B4). Hence, the products B2 and B3 are also substrates of the enzyme, since both contain a terminal GlcNAc on which it can act. All three substrates are therefore competitive inhibitors in the earlier sense, and can form a ternary complex with E⋅Bj, the free terminal β-d-galactose in the acceptor competing with the donor, UDP-Gal [33].
Before continuing, we make the parenthetic observation that reaction networks such as those in Fig 2 follow a binomial distribution pattern in the number of acceptors at each step. If the initial substrate has m sites on which an enzyme can act, then the m immediate acceptor-products of that substrate will each have m − 1 available sites. There will be a reaction hierarchy based on the combinatorial filling of available sites until the final product is reached at m = 0, with the number of substrates at the kth step following the familiar mCk pattern,
m C k = m ! k ! ( m - k ) ! .
After k steps, a glycan substrate originally with m sites will have m − k sites remaining. The resulting network of all possible reactions, for a single acceptor possessing m sites at which an enzyme can act, will have N(m) nodes and E(m) edges, given by N ( m ) = ∑ k = 0 m m C k and E ( m ) = ∑ k = 0 m m C k ( m - k ). Every node, whether substrate or product, will have degree m, with the in-degree of a node at the kth step being k and its out-degree being m − k. The number of possible pathways from initial substrate to final product will be
P ( m ) = ∑ k = 0 m m C k k ( m - k ) .
Every glycan will have up to m of each type of dissociation constant, for the enzyme of which it is a substrate, product or inhibitor.
Extending the derivation of the rapid-equilibrium random equation in the S1 Appendix, an additional term will be required in the denominator to represent the abortive complex(es). Since there are n substrates, there will be n2 ways in which to form E⋅Bk⋅Bi. A double summation over the indices i and k will be required, giving the additional term
∑ i = 1 n ∑ k = 1 n [ E · B k · B i ] = [ E · Ax ] K s Ax [ Ax ] ∑ i = 1 n ∑ k = 1 n [ B k ] K I B k [ B i ] K s B i
where K I B k is the dissociation constant of the kth acceptor from complex E⋅Bk⋅Bi.
The rate of appearance of the jth product will then be
v j = V j [ Ax ] [ B j ] K s Ax K m B j ( 1 + s B j ′ + s I ) + K m B j [ Ax ] ( 1 + s B j ) + K m j Ax [ B j ] + [ Ax ] [ B j ] (6)
with
s I = ∑ i = 1 n ∑ k = 1 n [ B k ] K I B k [ B i ] K s B i = ∑ k = 1 n [ B k ] K I B k ∑ i = 1 n [ B i ] K s B i .
When n = 1, this reduces to
v = V max [ Ax ] [ B ] K s Ax K m B + K m B [ Ax ] + K m Ax [ B ] + K m Ax K I B [ B ] 2 + [ Ax ] [ B ] (7)
The equation for the compulsory order mechanism will be identical, and the more computationally efficient representation, equivalent to Eq (4), is
v j = V j [ Ax ] [ B j ] K s Ax K m B j ( 1 + A E ′ + s I ) + K m B j [ Ax ] ( 1 + A E ) (8)
with two summation terms, A E = ∑ i = 1 n [ B i ] / K m B i and A E ′ = ∑ i = 1 n [ B i ] / K s B i.
A general scheme for the formation of ternary enzyme-acceptor complexes is given in Fig 3A. This scheme does dual service, in illustrating both the formation of n2 inhibitory complexes in an n-substrate environment, but also the two catalytic mechanisms involving compulsory-order and random-order binding of substrates, which in the latter case only occurs when j = k, and for substrate inhibition at high concentrations, when j = k = i. The scheme illustrates two aspects of multi-substrate competition: productive, in which catalysis occurs, and non-productive, where there is inhibition as a result of abortive complex formation at higher acceptor concentrations. In the productive case, the n acceptors compete with each other for the E⋅Ax complex, in either random-order or compulsory-order binding mechanisms. In the non-productive case, higher acceptor concentrations compete with the donor for binding to the free enzyme, as well as with each other, for an enzyme-acceptor complex, resulting in non-productive multi-substrate inhibition in compulsory-order mechanisms. In the case of a random-order mechanism, the acceptor may bind to either the free enzyme or to the E⋅Ax complex in pathways leading to the productive ternary (E⋅Ax⋅Bj) complex. Therefore high substrate inhibition may result from the mis-oriented binding of acceptor to the free enzyme or binding of a second B to the E⋅B resulting in an abortive ternary complex. The binding site at which competition occurs may differ, depending on the enzyme mechanism involved. Fig 3B displays three curves of v vs [acceptor], showing the relief of substrate inhibition that occurs as the donor concentration is increased, and in Fig 3C, the velocity-substrate surface defined by two K I B values, for n = 2.
The situation is more complicated when multiple binding sites exist on each molecule of acceptor. According to Fig 2, B1 is a substrate, but B4 is not, while B2 and B3 can bind as substrate inhibitors, though B1 cannot because it does not have a terminal GlcNAc. B4 acts as a competitive (product) inhibitor of UDP-Gal, with two possible inhibition constants, K I , 1 B 4 and K I , 2 B 4. The effective value of n is the number of edges, E(m), in the network of a substrate with m recognition sites, as defined in the previous section, which gives 16 summands in sI. For the network in Fig 2, therefore,
s I = [ B 4 ] K I , 1 B 4 [ B 1 ] K s , 1 B 1 + [ B 4 ] K I , 2 B 4 [ B 1 ] K s , 1 B 1 + [ B 3 ] K I , 1 B 3 [ B 1 ] K s , 1 B 1 + [ B 2 ] K I , 1 B 2 [ B 1 ] K s , 1 B 1 + [ B 4 ] K I , 1 B 4 [ B 1 ] K s , 2 B 1 + [ B 4 ] K I , 2 B 4 [ B 1 ] K s , 2 B 1 + [ B 3 ] K I , 1 B 3 [ B 1 ] K s , 2 B 1 + [ B 2 ] K I , 1 B 2 [ B 1 ] K s , 2 B 1 + [ B 4 ] K I , 1 B 4 [ B 2 ] K s , 1 B 2 + [ B 4 ] K I , 2 B 4 [ B 2 ] K s , 1 B 2 + [ B 3 ] K I , 1 B 3 [ B 2 ] K s , 1 B 2 + [ B 2 ] K I , 1 B 2 [ B 2 ] K s , 1 B 2 + [ B 4 ] K I , 1 B 4 [ B 3 ] K s , 1 B 3 + [ B 4 ] K I , 2 B 4 [ B 3 ] K s , 1 B 3 + [ B 3 ] K I , 1 B 3 [ B 3 ] K s , 1 B 3 + [ B 2 ] K I , 1 B 2 [ B 3 ] K s , 1 B 3 , (9)
in which K X , i B k denotes the ith dissociation constant of the kth acceptor, where X is either s (dissociation from E⋅Bi) or I (dissociation from an abortive ternary complex).
It has been observed that bistability can arise when an enzyme is inhibited by one of its substrates in an open system [34], in which substrate enters at a zero-order rate, and exits at a rate that is first-order in the concentration of that substrate. If the substrate can diffuse into the reaction medium according to
v diff = K ( [ B ] 0 - [ B ] ) , (10)
where [B]0 is the concentration of exogenous substrate, then multiple steady-state solutions for the concentration of substrate can coexist for venz = vdiff. This is illustrated in Fig 3D, where the number of points of intersection of the line (10) with the curve described by Eq (7) will depend on the values of [B]0 and the diffusion constant, K.
Bistability can be demonstrated through numerical simulation of the one-dimensional ODE system:
d b d t = K ( b 0 - b ) - V max a b K a K b + K b a + K a b + K a K s b 2 + a b , (11)
where a and b are the concentrations of the donor and acceptor, respectively. It is assumed that the donor concentration is constant, while the external concentration of b is chosen as the parameter to vary. The numerical continuation software AUTO, part of the ODE solver XPPAUT [35], was used to calculate the steady-state level of b for increasing b0. For the parameters a = 0.6, K = 0.075, Kb = 0.1, Ks = 0.05, Vmax = 1 and Ka = 0.6, bistability is obtained for 1.518665 < b0 < 2.325853 (Fig 4).
Within this range two stable steady states of acceptor concentration can coexist, as shown by upper and lower branches in b–b0 space. This can be confirmed by solving dvdiff/db = −K for b, using the parameters of Fig 3D, and computing the ordinate-axis intercept for vdiff at these two concentrations, which will be points of tangency of the two lines described by Eq (10) with the velocity–substrate curve. The values of b, computed in Mathematica (version 11.0.1; Wolfram Research, Inc.), are b* = {0.10755, 0.693546}. Substituting into (10), we evaluate b* + vdiff(b*)/K = b0, obtaining the corresponding solutions b0 = {1.51866, 2.32585}.
The reaction scheme shown in Fig 2 is modelled with five differential equations,
d b 1 d t = K ( b 0 - b 1 ) - ( v 1 + v 2 ) (12) d b 2 d t = K ( b 0 - b 2 ) + v 1 - v 3 (13) d b 3 d t = K ( b 0 - b 3 ) + v 2 - v 4 (14) d b 4 d t = K ( b 0 - b 4 ) + v 3 + v 4 (15) d a d t = K ( a 0 - a ) - ( v 1 + v 2 + v 3 + v 4 ) (16)
where the bi represent the acceptor concentrations [Bi], i = 1 … 3, a is the concentration of UDP-Gal, and the enzyme velocities v1 … v4 are described by Eq (8). As before, the model assumes free diffusion of substrates into the medium in which enzyme is active [36]. There will be additional terms in s B j and sI, since there will be two sets of constants for the initial oligosaccharide substrate B1, one set for each recognition site. The total enzymic rate of removal of B1, for saturating levels of Ax, will be
v 1 + v 2 = V 1 b 1 K m 1 + b 1 + V 2 b 1 K m 2 + b 1 . (17)
Assuming that the maximal velocities of each of v1 and v2 are the same, we can solve for substrate concentration at half-maximal velocity, to obtain apparent Km as the geometric mean of the individual Michaelis constants, K m app = K m 1 K m 2. Under the same assumption, for a substrate with m recognition domains, the apparent Km will be the solution to
1 = ∑ i = 1 m K m app K m j + K m app . (18)
Numerical simulation of the model also displayed bistability (Fig 5). Using a two-parameter continuation, the region of a0–b0 space under which bistability exists was determined (Fig 5B). The values of the external concentrations at the point of the cusp were found to be (b0, a0) = (0.07094, 0.5959). Bistability was also obtained by varying the diffusion constant, K (Fig 5C); a two-parameter continuation in a0–K space revealed a closed region of bistability (Fig 5D).
In this article a general equation for multi-substrate inhibition is derived, from which are deduced a number of properties of a system of reactions involving rate laws of this kind. Enzymes following a quasi-steady-state compulsory mechanism described by Eq (5) will not show this response with donor concentrations at saturating levels. While the nature of such competitive inhibition had been examined by Schnell and Mendoza [18], and our initial result was presented, without proof, by Umaña and Bailey [12], to our knowledge, this work is the first to present a derivation of a bisubstrate reaction equation with multi-substrate competition, with an extension to include substrate inhibition.
We have extended the treatment of multi-substrate enzymes obeying rapid-equilibrium random-order kinetics to systems exhibiting inhibition at high substrate concentrations. Special note was made of the additional complication of oligosaccharide acceptors, which will have multiple dissociation constants when multivalency is present, and it was shown that, in the bivalent case, an overall Michaelis constant can be predicted from the geometric mean of the individual Km values. The work of Degn [34] was applied to the transferase enzymes acting in a system held thermodynamically far from equilibrium, and it was shown that two stable solutions can exist over a range of external substrate concentrations.
Bistability was also shown to be possible for a system of reactions catalysed by the enzyme β-N-acetylglucosaminylglycopeptide β-1,4-galactosyltransferase (GalT) evincing both multi-substrate competition and multivalency. Our model provides a possible explanation for both the compulsory-order catalytic mechanism of this enzyme, reported by Qasba et al. [24], in which the donor binds before the acceptor, and the inhibition observed by Freilich et al. at high acceptor concentrations [26]. A derivation of the random-order two-substrate mechanism, under quasi-steady-state assumptions, will lead to a 2:2 rational function that is second order with respect to the concentration of acceptor in both the numerator and denominator [37]. Although this non-Michaelian function would give rise to a velocity-substrate curve similar to that observed with substrate inhibition, it conflicts with the available evidence for the catalytic mechanism. If, as we propose here, the acceptor binds as a substrate analog of the donor, at the donor site, followed by a further acceptor-binding step to form a dead-end ternary complex, the apparent paradox is resolved.
As a biological phenomenon, bistability has previously been identified in apoptosis [38], cancer [39], disease progression [40], cell cycling [41], cell motility [42] and differentiation [43]. It has also been reported in an open reconstituted enzyme system containing phosphofructokinase [44]. Multistability is well known in the context of ultrasensitivity [45], and similar phenomena, such as cooperativity and allostery, where enzymes possess switch-like behaviour [46]. In enzyme-kinetic models, a general condition for multistability is that the rate law be a non-monotonic function of the reactant concentrations. Hence, the competitive inhibition introduced by multi-substrate competition is not a necessary, or sufficient, condition for switching behaviour; rather, it is the formation of ternary enzyme-substrate complexes that can lead to non-monotonicity in the enzyme rate law. The first derivative of such a function, possessing at least one maximum or minimum, must undergo a change of sign, as the substrate or effector concentration is varied. Since the property is shown to be possible for a single enzyme, its origin can be distinguished from that based on network topology [47] or feedback regulation [48].
Our main result is consistent with the prediction by Neelamegham and Liu [49] that bistability could arise under circumstances where Michaelis-Menten kinetics, with nonlinearities caused by large numbers of possible substrates and products, were combined with feedback/feedforward regulation. We have considered only initial rate kinetics in this study, ignoring the effects of product concentration in the derivations in the S1 Appendix, although product inhibition effects will also play a role, as can be seen at the early stages of the proof. In neglecting the product concentrations, we have constructed the system in such a way that the primary cause of the bistability is more readily apparent. The models of Shen and Larter [50], who studied the membrane-bound enzyme acetylcholinesterase, not only displayed bistability, but also oscillatory behaviour when either autocatalysis or product inhibition were included. Higher order dynamic behaviour might therefore arise if our model was expanded to incorporate the effects of product concentrations.
The conventional approach to modelling metabolism has involved the construction of systems of ordinary differential equations using kinetic rate laws appropriate to the enzymes and transporters involved, as has been the case for most models of glycosylation to date [12, 14, 51], and in the present work. Such models assume an underlying deterministic process and a detailed knowledge of the parameters, which may not be available. Another approach is to model the transitions between the reactants in a network by a Markov chain, an application of which to glycosylation has recently appeared [52]. It is known that bistability can arise within noise-driven biochemical systems operating at the level of cellular volume [53], even where it is not predicted in the deterministic limit. Multistability as a general principle, therefore, and outside of the specific application to glycosylation, can exist within either modelling framework. As the volume size of the system decreases, the rate of switching between biologically realisable steady states increases exponentially [54], which has implications for several of the phenomena cited above, such as cellular differentiation and cancer. The transitions between steady states, perturbed by stochastic fluctuations, may additionally require that the system be close to the boundaries of the basins of attraction [55].
These results demonstrate that the complex interplay of enzyme and substrate can give rise to nonlinear behaviour in systems of reactions held far from thermodynamic equilibrium. The significance of the present study is that small changes in one condition, such as the amount of available sugar-nucleotide donor [56], might incur large and abrupt changes in the amount of product formed. Since GalT action influences the number of sites available for sialylation, such changes should have important implications for cancer progression and metastasis, which have been shown to be related to these processes [57], and for biotechnology, such as in the production of therapeutic antibodies [58], which can be influenced through control of metabolic flux [59]. More generally, the occurrence of bistability in metabolism could provide the basis for cellular long-term memory [60]. The commonly occurring pattern of substrate inhibition in transferases should complement the already known behaviours of models based on sigmoidal functions. For instance, it is known that different glycosylation enzymes associate, and co-locate with the Golgi, according to the ‘kin recognition’ model [61], and may therefore display cooperativity. Whether a combination of cooperativity and substrate inhibition could lead to higher order dynamic behaviour, such as oscillations in acceptor concentration, is an open question that deserves further study.
|
10.1371/journal.pntd.0003614 | Ecological and Sociodemographic Determinants of House Infestation by Triatoma infestans in Indigenous Communities of the Argentine Chaco | The Gran Chaco ecoregion, a hotspot for Chagas and other neglected tropical diseases, is home to >20 indigenous peoples. Our objective was to identify the main ecological and sociodemographic determinants of house infestation and abundance of Triatoma infestans in traditional Qom populations including a Creole minority in Pampa del Indio, northeastern Argentina.
A cross-sectional survey determined house infestation by timed-manual searches with a dislodging aerosol in 386 inhabited houses and administered questionnaires on selected variables before full-coverage insecticide spraying and annual vector surveillance. We fitted generalized linear models to two global models of domestic infestation and bug abundance, and estimated coefficients via multimodel inference with model averaging.
Most Qom households were larger and lived in small-sized, recently-built, precarious houses with fewer peridomestic structures, and fewer livestock and poultry than Creoles’. Qom households had lower educational level and unexpectedly high residential mobility. House infestation (31.9%) was much lower than expected from lack of recent insecticide spraying campaigns and was spatially aggregated. Nearly half of the infested houses examined had infected vectors. Qom households had higher prevalence of domestic infestation (29.2%) than Creoles’ (10.0%), although there is large uncertainty around the adjusted OR. Factors with high relative importance for domestic infestation and/or bug abundance were refuge availability, distance to the nearest infested house, domestic insecticide use, indoor presence of poultry, residential overcrowding, and household educational level.
Our study highlights the importance of sociodemographic determinants of domestic infestation such as overcrowding, education and proximity to the nearest infested house, and corroborates the role of refuge availability, domestic use of insecticides and household size. These factors may be used for designing improved interventions for sustainable disease control and risk stratification. Housing instability, household mobility and migration patterns are key to understanding the process of house (re)infestation in the Gran Chaco.
| Indigenous peoples are especially affected by Chagas and other neglected tropical diseases. One of the most numerous ethnic groups in the Gran Chaco region of South America is the Qom (Toba) people. The living conditions of Qom households most likely differ substantially from those of Creoles, and their association with house infestation by triatomine bugs has not been investigated. This is important because the major vector species have adapted to thrive in human sleeping quarters, and in addition to housing construction patterns, other ecological and sociodemographic factors may affect house infestation. We found that Qom households had much higher domestic infestation than Creole ones, in conjunction with more precarious housing, fewer poultry and livestock. The unexpectedly high local residential mobility of Qom households combined with the large fraction of recently-built houses (derived from a rapidly increasing population with a very young age structure during recent decades) may explain why domestic infestation was much lower than expected from the lack of recent insecticide spraying campaigns. Domestic infestation and bug abundance increased with overcrowding and refuge availability, and decreased with household education levels and insecticide use. These results are useful for designing improved interventions and household risk stratification.
| The strong association between neglected tropical diseases (NTDs), poverty and particular combinations of ecological, social, political and economic determinants explains the occurrence of global hotspots of NTDs [1]. One of such hotspots occurs in the Gran Chaco ecoregion in South America, where the prevalence rates of geohelminthic infections and Chagas disease are very high [1]. Chagas disease, caused by Trypanosoma cruzi, is considered the main regional vector-borne disease in terms of disease burden and affects 8–10 million people in Latin America [2]. Triatoma infestans, the main vector in the Southern Cone countries and southern Peru, has been the target of an insecticide-based regional elimination program that interrupted the transmission of human T. cruzi infection by T. infestans in various countries [2–4]. Progress in the Gran Chaco lagged behind and vector-mediated transmission of T. cruzi still occurs albeit at lower incidence levels than 20 years ago [5–8].
The Gran Chaco is home to more than 20 ethnic groups [9]. Indigenous populations usually are among the most marginalized groups, with more precarious health and living conditions than other peoples [10–12]. Indigenous communities of the Gran Chaco showed high seroprevalence of human T. cruzi infection [13–18]. One of the most numerous ethnic groups in this region is the Qom (Toba) people [19]. Qom households were exposed to a greater risk of T. cruzi infection than Creole households in a well-defined rural section of Pampa del Indio (Argentine Chaco) mainly inhabited by Creoles (denominated Area I), but there were large heterogeneities between and within ethnic groups [20,21]. Further studies on the ecological, biological and social (eco-bio-social) determinants of vector-borne diseases are needed [22], more so in the case of vulnerable indigenous populations affected by Chagas disease and other NTDs.
The main identified determinants of house infestation with the major domestic vectors of T. cruzi (T. infestans, Rhodnius prolixus, Panstrongylus megistus, and Triatoma dimidiata) include housing construction characteristics that create refuges for the bugs to hide in (e.g., cracks in walls, thatched roofs, precarious peridomestic structures); the presence and number of human and domestic animal hosts (dogs, chickens) that serve as bloodmeal sources, and little or no domestic application of insecticides by house residents [20,23–32]. These factors are the expression of various underlying processes that ultimately create conditions that facilitate house infestation and T. cruzi transmission [33]. A full understanding of complex systems [34] involving infectious diseases requires more integrative approaches such as the ecosystem approach to human health (ecohealth) [35], which gives proper attention to eco-bio-social factors and their eventual interactions. However, very few studies have explicitly addressed these factors simultaneously in relation to Chagas disease [31,36,37]. This limited knowledge curtails our ability to design and implement innovative vector and disease control strategies adapted to resource-constrained settings. The current study therefore addressed traditional ecological determinants and selected sociodemographic factors related to poverty and ethnicity.
As part of a longitudinal study on the eco-epidemiology and control of Chagas disease in northeastern Argentina, we expanded the scope and geographic scale of our previous studies [20,21,37] conducted in Area I of Pampa del Indio to focus on Qom communities living in ancestral territories which also included a Creole minority (denominated Area III). The living conditions of Qom households most likely differed substantially from those of Creoles, and their association with house infestation has not been investigated at a sufficiently large spatial scale. The objective of the current study was to identify the main ecological and sociodemographic determinants of domestic infestation and abundance of T. infestans (two surrogate indices for transmission risk) in Area III, where Qom communities predominated, using generalized linear models in a multimodel inference frame with model averaging. In addition to the above-mentioned factors known to be closely associated with house infestation in multiple settings, we examined the effects of distance to the nearest infested house, residential overcrowding, household education level, wealth indicators, and preventive practices. The first two factors were predicted to exert positive effects on domestic infestation and bug abundance whereas the remaining factors were expected to exert negative effects. We also re-examined whether ethnic background modified both response variables when other relevant risk factors were accounted for. Our study highlights the relevance of various ecological and sociodemographic factors whose effects have not been investigated simultaneously, and provides guidance on improved control interventions specifically adapted to the Gran Chaco.
Field work was conducted in a rural section (95 km2) of Pampa del Indio municipality (25° 55’S 56° 58’W), Chaco province, Argentina (Fig. 1). The municipality was inhabited by approximately 22,000 people by late 2013, and 45% of residents belonged to the Qom ethnic group according to local municipal authorities. Official decennial census records in 2001 and 2010 indicated that the population of Pampa del Indio municipality increased remarkably from 11,558 to about 18,000 people, respectively (annual population growth rate, 4.9%). The climate, landscape and demographic features of a contiguous section of the municipality inhabited mainly by Creole households were described elsewhere [20,21]. The last insecticide spraying campaign conducted in the municipality occurred in 1997–1998 according to the Chagas disease control program from Chaco province. Selection of the study area took into account the lack of recent history of community-wide insecticide spraying; preliminary evidence of house infestation ranging from 30 to 40%; the predominance of indigenous households; and the presence of at least 350 adjacent households in order to achieve a sufficiently large study base for statistical inference.
A household is defined as all the people who occupy a housing unit including the related and nonrelated family members [38]. A house compound was defined as the set of domicile (i.e., an independent structure used as human sleeping quarters, S1 Fig.), patio and other structures included within the peridomestic area (kitchens, storerooms, latrines, corrals, chicken coops and chicken nests (“nidero”), ovens, trees where chickens roosted, others) as illustrated elsewhere [20]. House compounds sometimes had more than one domicile used as sleeping quarters by related family (S1 Fig.).
Before initiating field operations local key actors were interviewed to gather background data that may allow a better assessment of the preintervention situation; discuss the initial and long-term goals of the research program (see below); and assist the interpretation of the study outcomes. Local key actors included the mayor, health and education authorities and other personnel, rural health-care workers and school teachers, representatives of third-sector organizations, and community leaders. The stated long-term goals of the research program were to interrupt the human transmission of T. cruzi through intensified vector control, human diagnosis and treatment, and to promote long-term sustainability of disease control efforts through local empowerment.
A cross-sectional survey aimed at enumerating all house compounds in the area and assessing house infestation was conducted in October 14–31, 2008. The study area included seven villages with 407 inhabited houses, 19 abandoned dwellings and 17 public buildings (4 primary schools, 1 health-care post, 6 churches and 6 community centers) (Fig. 1). One member of the research team explained to each householder the aims of the survey and requested permission to access their premises and identify the house with a numbered aluminum plate. The location of each house was georeferenced with a GPS receiver (Garmin Legend). Householders were asked for the presence of triatomines within their premises after showing them dry specimens of T. infestans, Triatoma sordida and other Reduviidae to prevent confusion with other insects; from these reports we derived the index “householders’ notification of house infestation”. All households were provided with a labeled self-sealing plastic bag to contain any triatomine they sighted, and instructed on how to manipulate the bugs safely. This additional source of bugs was denominated “householders’ bug collections”. Householders’ bug collections were only considered if the date and collection site were reported to us. The study protocol was approved by the Dr. Carlos A. Barclay Independent Ethical Committee for Clinical Research, Buenos Aires, Argentina.
A total of 386 inhabited houses (94.9%) were included in the current study of triatomine infestation; 21 houses closed during the survey were not searched for bugs. In all of the 386 study houses the following methods were performed to assess bug infestation: i) inspection by timed-manual searches; ii) collection of bugs that were spotted during insecticide spraying operations; and iii) promotion of householders’ collection of any triatomine they sighted (as explained above). Multiple methods were used as a cross-check of the outcome of timed-manual searches. All domestic and peridomestic sites of the study houses were searched for triatomine bugs (timed-manual collections) by four teams including one supervisor and two or three skilled bug collectors who used 0.2% tetramethrin (Espacial, Argentina) to dislodge the insects as described [20]. Each domicile and peridomestic site was searched by one person for 15 min. Immediately after the vector survey, vector control personnel sprayed every house with suspension concentrate deltamethrin (K-Othrin, Bayer) or beta-cypermethrin (Sipertrin, Chemotecnica) using standard doses (25 and 50 mg/m2, respectively) and routine procedures [39]. Bugs sighted during insecticide spraying operations were also collected.
The collected bugs were stored in plastic bags labeled with the house number and specific bug collection site and were transported to the field laboratory where they were identified taxonomically and counted according to species, stage or sex. Two to six weeks after bug collection, feces of all the third-instar nymphs and older stages that were alive were microscopically examined for infection with T. cruzi at 400× as described [21]. The bugs examined for infection were collected from 72.8% of the infested houses.
This survey was conducted in parallel to the vector survey in October 2008. An adult household member fluent in Spanish was asked for information on the following items: full name of householder (i.e., head of household) [40]; the number of resident people by age class (0–5, 6–14, and 15 or more years of age); the number of domestic animals of each type (dogs, cats, chickens, other poultry, goats, pigs, cows, and equines) and their resting places; use of domestic insecticides (type, frequency, purpose); and date of the last insecticide spraying of house premises conducted by vector control personnel or the local hospital or any other third party using manual compression sprayers. Because the study area encompassed traditional Qom communities, assignment of a household to ethnic group was based on whether they spoke Qom language (Qomlaqtaq); participated in traditional Qom organizations; and took into account their physical features. When in doubt, assignments to ethnic group were subsequently checked with local Qom health-care personnel and were corroborated in all cases. Households with a mixed ethnic background were considered to be Qom because they resided in ancestral indigenous territories and fulfilled the above mentioned attributes.
A sketch map of the spatial location of all structures in each house compound was performed, and each structure was given a unique code according to its use. We recorded the building materials used in roofs and walls, presence of wall plaster, condition of wall surface, and plaster material. The availability of refuges for bugs was determined visually by a skilled member of the research team and scored in one of five levels ranging from absence to very abundant refuges [20]; only the three top categories were recorded in domiciles.
As our knowledge of the study area increased during the vector surveillance phase, additional sociodemographic variables potentially associated with house infestation were taken into consideration and recorded mostly in November 2012: educational level attained by each household member (number of schooling years completed); land ownership (no ownership; individual: the householder owned the land they inhabited; familial: a relative owned the land; communal: the community owned the land which therefore could not be sold); agricultural activities (present and past); monthly public welfare support; household electricity and time since first connection; age of house (years since construction); size of each domicile’s area; source of drinking water; presence of window screens (wire mesh); use of bed nets; and participation in a local social organization.
The data collected in 2012 were back-corrected to extant conditions in 2008 (e.g., access to electricity, age of house, agricultural activities). Overall changes in several respects (e.g., drinking water source, domicile’s area, participation in social organizations) were negligible during the four-year period. For some of the back-corrected variables it was possible to assess the validity of the reports. The comparison of domestic area and age of house recorded both in 2009 and 2012 showed minor differences. Land tenure, access to electricity and householders’ reports of time since last insecticide spraying were checked with other local sources of information and whether they were spatially clustered. Comparison between the list of houses sprayed with insecticides in 2006 (identified by the name of the head of household) and the date of the last insecticide spraying each individual household reported to us in 2008 showed either large or perfect agreement in two communities (75% and 100%) and a very low degree of agreement in another community (8%).
The recorded data were used to compute household-level surrogate indices for wealth, educational level and overcrowding. The goat-equivalent index represents a small stock unit to quantify the total number of livestock (cows, pigs, goats) and poultry owned by the household in terms of goat biomass. To calculate this index the average weight of each type of animal was considered (cow, 453 kg; pig, 159 kg; goat, 49 kg; chicken, 2.5 kg) [41]. The household educational level was defined as the mean number of schooling years attained by household members aged 15 years old or more [42]. The overcrowding index was defined as the number of human occupants per sleeping quarter; the presence of 3 or more occupants per room is taken as critical overcrowding [43]. Housing quality (a three-level categorical variable) was represented by the combination of mud walls (versus brick-cement walls) and tarred-cardboard sheets on the roof (versus corrugated metal-sheets); no house had brick-cement walls and tarred-cardboard sheets.
As part of annual vector surveillance after community-wide insecticide spraying in October 2008, all extant households in the study area were re-surveyed in August 2009, April 2010 and November 2012, whereas a sample of 86 houses was re-surveyed in December 2011. For each house we recorded its current and previous existence; fate (destruction, movement and construction); destination of moving households and underlying reasons (whenever possible), among other variables. The sociodemographic information was collected at every new house as in the baseline survey.
House infestation data only included inhabited houses because no local public building or abandoned house was found to be infested. Similarly, latrines and trees used by chickens were not infested by T. infestans and therefore were not included in the number of peridomestic sites per household. The prevalence of house (or site-specific) infestation by T. infestans was calculated based on the finding of at least one live bug (except eggs) by any of the bug collection methods used (i.e., timed-manual searches, during insecticide spraying operations, and householders’ bug collections) relative to the total number of houses (or sites for each ecotope) inspected. The abundance of triatomine bugs was calculated as the number of live bugs collected per 15 min-person among houses positive by timed-manual searches. If a house compound had more than one domicile, the average domestic bug abundance was calculated as the total number of live triatomines collected per 15 min-person across domiciles divided by the number of domiciles inspected. A matrix of distances to the nearest infested house was calculated using qGIS [44]. Agresti–Coull binomial 95% confidence intervals (95% CIs) were used for infestation prevalence [45].
Householders’ notification of the domestic presence of T. infestans and timed searches of domestic infestation were compared using the kappa index implemented in Stata 12 [46]. Kappa index values greater than 0.6 may be considered substantial to perfect agreement and values less than 0.4 represent a poor agreement beyond chance.
Risk factor analyses of the presence and relative abundance of T. infestans were restricted to human sleeping quarters because peridomestic infestations were relatively few. Owing to the occurrence of house compounds with more than one domicile (including related family) and that several variables were measured at the household level, in these cases data for all domiciles were pooled to obtain a single observation per compound. Availability of refuges for bugs and age of house were averaged over domiciles within a house compound, and the total domestic area was the sum of each domicile’s area. The number of domestic hosts (dogs or cats and poultry, mostly chickens) used in the analyses (not in the census) only included animals reported to rest or nest inside domiciles. Bivariate logistic and negative binomial regressions on each explanatory variable were performed with domestic infestation and bug abundance as response variables, respectively. Relative bug abundance (RA), labeled in Stata output as 'incidence-rate ratios’, and their CIs were calculated from the estimated coefficients (b) of the negative binomial regression as eb.
The association between selected explanatory variables and both response variables were tested through multiple logistic and negative binomial regressions, respectively. The global models included 10–12 explanatory variables with complete data selected a priori based on background evidence (e.g., [20,25,30]) and additional hypothesis on the predicted effects of selected sociodemographic determinants as mentioned above. Some variables measured in 2012 (i.e., age of house, electricity, time since last insecticide spraying) had a large number of missing data and therefore were not included in these analyses. We also compared the fit of the negative binomial models for bug abundance with those returned by mixture and two-part models for zero-inflated distributions, and found strong evidence of the superiority of the negative binomial regression model (S2 Text).
Two global models were analyzed. The first model included 10 explanatory variables (from 386 households) which described building characteristics (housing quality, refuge availability), domestic host availability (number of persons, number of dogs or cats and presence of poultry indoors), socioeconomic features (ethnicity, goat-equivalent index), household preventive practices (reported insecticide use), peridomestic infestation by T. infestans, and distance to the nearest infested house. The second model included 12 variables (i.e., the 10 variables mentioned before, residential overcrowding and household educational level) recorded at 274 households. Some continuous variables were rescaled in order to give more meaning to the unit of increment of risk estimates: distance to the nearest infested house (one unit every 50 m), household educational level (every 6 years) and the goat-equivalent index (every 10 goats). For comparative purposes we also analyzed the second data set after removing overcrowding and educational level data. On a post hoc basis we investigated the effects of the interactions between ethnicity and every other factor in the global models on both response variables, which proved not to be significant. These terms were added one by one to each global model and tested separately to avoid convergence problems.
Potential multicollinearity among explanatory variables was evaluated through the variance inflation factor (VIF) and condition numbers as implemented in Stata 12. The condition numbers were less than 10 and VIF < 2 for all explanatory variables, indicating that the significant correlation found between some pairs of variables (ethnic group with housing quality, refuge availability, goat-equivalents, which had correlation coefficients ranging from 0.35 to 0.4) would not cause serious multicollinearity.
We used an information theoretic approach and Akaike’s information criterion (AIC) to identify the best-fitting models describing variations in domestic infestation and abundance of T. infestans, given the data collected. Multimodel inference was especially conceived to account for model selection uncertainty; it allows a quantitative ranking of the models and identification of the set of models having best support given the data [47,48]. Because the ratio between the number of parameters and the number of observations (i.e., houses) was less than 40, we used the AIC corrected for small sample size (AICc). Akaike differences (ΔAICc) were calculated for each model as ΔAICc = AICc—AICmin; models with ΔAICc > 2 were considered to have less support than the best model (AICmin), given the data and models analyzed. Several models had substantial support; therefore, we performed multimodel inference through model averaging. The Akaike weight (wi) of a model represents the support or probability of being the “best model”. The relative importance (RI) of each variable is defined as the sum of Akaike weights in each model in which the variable is present; RI takes values from 0 to 1. The overall fit of the logistic models was assessed by the Hosmer-Lemeshow test using the model-averaged coefficients and pooling the data in 10 equal-sized groups. Odds Ratios (ORs) and their 95% confidence intervals were calculated from model-averaged coefficients. Unconditional standard errors were calculated according to equation 4 in [49] with the default option (revised.var = TRUE). The area under the receiver operating curve (ROC) was also calculated; a value of 1 indicates a perfect fit. Sensitivity and specificity were assessed using the observed infestation prevalence of each data set as the cutoff values. The analyses and calculations were performed in R (version 2.15.1) [50]. Package MuMIn (version 1.9.5) was used for multimodel averaging; ResourceSelection (0.2–2) for performing the Hosmer-Lemeshow test; and ROCR (version 1.0–5) for calculating sensitivity, specificity and the area under the ROC curve.
The spatial distribution of domestic infestation was assessed through global and local point pattern analyses (PPA) [51]. The former estimates the spatial aggregation of the outcome event across the entire study area whereas the latter detects the location of aggregated events. The spatial distribution of houses was examined to determine whether the potential aggregation of house infestation was influenced by a non-random dispersion of house locations.
The global spatial analysis of domestic infestation was performed in Programita using the weighted K-function [52] and random labeling as the null hypothesis (i.e., to assess the spatial distribution of infested houses given the fixed spatial distribution of all houses). The maximum distance considered was 2,000 m (i.e., one-third of the smallest dimension of the area) [51], and the cell size was 40 m. A total of 999 Monte Carlo simulations was performed and the 95% confidence envelope was calculated with the 25th upper and lower simulations.
Local spatial aggregation of infestation was tested through the Getis statistic (G*) [53] implemented in PPA [54]. This analysis distinguishes between positive and negative aggregation of events (i.e., infested houses); parameter settings were the same as for the global analysis.
The house-to-house census enumerated a total of 2,389 inhabitants in 386 inhabited houses as of October 2008, and 2,356 persons in 445 inhabited households in November 2012. The population included 18.0% up to five years of age; 27.7% between 6 and 14 years; and 54.3% with 15 or more years of age at baseline, and displayed a nearly indistinguishable age distribution in 2012. The mean age was 20.3 yr whereas the country-wide average was 34.4 yr. The number of men per 100 women was 109.2 whereas the average for Argentina was 95.8 as of 2010 [55].
A summary of the housing and sociodemographic characteristics of the study population by ethnic group is shown in Table 1. The detailed frequency distribution of study variables appears in Tables 2, 3 and S1. Qom households comprised 89.6% of the inhabited houses. Unlike Creoles, most Qom households lived in houses with mud walls and a tarred-cardboard roof, in small-sized (< 30m2), recently-built domiciles with high refuge availability, < 2 peridomestic structures, and with little access to electricity (Table 1). Qom households were larger, more often experienced critical overcrowding (S1 Text), and had lower household educational level than Creoles. The average goat-equivalent index of Creoles (median, 68.9; first-third quartiles, 5.7–168.6) was 69 times larger than that of Qom households (1.0, 0.1–7.7).
Most Creole households applied insecticides in domestic premises (90.0%) and had window screens (59.5%), unlike Qom households (Table 1, S1 Text). Creole households applied high-concentration pyrethroid or carbamate insecticides (27.5%) much more frequently than Qom households (5.8%). Householders’ reports indicated that 68.0% of houses had never been sprayed with insecticides by vector control personnel whereas 21.5% had been sprayed two years before (Table 3). Local health personnel reportedly sprayed with insecticides 36 and 49 houses, mainly from Cuarta Legua villages in 2000 and 2006, respectively.
A key feature of the study population was the very frequent mobility of households to a new house (i.e., housing instability). Of all the inhabited houses enumerated in 2008, 20.2% (78) were demolished or abandoned by 2012, whereas 142 new houses were built over the four-year period, of which only 52 (36.7%) were present in 2012. Most of the new houses (90.1%) and the demolished or abandoned ones (98.7%) belonged to Qom households. Among the latter, movers were relatively disadvantaged compared to nonmovers. On average, movers had a lower goat-equivalent index (median, 0.5 versus 1.9); smaller domiciles (24 versus 36 m2); more recently-built houses (69.2% versus 45.6%); smaller household size (5.2 versus 6.8); and fewer peridomestic sites (1.3 versus 2.1).
Triatoma infestans was found by timed-manual searches in 108 (28.0%) of the 386 inhabited houses and in 6.9% of the 1,744 sites inspected. The median relative abundance was 3 bugs (first-third quartiles, 1–11) per unit of catch effort. Fifth-instar nymphs (24%), males (20%) and females (16%) were the stages most frequently captured. When the finding of bugs by any collection method was considered, the prevalence of house infestation slightly rose to 31.9% (123 of 386), and was 27.2% (105 of 386) in domestic sites and 7.8% (30 of 386) in peridomiciles. A total of 2,362 T. infestans was caught. Triatoma sordida was found in 4.5% (17 of 386) of houses exclusively in peridomiciles.
The contribution of each collection method to detection of domestic infestation is shown in S3 Table. Although the majority of domestic infestations was detected by timed-manual searches, bugs collected by householders (almost exclusively in domestic areas) and during insecticide spraying operations contributed to additional detection of 19 infested houses that timed searches had missed. Detection of domestic infestations by timed-manual searches and householders’ notifications were in poor agreement (kappa index = 0.3).
The ecotopes most frequently infested at site level (as determined by any bug collection method) were domiciles (23.1%), storerooms (14.0%), kitchens (6.3%), chicken nests (6.0%), and chicken coops (4.6%) (Fig. 2). The median abundance of T. infestans per unit of catch effort was higher in kitchens, storerooms and chicken nests, but did not differ significantly among ecotopes by negative binomial regression (P > 0.1 in all cases).
The overall prevalence of bug infection with T. cruzi was 25.0% among 719 live bugs examined, and ranged from 23.9% (150/628) in domiciles to 33.0% (30/91) in peridomiciles. Infected bugs were collected in domiciles of 45.2% (28/62) of the houses with bugs examined for infection, and in peridomiciles of 44.4% (8/18) of them.
Although house infestation occurred across the study area, some communities showed larger domestic infestation than others (range, 12.2–50.8%) (Fig. 3, S1 Table). Domestic infestation was significantly aggregated at a global scale at distances ranging from 600 to 2,000 m (S2 Fig.); this means that infested houses were clustered, and on average, for every infested house there was a higher probability of finding another infested house within 600–2,000 m than expected by chance. Local spatial analyses of bug abundance identified clusters of houses located within 40–600 m in some communities (Cuarta Legua, Pampa Chica). The apparent cold spot at the NE angle (Pampa Grande village) was not statistically significant (P > 0.05).
Wall and roof materials were significantly associated with domestic infestation (Table 2). The prevalence of domestic infestation increased significantly with increasing refuge availability levels and numbers of human residents, and declined steadily with increasing age of house and domestic area. Infested domiciles had a significantly smaller area (37.8 ± 27.2 m2, n = 84) than non-infested ones (50.6 ± 39.9 m2, n = 200; Mann-Whitney test, P < 0.001). Domestic infestation was higher in houses with at least one infested peridomestic site and fewer peridomestic structures (S1 Table). Domestic bug abundance only was significantly associated with refuge availability, domestic area and the number of dogs.
Qom households had a nearly threefold domestic infestation (29.2%) than Creoles’ (10.0%), whereas domestic bug abundance was similar between ethnic groups (Table 3). Domestic infestation increased steadily with increasing residential overcrowding from 0% up to 42.9%, and decreased with increasing household educational level from 30.1% to 0%. Households reporting insecticide use had a significantly lower infestation (21.2%) than those that did not (32.7%). Bug abundance was also significantly associated with residential overcrowding, household educational level, the goat-equivalent index, number of peridomestic sites, land ownership, and access to electricity (Table 3, S1 Table). The signs of the individual effects were the same as for domestic infestation. Households with no window screens had increased domestic infestation and bug abundance, whereas the use of bed nets was inversely associated (S1 Table). No significant association was found between domestic infestation and time since last insecticide spraying (Table 3).
Using the first global model including 386 houses, we identified 11 and 6 models with considerable support (ΔAICc < 2) for domestic infestation and bug abundance, respectively. Refuge availability (RI = 1.00), distance to the nearest infested house (RI = 1.00–0.83) and insecticide use (RI = 0.75–0.69) were the most important factors (Table 4). Refuge availability exhibited a strong positive association whereas insecticide use and distance to the nearest infested house had a negative one. The presence of poultry indoors (RI = 0.75) was only moderately and directly associated with domestic bug abundance. The number of people (RI = 0.73) had moderate importance and marginally positive effects on infestation only, with CIs including the null value. Other factors presented lower RI for both response variables. Ethnicity showed low RI and widely variable CIs. The average logistic model for infestation (Hosmer-Lemeshow test, χ2 = 8.03; d.f. = 8; P = 0.43) and the area under the ROC curve (0.73) indicated a good fit. The model had moderate specificity (0.62) and sensitivity (0.71).
For the second global model including 274 houses, refuge availability (RI = 1.00), overcrowding (RI > 0.98) and distance to the nearest infested house (RI = 0.78–0.88) were the most important factors and showed strong to moderate effects, whereas household educational level had moderate importance (RI = 0.74–0.68) and rather small negative effects on domestic infestation only (Table 4). The average logistic model for infestation (Hosmer-Lemeshow test χ2 = 3.66; d.f. = 8; P = 0.89) and the area under the ROC curve (0.79) indicated a good fit. This model showed higher specificity (0.71) and similar sensitivity (0.72). Removing overcrowding and household educational level data yielded results that were qualitatively similar to those in the first global model (not shown).
Additional analyses including only Qom households identified the same set of factors with high and moderate RI in both global models (not shown).
Our study identified important risk factors for domestic infestation and bug abundance in a rural area inhabited mainly by Qom populations through a multimodel inference framework. Some of these factors were novel and pertain to the sociodemographic domain, such as distance to the nearest infested house, residential overcrowding and household educational level, and accorded with predictions. We also corroborated the high RI of refuge availability and lack of domestic use of insecticides on both response variables, and the moderate RI of the number of people on domestic infestation recorded in the neighboring Area I [20]. These findings are in general qualitative agreement with the outcomes of studies on various species of Triatominae regardless of large differences between rural settings, ethnic composition, analytic methods and variables examined [20,25,27–32,37]. Surprisingly, the baseline prevalence of domestic infestation was much lower than expected on the basis of the near absence of insecticide spraying campaigns over the previous decade. This unexpected finding called for a post hoc explanation.
Domestic infestation and bug abundance increased with proximity to the nearest infested house although the effect was rather small. These observations are in agreement with predictions from patterns showing spatial aggregation of house infestation before and after insecticide spraying elsewhere in Argentina [20,56,57] but not in peri-urban habitats of Arequipa, Peru [58]. Subsequent studies in Arequipa, however, revealed that most of the infestations detected after the insecticide spraying campaign occurred in untreated houses which later served as sources of the insects that recolonized their neighbors [59]. The short distances between houses in several of our study villages clearly facilitated the invasion of T. infestans by flight or walking dispersal [60–62]. Both domestic infestation and bug abundance were spatially aggregated, with hotspots including heavily infested domiciles within 40–600 m; the upper bound is well within the observed flight dispersal range of T. infestans in a mark-recapture experiment [63]. Estimates based on the duration of sustained tethered flights suggest the flight range of T. infestans might exceed 2,400 m [64].
Residential overcrowding, a surrogate index for socioeconomic status and health conditions [65], was closely and positively associated with domestic infestation and bug abundance. This index approximates human density in sleeping quarters, and incorporates both household size and number of rooms in the domicile. Overcrowding is also expected to facilitate host finding and the human-feeding success of T. infestans, and most likely underlies the positive relation between the number of human occupants and domestic infestation by different species of triatomine bugs [24,28,58,66]. Likewise, the household seropositivity to T. cruzi increased steadily and significantly with decreasing size of domestic area –a putative index of stable settlement and well-being in Creole communities of the dry Chaco [67]. In our study, the average prevalence of domestic infestation also showed a strong positive trend with decreasing domestic area and age of house, and increasing number of people, although all CIs included the null values.
Household educational level averaged less than 6 years of schooling and was moderately important for domestic infestation, with negative effects whose size was rather small. Household education is considered a generic measure of household socioeconomic status, which is among the main social determinants of health inequalities [65]. The causal pathway linking increasing household educational levels to decreasing domestic infestation may be associated with access to information and receptivity to health education messages, which may translate into healthier practices [65]. For example, lower education levels correlated with increasing severity of Chagas disease cardiomyopathy, among other factors [68]. Even though education does not exclusively occur through formal instruction and duration of schooling does not specify its quality, educational level is a simple metric for comparison between households. Another measure of household socioeconomic status and wealth, the goat-equivalent index, showed a rather moderate RI for domestic bug abundance and marginally negative effects. The goat-equivalent index differed largely between ethnic groups (69×) and between movers and nonmovers; nearly one in every four Qom households owned no poultry and most had no livestock –stark measures of reduced livelihoods in a context where employment was rare and hunter-gatherer habits are no longer productive or feasible.
The application of domestic insecticides by householders was a moderately important factor negatively associated with domestic infestation and bug abundance as in other surveys [20,25,30], despite the fact that low-concentration sprays were mainly used against mosquitoes (S1 Text), as in other settings [69], and probably had very limited effects on triatomine bugs. Domestic insecticide use may be a surrogate for householders’ economic and behavioral aspects; its use implies both the capacity to purchase insecticides and willingness to take protective actions in response to nuisance insects. Two other protective practices (window screens and bed nets) showed opposite associations with ethnicity and domestic infestation in bivariate analyses. Window screens were restricted to brick-cement, Creole houses, and their presence was negatively associated with domestic bug infestation, as expected [70]. Conversely, only Qom households used bed nets, and their use was significantly and positively related to domestic infestation. This may be an example of reverse-causality effects which frequently limit the interpretation of associations derived from cross-sectional surveys. How well window screens and bed nets acted against triatomine bugs in the study setting remains to be determined.
House infestation with T. infestans (31.9%) was much lower than expected despite the fact that most houses provided suitable conditions for triatomines and the lack of insecticide spraying campaigns over the previous decade. Under similar circumstances, up to 90% of houses were infested elsewhere in the Gran Chaco [6,25,71]. This suggested that other additional factors may have affected the process of house reinfestation. Among the possible candidates we included: housing instability combined with a large fraction of recently-built houses; partial housing improvements (as shown by metal roofs); selective insecticide sprays; fewer peridomestic sites and fewer peridomestic foci of T. infestans than in other resource-constrained rural areas including Area I [20,57]. Indeed, the absence of a strong positive relationship between the occurrence of peridomestic foci and domestic infestation or bug abundance contradicts other findings elsewhere in the dry Chaco [7,56,57,72] and with various species of Triatominae [73,74], and may be explained by the paucity of peridomestic sites and domestic animals in Area III.
Housing instability was evidenced by the large mobility of Qom households within the study area and the municipality, and agreed with the large fraction of houses having less than 5 years of age at baseline. Residential mobility was virtually restricted to Qom households, indicating distinct patterns of settlement, housing occupation and displacement over time related to ethnicity. The backdrop for these patterns is the high population growth rate of local Qom population during recent decades, as evidenced by the very young age structure and decennial census figures, combined with intense out-migration to cities [19], in a context of structural poverty. Over 95% of the study population was native to the municipality of Pampa del Indio. The approximate stability of total population numbers in the study area is explained by out-migration with different destinations (S1 Text). Moreover, the highly skewed sex ratio toward males may be explained by gender-biased migration rates and accelerated population growth, but other competing hypotheses should be further investigated. The nature and implications of this rural-to-rural (local) movement are essentially different from unidirectional, rural-to-peri-urban migrations affecting house infestation elsewhere [75,76], and pose special challenges for research and vector control (see below). Housing instability, household mobility and migration patterns are key to understanding the process of house (re)infestation and to designing locally adapted vector surveillance systems in the Gran Chaco region.
Selective insecticide sprays performed by the local health system two years before our baseline survey were expected to lower house infestations, but a closer look revealed that the sprayed villages ranked at the top of the baseline infestation list (32.8–50.8%). The limited village-level effects of these treatments may in part be attributed to incomplete house coverage (37.1%); the occurrence of moderate resistance to pyrethroids in local T. infestans populations [39]; and fast reinfestation from untreated, neighboring sources (e.g.,[57,59]). Unfortunately, the selection criteria used to decide which villages and houses were sprayed could not be recalled by the participants in charge. If the sprayed villages had been selected based on having high infestation prevalence then we would expect them to have higher infestation rates than other villages two years later (as observed), given the partial spray coverage and strong association between preintervention and postintervention house infestation rates recorded in different settings (e.g.,[6,59,72,77]). Although householders reported additional insecticide treatments not registered by the local or provincial health system, 68.0% of responding households reported their current premises had never been treated with insecticides. Such low rates of spray coverage are roughly consistent with the average age of houses; the high rates of house destruction and reconstruction, and the dates and partial coverage of previous insecticide spraying campaigns.
Qom households had a threefold prevalence of domestic infestation than Creoles’, but the multivariate analyses revealed a large degree of uncertainty around the adjusted OR. The differential infestation between ethnic groups was greater than in Area I (34% versus 20%, respectively, Fig. 3 in [20]), where the Qom minority had similar domestic infestation levels as in Area III. Both areas and ethnic groups differed substantially in the mean number of poultry, livestock and other attributes (S2 Table compared with Table 2 in [20]), indicating that their living conditions were heterogeneous in several respects (inter-house distance, domestic area, household educational level, and domestic infestation, Table 1). Both in Area I and III, ethnicity had low RI when other important variables with a priori support were included in the global models, and the sizable differences in domestic infestation between ethnic groups presented a large degree of uncertainty. Other unidentified factors may underlie these heterogeneities within ethnic groups. Qom households likely were more infested because of the convergence of multiple factors intimately related to structural rural poverty (e.g., poor housing quality, overcrowding, less frequent insecticide use) rather than to direct ethnic or cultural effects facilitating house invasion and colonization by triatomine bugs. In fact, the increased mobility of Qom households most likely reduced house infestation and bug population size, and perhaps increased the spatial propagation of bugs. After accounting for the effects of other factors with high RI, ethnicity per se was a poor predictor of house infestation status.
Our study has some limitations and strengths. Because timed-limited manual searches underestimate the true house infestation rates at low levels of bug abundance [78], the additional bug collection methods were used to refine the assessment of infestation status. In general, householders’ bug collections (not bug notifications) have usually been more sensitive than timed-manual searches in domiciles across settings [8,78], but this pattern may vary with prior experience and promotion of vector surveillance activities (absent in our study area). Householders’ actual degree of compliance with bug collections usually remains unknown and is hard to gauge. Although some bias may result from the non-systematic bug collections performed by householders, searches conducted during insecticide spraying operations were systematic. Householders’ notification of domestic infestation by T. infestans had a poor agreement with timed-manual searches. This discrepancy is unlikely to be due to confusion with other insects because householders clearly distinguished T. infestans from other similar insects such as T. sordida (which showed low infestation prevalence in the area), and was perhaps related to notifying infestations that occurred in the past.
Selected sociodemographic variables measured four years after the baseline survey included a large number of missing data because the households that moved out were no longer reachable. Missing data frequently involve households with lower socioeconomic status [79], and in our study they apparently occurred at random (S1 Text). If household mobility was inversely associated with socioeconomic status (as suggested by the limited comparison between movers and nonmovers), the smaller sample with 274 houses could potentially be biased. However, the multimodel analysis of the global model reached similar qualitative conclusions regardless of the number of houses included, which suggests the magnitude of the bias was not serious. Variables based on householders’ reports of past events (e.g., insecticide sprays) may be affected by recall bias and social or cultural barriers against effective communication, including Spanish language skills. To minimize the latter, the questionnaires had previously been tested in other areas of the municipality and interviewees were acceptably fluent in Spanish. Some of the variables (e.g., land ownership, welfare support) may be open to social propriety issues and responses may have a variable degree of validity; for communal land ownership, however, responses matched other local sources of information. Information on the patterns of settlement, the extent of land owned by each household, income and employment may provide valuable, additional indicators of wealth and livelihood, but may also be affected by response bias. Major strengths of our study include the large number of Qom households surveyed over time and detailed household-level information on a sizable number of ecological and sociodemographic factors related to house infestation.
Our study documented threats of active vector-borne transmission of T. cruzi in approximately 27% of the households (as determined by the occurrence of domestic infestations), and identified manageable variables that may be targeted for improved interventions and risk stratification. Improving housing quality and living conditions is urgently needed and largely exceeds Chagas disease vector control because housing improvements will impact positively on family health. Reducing the presence of chickens in human sleeping quarters [20,21,25,26,64,67] and applying insecticides in more effective ways when required may contribute to improved vector control. Although these factors are frequently construed as environmental or ecological, the types of housing, land ownership, habits of raising livestock or poultry, frequency of insecticide use and type of preventive practices have historical, sociodemographic, cultural and political roots.
The household mobility patterns recorded have serious implications for vector and disease control. In the preintervention context of an infested area under marginal vector control (as in 2008), the mobility of Qom households implied the potential carriage of bugs in their belongings to the new houses, while leaving bugs behind in the rubble of knocked down walls. The recently-built houses represented new habitat patches susceptible to bug invasion and colonization, and therefore decreased the fraction of all houses effectively protected by the long-lasting residual effects of recent insecticide sprays. On the flip side, the processes of house destruction and reconstruction are expected to cause major negative impacts on the local abundance of bugs by increasing bug mortality and dispersal.
The mobility of some indigenous populations may pose special challenges to traditional housing improvement programs relying on stable settlement and secure land tenure. More knowledge of the drivers of household mobility, migration and the desired types of housing of Qom and other indigenous peoples which had a nomadic or seminomadic tradition are needed. The design of Chagas disease prevention programs and other health interventions directed to indigenous populations should address their specific needs and beliefs [80,81]. Improving housing quality in isolation, while traditional agricultural activities continue in decline and other sources of local employment are rare, may not stop the rural-to-peri-urban exodus across ethnic groups.
The links between household educational levels and domestic infestation require more elaboration and specific research on the mechanisms involved. This area offers new opportunities for innovative interventions through health education and promotion workshops [82] that include, but are not restricted to, community-based vector and disease surveillance, control and treatment. Better access to formal education may also contribute directly and indirectly to primary and secondary disease prevention (e.g., by increasing awareness of treatment opportunities). The large fraction of Qom and Creole households who managed to keep their premises free from triatomine bugs using the scarce means available to them holds promise for further improvements with a modest investment of resources. Households performing good practices of vector control may contribute as agents of change to further reduce infestation and transmission risks in community-based control programs.
The strong heterogeneities in the distribution of ecological and sociodemographic factors associated with house infestation may be used for risk stratification and targeted interventions. Large households residing in small-sized, precarious houses, with few or no livestock or poultry and lower educational levels, appear to be especially vulnerable for Chagas and other infectious diseases. These households and the affected communities may benefit from targeted disease prevention activities channeled through a more vigorous, adequately staffed, primary healthcare system deployed in the affected rural areas.
|
10.1371/journal.pgen.1004170 | Arf4 Is Required for Mammalian Development but Dispensable for Ciliary Assembly | The primary cilium is a sensory organelle, defects in which cause a wide range of human diseases including retinal degeneration, polycystic kidney disease and birth defects. The sensory functions of cilia require specific receptors to be targeted to the ciliary subdomain of the plasma membrane. Arf4 has been proposed to sort cargo destined for the cilium at the Golgi complex and deemed a key regulator of ciliary protein trafficking. In this work, we show that Arf4 binds to the ciliary targeting sequence (CTS) of fibrocystin. Knockdown of Arf4 indicates that it is not absolutely required for trafficking of the fibrocystin CTS to cilia as steady-state CTS levels are unaffected. However, we did observe a delay in delivery of newly synthesized CTS from the Golgi complex to the cilium when Arf4 was reduced. Arf4 mutant mice are embryonic lethal and die at mid-gestation shortly after node formation. Nodal cilia appeared normal and functioned properly to break left-right symmetry in Arf4 mutant embryos. At this stage of development Arf4 expression is highest in the visceral endoderm but we did not detect cilia on these cells. In the visceral endoderm, the lack of Arf4 caused defects in cell structure and apical protein localization. This work suggests that while Arf4 is not required for ciliary assembly, it is important for the efficient transport of fibrocystin to cilia, and also plays critical roles in non-ciliary processes.
| Primary cilia are ubiquitous sensory organelles that play vital roles in an ever-growing class of human diseases termed ciliopathies including obesity, retinal degeneration and polycystic kidney disease. The proper function of the primary cilium relies on a cell's ability to target and concentrate specific receptors to the ciliary membrane – a unique subdomain of the plasma membrane yet little is known about how receptors are trafficked to the primary cilium. Mutations affecting the ciliary localized receptor fibrocystin (PKHD1) cause autosomal recessive polycystic kidney disease, which affects approximately 1∶20,000 individuals. Previously we identified a motif located in the cytoplasmic domain of fibrocystin that is required for its ciliary localization. In this work we demonstrate that the ciliary targeting sequence (CTS) of fibrocystin interacts with the small G protein Arf4 and this interaction is important for the efficient delivery of the CTS to cilia in cultured cells. Disruption of Arf4 in mice results in defects in the non-ciliated visceral endoderm and death at mid-gestation indicating Arf4 has vital functions in addition to ciliary protein trafficking.
| Cilia play diverse motility and sensory functions throughout the eukaryotic kingdom, but play especially critical roles in vertebrates where severe defects lead to embryonic lethality while mild defects cause a wide range of syndromes affecting every organ system. Both the motility and sensory functions of cilia are important for health and development, but it is now recognized that sensory defects underlie the most severe maladies affecting humans. The sensory functions of cilia rely on a cell's ability to target and concentrate a specific set of receptors to the ciliary membrane. While contiguous with the plasma membrane of the cell, the ciliary membrane is a distinct compartment to which the cell targets and concentrates a unique complement of proteins [1], [2]. The list of membrane proteins found in the ciliary compartment is constantly growing; among the most studied ciliary proteins are the polycystins and fibrocystin that are defective in human polycystic kidney disease, rhodopsins and opsins that are critical for vision and the patched and smoothened receptors of the hedgehog pathway.
The mechanism that targets membrane proteins specifically to the ciliary compartment is an active area of study but very little is definitively known [3]. It appears that ciliary membrane proteins contain cis-acting motifs that cause them to be localized to cilia. We identified one of these ciliary targeting sequences (CTS) in fibrocystin, the gene product of the human autosomal recessive polycystic kidney disease gene (PKHD1) [4]–[7]. Like many other CTSs, the fibrocystin CTS contains lipid-modified residues that target the protein to lipid rafts, which appears to be part of the ciliary trafficking pathway. We proposed that this sequence might interact with proteins that are important for sorting or transport to the ciliary membrane compartment. In support of this idea, we found that the fibrocystin CTS interacted with Rab8, a protein widely recognized as important to ciliary membrane protein trafficking [7]–[9]. In the present work we asked if the fibrocystin CTS could interact with Arf4 as work of Deretic and colleagues has shown that this protein interacts with the CTS of opsin and is important for the formation of rhodopsin carrier vesicles at the Golgi complex [10], [11].
Arf4 is a small G protein in the Arf subfamily of Ras-related small G proteins. Mice have six members of this family while humans have lost Arf2 and have five members. Arf1 and Arf6 have been best-studied and are thought to organize membrane protein cargos into coated vesicles for transport to specific lipid domains in the cell [12]–[15]. Arf1 forms coated vesicles at the Golgi complex crucial for trafficking between the ER and Golgi and throughout the cell while Arf6 is thought to operate at the plasma membrane and regulate endosomal-membrane traffic. Arf4, which was thought to evolve from an Arf1-like precursor when metazoans arose, is a relatively unstudied member of the family [13]–[15]. Arf4 was first proposed to be important for ciliary trafficking when it was found to interact with the C-terminal tail of rhodopsin where the CTS is localized [10]. Depletion of Arf4 from an in vitro budding assay showed that it was important for the formation of rhodopsin carrier vesicles [11]. The CTSs of rhodopsins are contained in the last five amino acids (QV[S/A]PA) at the C-terminal end of the protein. The V and P residues are mutated in human patients with autosomal dominant retinitis pigmentosa. These residues were found to be important in an in vitro assay for the formation of rhodopsin carrier vesicles, thus the sequence has become known as a VXPX motif. A similar RVXP motif is present in the CNGB1b subunit of the CNG channel, another ciliary-localized protein [16]. The VXPX sequence is part of the CTS in polycystin-1 and polycystin-2 and it is hypothesized that VXPX motifs function as Arf4 binding sites for transport to the cilium [12], [17], [18].
In this work we ask if Arf4 plays a role in trafficking the fibrocystin CTS to the cilium and probe the function of Arf4 in vivo by analyzing a mutant mouse. Although the fibrocystin CTS does not contain a VXPX motif, it does bind to Arf4. Arf4 is not required for the trafficking of the fibrocystin CTS to cilia, but knockdown of Arf4 increases the time needed for the protein to travel from the Golgi complex to the cilium. Deletion of Arf4 in the mouse does not affect the formation or function of nodal cilia, but causes embryonic lethality at mid-gestation, probably due to trafficking defects in the visceral endoderm.
Fibrocystin (polyductin), the human autosomal recessive polycystic kidney disease gene product, is targeted to cilia by an 18-residue ciliary targeting sequence (CTS) located in the cytoplasmic C-terminal tail of the protein. Previously we showed that this sequence interacted with Rab8 and proposed that it may function by interacting with proteins involved in the sorting and transport of ciliary membrane proteins [7]. Recent work indicates that Arf4 is required for trafficking of other ciliary proteins including rhodopsin and the polycystins [10], [11], [18]. Arf4 is one of six members of the Arf subfamily of small G proteins. Our previous analysis showed that in non-ciliated cells, the GFP-CTS localized to small puncta that appeared to be lipid microdomains rich in GM1 gangliosides. Interestingly, Arf4 exhibited significant co-localization with the CTS-GFP in these puncta while the rest of the Arf family did not (Figure 1A). To determine if this colocalization represented a physical interaction, we immunoprecipitated each of the Arfs and asked if the CTS-GFP was co precipitated. Arf4 strongly precipitated the CTS, while the other Arfs either precipitated no CTS (Arf1, 3, 5, 6) or only a small amount (Arf2) (Figure 1B). Because ARF2 is a pseudogene in humans we did not pursue the Arf2 interaction any further. The failure of the CTS to interact with Arf5 suggests that the Arf4 interaction is likely to be specific as Arf4 and Arf5 share over 90% identity at the amino acid level.
To further characterize the interaction between Arf4 and the CTS we tested the ability of a series of CTS deletion constructs to interact with Arf4 in the co-expression immunoprecipitation assay (Figure 1C). In this assay, the CTS is sufficient for binding as the 3–20 construct was precipitated by Arf4 while the 63–193 construct, which lacks the CTS, was not bound. The 1–19 and 4–20 constructs also interact with Arf4 but do not target to cilia, indicating the CTS has functions in addition to Arf4 binding.
Next we examined the ability of Arf4 to interact with a series of alanine-scanning mutations within the CTS (Figure 1D). In general, the ability of Arf4 to bind the CTS correlated with the ability of the CTS to traffic to cilia. For example, mutations affecting the palmitoylated cysteines (CCC>AAA) or conserved basic residues (KTRK>AAAA) prevent ciliary targeting and likewise inhibit Arf4 binding. Mutations that exhibit little effect on the ciliary targeting of the CTS (LV>AA) similarly do not affect Arf4 binding. However the concordance was not perfect as the (IKP>AAA) mutation blocked ciliary targeting but had little effect on Arf4 binding, again supporting the idea that while the CTS is an Arf4 binding site, this is not its entire function.
Arf4 cycles between GTP and GDP bound states. To explore the role of this cycling on the ciliary trafficking of the CTS we measured the effects of constitutively active and dominant negative Arf4 on ciliary targeting and interaction with the CTS (Figure S1). Constitutively active Arf4 (Q67L) co-localized with IFT20 at the Golgi Complex while wild type or dominant negative (T13N, T48N) Arf4 displayed a punctate distribution in the cytoplasm (Figure S1A). The mutant forms of Arf4 retained the ability to bind the CTS (Figure S1E). Expression of dominant negative Arf4 reduced the percent of ciliated cells (Figure S1B) and ciliary length (Figure S1C). Surprisingly, over-expression of any Arf4 constructs completely inhibited CTS trafficking to the cilium (Figure S1D). These data indicate that increases in Arf4 levels prevent ciliary trafficking of the CTS possibly by sequestering it in the cytoplasm.
Prior studies in frog photoreceptors [10], [11] and cultured mammalian cells [18] suggest that Arf4 functions at the Golgi complex to direct rhodopsin and polycystin-1 to the cilium. Our initial immunofluorescence and biochemical results suggested that Arf4 may also be required for the ciliary targeting of fibrocystin. To test the role of Arf4 in the trafficking of fibrocystin, we developed a pulse chase assay to measure its movement through the endomembrane system and delivery to the cilium. To do this, we created a chimeric molecule containing the extracellular domain of CD8 fused to the C-terminal tail of fibrocystin (including the transmembrane domain) with a SNAP tag [19] on the C-Terminal end (Figure 2A) and expressed this in mouse kidney collecting duct cells (IMCD3). The extracellular domain of CD8 contains a signal sequence that when combined with the transmembrane domain of fibrocystin produces a type 1 membrane protein with membrane topology the same as native fibrocystin except that the large (3,851 residue) extracellular domain is replaced with the CD8 epitope. The SNAP tag is a fragment of the DNA repair protein O6-alkylguanine-DNA alkyltransferase that can be covalently modified with benzyl guanine derivatives and allows for pulse chase experiments [20], [21]. We developed a protocol to follow the movement of this chimeric protein through the endo-membrane system. At the beginning of the experiment all existing SNAP sites are blocked with a non-fluorescent benzyl-guanine so that only newly synthesized protein will be labeled by the fluorescent benzyl-guanine. The newly synthesized SNAP-CTS is first detected in the endoplasmic reticulum as expected for a trans-membrane receptor. The protein then moves to the Golgi complex where it can be trapped using a 19°C temperature block [22]. Shifting the cells back to 37°C allows the accumulated protein to exit the Golgi and traffic to the cilium where it can be detected within 30 min of release.
To determine if Arf4 is involved in trafficking of the CTS, cells expressing the CD8-CTS-SNAP construct were treated with siRNA to reduce the level of Arf4. Arf4 mRNA level was reduced greater than 90% as compared to cells treated with a control scrambled siRNA (Figure 2B). The Arf4 knockdown did not affect the percent of ciliated cells nor did it affect ciliary length (data not shown). Arf4 knockdown did not affect the total amount of CD8-CTS-SNAP in the cilium as measured by CD8 fluorescence (Figure 2D). To determine if the reduction of Arf4 affected the rate of delivery to the cilium, we measured the time that it takes for the CD8-CTS-SNAP construct to move from the Golgi complex to the cilium by using the pulse chase protocol described above. Newly synthesized protein was accumulated in the Golgi complex by a 19°C block and then released by shifting cells to 37°C (Figure 2 C, D). In control cells, after release from the Golgi block the CD8-SNAP-CTS moves quickly to the cilium and is detectable at the 1 hr time point with the ciliary level peaking at about 2 hrs post block (Figure 2C, insets). In contrast, when Arf4 is depleted, little CD8-CTS-SNAP is detectable in the cilium within the first two hrs after release from Golgi block but protein is detectable at 4 hrs. These data indicate Arf4 is not absolutely required for the delivery of the CTS to the cilium but does play a kinetic role in the steps between the Golgi complex and the cilium.
Our initial data indicates that Arf4 interacts with the CTS of fibrocystin and this interaction is required for the efficient delivery of newly synthesized CTS to the cilium. Data in the literature suggests thatArf4 is important for the trafficking of rhodopsin and the polycystins to cilium and it has been suggested that it is a global regulator of ciliary cargo [10], [11], [18]. To determine if Arf4 is a global regulator of ciliary protein trafficking we created an Arf4 genetrap mouse with the prediction that if it plays this role, the mouse should have ciliopathy phenotypes. Embryonic stem cells harboring a LacZ insertion in the Arf4 locus just downstream of exon 3 were obtained from the Sanger Institute and used to create an Arf4 genetrap mouse (Figure 3). The allele we generated expresses less than 1% of control Arf4 mRNA and is, at minimum, a strong hypomorph (Figure 3D).
Mice lacking cilia typically die mid-gestation around day embryonic day (E) 10 with a failure to undergo embryonic turning and have severe disruptions in left-right patterning [23]. Similar to this, the Arf4 mutant mice are embryonic lethal at mid-gestation with no live embryos detected after day E10.5 (Figure 3E). At E9.5 the mutant embryos were smaller than either wild type or heterozygous embryos and almost never completed embryonic turning to assume the characteristic fetal body position that is observed in almost all wild type and heterozygote embryos by this time (Figure 3B, C).
To investigate if Arf4 affects ciliary assembly we performed scanning electron microscopy (SEM) of the node, which is thought to be the first ciliated structure in the embryo. To preclude differences caused by a developmental delay in the mutant embryos, embryos from multiple litters were dissected and those at the 3–4 somite stages were examined by SEM (Figure 4A). No differences were observed in either length or number of cilia present on the mutant nodes as compared to wild type or heterozygous embryos (Figure 4B, C) indicating Arf4 is not required for ciliary assembly. Nodal cilia beat to create a leftward flow required to break the left/right symmetry of the developing embryo [23]. This leads to asymmetric gene expression patterns on the left versus right side of the embryos and eventually leads to the right-right pattern of the abdominal and visceral organs. One of the earliest physical manifestations is the looping of the heart tube, which under normal conditions adopts a characteristic D-loop by day E9.5. The developing heart in Arf4 mutant embryos always adopts a D-loop indicating the nodal cilia present in the Arf4 mutant embryos are functional in breaking left/right symmetry (Figure 4D, E).
Arf4 mutant mice die between embryonic days 9 and 10. This embryonic lethality does not appear to be connected to ciliary dysfunction leaving the cause of death unknown. To identify the site of pathology, we took advantage of the β-galactosidase insertion that was used to generate this allele and performed X-Gal staining to identify the sites of high Arf4 expression (Figure 5). As expected, no staining was observed in the wild type embryos. In the mutants and heterozygotes, the majority of the staining was observed in the visceral yolk sac with only mutants exhibiting label in the embryo on E8 and E9 (Figure 5). At day 8, the label was observed in the allantois, paraxial mesoderm and in the forming definitive endoderm in hindgut region (Figure 5C, D; Figure S2). By E9.5 the definitive endoderm in the posterior foregut, including the liver bud was most strongly stained (Figure 5F). The yolk sac consists of two layers; the outer visceral endoderm is composed of highly polarized cells covered with microvilli on their apical surface, while the inner mesoderm gives rise to the developing blood islands in early development of the circulatory system [24]. The β-galactosidase activity was exclusively found in the visceral endoderm layer of the yolk sac (Figure 5F, bottom row).
As Arf4 is most highly expressed within the visceral endoderm during development, we examined this tissue further by immunofluorescence and electron microscopy (Figure 6). To determine if these tissues were ciliated, we stained yolk sacs and sectioned embryos for cilia, and imaged by confocal microscopy. We did not detect cilia on the visceral endoderm at embryonic day 8.5 or 9.5. However cilia are present on the adjacent inner layer of mesoderm cells in both wild type and mutant embryos at these stages in development (Figure 6A, B).
The visceral endoderm serves as the major secretory and absorptive tissue of the developing embryo prior to placental formation [25]. As expected for a highly absorptive tissue, the visceral endoderm has a well-developed brush border on the apical surface and large apical vacuoles/lysosomes that facilitate uptake and breakdown of macronutrients from the maternal blood supply (Figure 6C). In Arf4 mutants the apical/basolateral polarity appears intact and the brush border remains, but the microvilli are less organized and the apical vacuoles are missing, suggesting that visceral endoderm function may be compromised. In addition, bulbous misshapen microvilli are often observed along with small vesicles that are surrounded by a fuzzy coat, which is not seen on the microvilli (Figure 6C, middle row). The cell-cell contacts also appear to be compromised as more space is seen between the mutant cells by TEM; moreover microprojections that form the interdigitations between the lateral surfaces of adjacent cells can be observed by SEM on the apical surface of the mutants but not the controls (Figure 6C, bottom row). It is interesting to note that the amnion is not detectably altered in the Arf4 mutants (Figure 6D) indicating that not all mutant cells are disrupted by the lack of Arf4.
The visceral endoderm carries out its absorptive function by localizing megalin (Lrp2) and other scavenger receptors on its apical surface. These receptors bind to substrates such as vitamins, lipoproteins and signaling molecules (reviewed in [26]), which are then internalized and transcytosed or broken down in the lysosome. Trafficking defects within the visceral endoderm result in embryonic lethality and are often associated with mislocalization of megalin [27], [28]. At E8.5, megalin is normally concentrated along the apical surface of the visceral endoderm (Figure 6E). Arf4 mutants have significantly reduced megalin staining at the apical surface and some of the protein appears in the cytoplasm. This suggests megalin trafficking is disrupted in Arf4 mutants and supports the ultrastructural studies of the visceral endoderm that indicate nutrient uptake is compromised within this cell layer.
Megalin is a type 1 membrane protein with a large extra cellular domain, single transmembrane span and a short cytoplasmic C-terminal tail similar to the structure of fibrocystin. Because of the similarity in structure and the observation that Arf4 mutant embryos have reduced megalin on the apical surface of the visceral endoderm, we asked if Arf4 interacts with megalin. Co-immunoprecipitation indicates Arf4 interacts with the intracellular domain of megalin (Figure 6F). Similar to what we observed with fibrocystin, the highly similar protein Arf5 did not interact with megalin. These data suggest Arf4 is involved in not just trafficking of ciliary cargo but also a larger class of trans-membrane receptors – including megalin.
The primary cilium is a sensory organelle and its proper function relies on the correct complement of receptors localized specifically in the ciliary membrane. Little is known about how proteins are sorted to the ciliary compartment but understanding this process is critical as defects in the signaling functions of primary cilia underlie a diverse group of human pathologies known collectively as ciliopathies. These diseases range from developmental defects of the brain, heart and other organs to chronic ailments including retinal degeneration, obesity and polycystic kidney disease. To study ciliary protein sorting, we focused on analysis of the trafficking of the transmembrane protein fibrocystin to the primary cilium. Mutations in the fibrocystin gene (PKHD1) are responsible for autosomal recessive polycystic kidney disease, a disorder afflicting approximately 1 in 20,000 individuals and a cause of significant mortality during the first year of life [29], [30]. The ciliary targeting sequence of fibrocystin is an 18 amino acid sequence contained in the cytoplasmic tail [7]. We had previously shown that the CTS interacts with the small G protein Rab8. In this work we studied the interaction of the CTS with another small G protein, Arf4. Arf proteins group vesicular cargo and through interactions with coat proteins form transport vesicles [12]–[14]. The proposed sorting ability of Arf proteins make them attractive candidates as specificity factors and recent work suggests Arf4 is involved in targeting rhodopsin and polycystin-1 to the cilium [10], [11], [18]. We found that Arf4 was capable of interacting with the fibrocystin CTS in a co-immunoprecipitation assay. The interaction is specific as Arf4 is the only member of this highly conserved family that precipitated significant amounts of the CTS. The analysis of deletion and alanine scanning mutants within the cytoplasmic tail of fibrocystin showed that the Arf4 interaction site was localized within the CTS. The ability of Arf4 to bind to the mutated CTSs roughly correlated to the ability of the CTS to enter the cilium suggesting that the Arf4/CTS interaction was functionally important.
Using SNAP-tagging technology we found that Arf4 is not absolutely required for the delivery of the fibrocystin targeting construct to cilia as the steady state level of the protein was not affected by knockdown of Arf4. This is in contrast to the reported effect of knockdown on the trafficking of polycystin-1 where Arf4 knockdown significantly reduced ciliary levels of polycystin-1 [18]. However, in our hands, the delivery of newly synthesized fibrocystin fusion protein was slower in the knockdown cells indicating that Arf4 is needed for the efficient delivery to primary cilia. The delayed, but eventual delivery of the CTS to the cilium may be a result of residual Arf4 protein (∼10% of the mRNA remains in the knockdown cells) or it may indicate an alternative pathway to the cilium that does not utilize Arf4. Smoothened appears to enter the ciliary compartment by first traveling to the plasma membrane before moving into the cilium [21] and so it is possible that this route can also be utilized by fibrocystin.
Our work, and work from others, indicates that Arf4 may play a key role in sorting transmembrane receptors to the cilium. This suggests that Arf4 may be an important player in human diseases such as retinal degeneration and polycystic kidney disease. To better understand the function of Arf4 and its possible role in ciliopathies, we created an Arf4 mutant mouse. If the primary function of Arf4 is specific to cilia, we would expect the mutant mice to exhibit phenotypes in common with established mutations that affect cilia. Mice with strong defects in ciliary assembly die at mid-gestation with severe left-right abnormalities while those with more mild ciliary defects survive longer and display phenotypes indicative of hedgehog signaling dysfunction. Arf4 mutant mice die between embryonic days 9 and 10, which is similar to the time when mice with severe ciliary assembly defects die. However, ciliary assembly is normal in the embryonic node and the nodal cilia are functional as all embryos broke left-right symmetry properly and formed a D-looped heart. The embryonic lethality but lack of ciliary defects suggested Arf4 might have functions in addition to the ciliary targeting of transmembrane proteins. Expression analysis between embryonic day 7 and 10, around the time the Arf4 mutant embryos were dying, indicated that the major site of Arf4 expression was in the visceral endoderm. Examination of the visceral endoderm indicates this tissue is not ciliated at this time in development although cilia are present on the adjacent mesoderm. The visceral endoderm is the major secretory and absorptive tissue of the developing embryo prior to chorioallantoic placenta formation [24] and defects within this cell layer often result in embryonic lethality [27], [28]. Arf4 mutant embryos have multiple defects within the visceral endoderm. Ultrastructural analysis of the visceral endoderm indicates Arf4 mutant embryos have defects in cell-cell contacts and organization of the brush border, but most strikingly, they lack the large lysosomes normally present in healthy tissue. The absence of lysosomes could be a direct effect of the Arf4 mutation, but is more likely an indirect consequence of a failure to absorb nutrients from the adjacent maternal blood supply. A failure to uptake nutrients is consistent with the observed growth restriction evident by embryonic day 7 and likely accounts for the lethality around day 9. The apical surface of the visceral endoderm is covered with microvilli that contain a number of scavenger receptors that bind ligands including vitamins and lipoproteins required by the developing embryo. We examined the distribution of one of these receptors, megalin within the visceral endoderm. Megalin is normally localized to the apical surface of the developing visceral endoderm however in Arf4 mutants, megalin fails to localize to the apical surface. Megalin is a large single span transmembrane receptor with membrane topology similar to fibrocystin. Co-immunoprecipitation assays indicate that Arf4 can interact with megalin similar to what we observed between Arf4 and fibrocystin. This interaction and the observed defects in megalin trafficking in the Arf4 mutant suggest that Arf4 is required to target megalin to the apical surface. While our work was in progress, another gene trap allele of Arf4 was generated and used to study dendritic spine formation in the brain. Similar to our allele, homozygotes were lost prior to birth but no further analysis was done to characterize the homozygous phenotypes [31].
RNAi studies suggested that the Arf family of proteins was highly redundant and it was not predicted that the genetic loss of any one would have a strong phenotype [32], [33]. The observation that Arf4 null mice die mid gestation indicates that this is not completely correct. To date, Arf6 is the only other Arf family member that has been mutated in the mouse. Like Arf4, Arf6 null mice die during development, although they survive longer than the Arf4 mice. The major defect in Arf6 null mice was in the liver, where the lack of Arf6 caused increased rates of apoptosis resulting in a significantly smaller liver with lethality around embryonic day 15. [34]. The fact that both Arf4 and Arf6 mice survive through early gestation suggests that neither of these genes are essential genes at the cellular level, but do play critical functions in particular cells at particular times in development. In the case of the Arf4 mouse, the major defect was in the visceral endoderm, a tissue with a very high rate of internalization and trafficking of lipid and protein molecules. It is possible that this is the first point in development that requires this level of internalization and trafficking. The analysis of a floxed allele will be of interest to determine if Arf4 is required in adult cells with a high rate of flux such as the intestine or kidney proximal tubule, or even the rod and cone photoreceptor cells where a high flux is needed to maintain the outer segment.
The literature on Arf4 has mostly focused on proposed roles in trafficking proteins to the ciliary membrane compartment. However, our finding that the highest level of expression is in the visceral endoderm, which is not ciliated at the time of high expression, suggests that Arf4 has functions outside the targeting of ciliary cargo. This is consistent with recent studies showing a role for Arf4 in Golgi stress responses [35] and the finding that class II Arfs (Arf4, Arf5) play roles in the trafficking of dense core vesicles [36], [37] and secretion of Dengue virus particles [33]. In the case of the dense core vesicle transport, Arf4 and Arf5 interacted with two calcium dependent activator proteins for secretion (CAPS1 and CAPS2). This interaction was required for the efficient trafficking of dense core vesicles as knockdown of either Arf4/5 or CAPS1/2 significantly reduced chromogranin secretion [36], [37]. In the case of Dengue virus production, Arf4 and Arf5 were required for the secretion of subviral particles and the Arfs were thought to act though an interaction with prM glycoprotein of the virus. Interestingly, the prM glycoprotein contains a VXPX motif in the C-terminus similar to the Arf4 binding site in rhodopsin [10]. However mutation of the VXPX motif did not disrupt interaction with Arf4 indicating that it is not the binding site [33]. This is consistent with our finding that Arf4 binds to the CTS of fibrocystin, which does not contain a VXPX motif and studies of nephrocystin-3, which contains a VXPX motif, but this motif is not necessary for ciliary targeting [38].
In conclusion, we have shown that Arf4 plays a role in the efficient transport of the fibrocystin CTS to the cilium, but it is not required for ciliary assembly and in the mouse has critical functions in non-ciliated cells. Thus, our work, and other published work, suggests that Arf4 function is not restricted to ciliary assembly but rather plays a broader role in cellular trafficking.
Mouse work was approved by the University of Massachusetts Medical School IACUC.
An Arf4-mutant ES cell line was obtained from the Sanger Center and used to generate Arf4Gt(AY0614)Wtsi mutant mice. The animals used in this study were a mix of 129 and C57Bl6 backgrounds. Embryonic ages were determined by timed mating with the day of the plug being embryonic day 0.5. Genotyping was carried out with the following primer pairs: Arf4-1 AGCAGCCTCATTGTCCTAGC+Arf4-2 CCTCCCCACAATTCAACAAT (product size = 189 bp in wildtype) and Geo-3 GATCGGCCATTGAACAAGAT+Geo-4 CAATAGCAGCCAGTCCCTTC (product size = 280 bp in mutant).
IMCD3 (ATCC) were grown in 47.5% DMEM 47.5% F12, 5% fetal bovine serum, with penicillin and streptomycin at 37°C in 5% CO2. Cells were transfected by electroporation (Bio-Rad, Hercules CA). Stable cell lines were selected by supplementing the medium with 400 µg/ml of G418 (Sigma, St. Louis, MO). Clonal lines were selected by dilution cloning after drug selection.
For scanning electron microscopy (SEM), timed pregnant females were euthanized by approved IACUC protocols, embryos dissected in DMEM/F12 supplemented with 5% fetal bovine serum, fixed overnight in 2.5% glutaraldehyde in 0.1M sodium cacodylate. Fixed embryos were rinsed twice with 0.1M sodium cacodylate, osmicated in 1% osmium tetroxide, dehydrated in a graded ethanol series and critical point dried (Autosamdri-815, Series A, Tousimis Research Corp.). Dried embryos were sputter coated with iridium to a thickness of 3 nm (Cressington 208 HR Sputter Coater, Ted Pella, Redding, CA, USA) and examined in a scanning electron microscope (FEI Quanta 200 FEG SEM) [39]. For comparison of nodal cilia, embryos were developmentally matched by counting somite number.
For transmission electron microscopy (TEM), samples were fixed, osmicated and dehydrated as described above. Dehydrated samples were then infiltrated first with two changes of 100% propylene oxide and then with a 50%/50% propylene oxide/SPI-Pon 812 resin mixture. The following day, three changes of fresh 100% SPI-Pon 812 resin were done before the samples were polymerized at 68°C in plastic capsules. The samples were then reoriented and thin sections were placed on copper support grids and contrasted with Lead citrate and Uranyl acetate. Sections were examined using a Phillips CM10 TEM with 80 Kv accelerating voltage, and images were captured using a Gatan TEM CCD camera.
Cells for immunofluorescence microscopy were grown, fixed, and stained as described [40]. For visceral endoderm immunofluorescence, E8.5 embryos were fixed for 15 minutes at room temperature with 4% paraformaldehyde in PBS rinsed twice in PBS, equilibrated in 30% sucrose overnight and embedded in Tissue Freezing Media (Triangle Biomedical Sciences). Cryosections (10 µm) were blocked for 1 hour in 1% bovine serum albumin, incubated with primary antibodies overnight at 4°C.
Primary antibodies included anti acetylated-tubulin (611B1, Sigma, St. Louis MO), anti-FLAG (Sigma), anti-MmIFT20, anti-MmIFT88 [41], MmIFT27 [42], anti-golgin97 (CDF4, Molecular Probes) anti-BIP (clone 40, BD Transduction Laboratories), anti-Rab11 (clone 47, BD Transduction Laboratories), anti-TfR (clone H68.4, Invitrogen), anti-giantin [43] and anti-megalin (P-20, Santa Cruz Biotechnology).
Widefield images were acquired by an Orca ER camera (Hamamatsu, Bridgewater, NJ) on a Zeiss Axiovert 200 M microscope equipped with a Zeiss 100× plan-Apochromat 1.4 NA objective. Images were captured by Openlab (Improvision, Waltham, MA) and adjusted for contrast in Adobe Photoshop. If comparisons are to be made between images, the photos were taken with identical conditions and manipulated equally. For the quantification of SNAP and CD8 in the cilia, the length, area, and average fluorescence intensity of the cilia was measured using the measurement tools of Openlab. To determine significance of differences, data from three independent experiments were subjected to an unpaired Student's T test. Confocal images were acquired by a Nikon TE-2000E2 inverted microscope equipped with a Solamere Technology modified Yokogawa CSU10 spinning disk confocal scan head. Z-stacks were acquired at 0.2 µm or 0.5 µm intervals and converted to single planes by maximum projection with MetaMorph software. Bright field images were acquired using a Zeiss Axioskop 2 Plus equipped with an Axiocam HRC color digital camera and Axiovision 4.0 acquisition software.
The construct (pJAF270) used for SNAP trafficking assays was constructed by fusing the extracellular domain of mouse CD8a [44] to the last 17 extracellular residues of mouse fibrocystin through the first 27 intracellular residues, the SNAP tag was cloned onto the c-terminal end of the CTS creating CD8-CTS-SNAP. Mouse kidney cells stably expressing CD8-CTS-SNAP were incubated with 0.04 µM cell permeable non-fluorescent BG-Block (New England Biolabs) for 30 minutes to block all SNAP epitopes. Following 3 washes with complete growth media cells were allowed to synthesize new CD8-CTS-SNAP for 1.5 hrs before the addition of HEPES pH 7.4 to 20 mM and cycloheximide to 150 µg/ml, then shifted to 19°C for two hrs to accumulate CD8-CTS-SNAP at the Golgi complex. Cells were returned to 37° and allowed to traffic CD8-CTS-SNAP for the indicated periods of time before being fixed and stained.
For siRNA knockdown, cells were transfected by RNAiMAXX (Invitrogen) with SMARTpool siRNA (Dharmacon) targeting Arf4 (L-060271) or a non-targeting control (D-001810) and assayed for knockdown 48 hours post transfection.
FLAG-tagged Arf1-6 (pJAF215, pJAF213, pJAF214, pJAF216, pJAF210, pJAF211), were constructed by PCR amplifying the open reading frames and inserting them into p3XFLAG-CMV-14 (Sigma, St. Louis, MO). Point mutations in Arf4 (Arf4T31N = pJAF221, Arf4T48N = pJAF222, Arf4Q71L = pJAF223) were generated by inverse PCR using the Quick Change II site directed mutagenesis kit (Stratagene) starting from pJAF216. Cells were transfected with FLAG-tagged Arf and GFP-tagged CTS deletion and alanine scanning mutants used in [7], CD8-PKHD1 (pJAF268), or CD8-Megalin (pJAF281) and 48 hours later, FLAG immunoprecipitation was carried out as described in [40].
Embryos were fixed in 0.2% glutaraldehyde, 2% formalin, 5 mM EGTA and 2 mM MgCl2 in 0.1M phosphate buffer pH 7.3 for 10 minutes at room temperature then rinsed three times in wash buffer containing 0.1% sodium deoxycholate, 0.2% IGEPAL, 2 mM MgCl2 in 0.1M phosphate buffer for 30 minutes each wash. Fixed embryos were stained overnight at 37°C in 1 mg/ml X-gal, 5 mM potassium ferrocyanide, 5 mM potassium ferricyanide diluted in wash buffer.
RNA was isolated from individual E9.5 embryos or from IMCD3 cells using RNeasy kits (Qiagen), including on-column DNA digestion. First strand cDNA was synthesized from 100–500 ng of total RNA using a SuperScript II First-Strand Synthesis System (Invitrogen, Carlsbad, CA) and random hexameric primers. Quantitative real-time PCR primers were designed to produce amplicons between 100–150 nucleotides in length, using the online primer3 web PCR primer tool (http://fokker.wi.mit.edu/primer3/input.htm) and the IDT Primer Express software tool (http://www.idtdna.com/Scitools/Applications/Primerquest/). Primers were synthesized by Integrated DNA Technologies Inc (Coralville, IA) and are listed in Table 1. qRT-PCR analysis was performed using an ABI Prism 7500 sequence detection system (Applied Biosystems, Foster City, CA). Each reaction contained 5–12.5 ng first strand cDNA, 0.1 µM each specific forward and reverse primers and 1× Power SYBR Green (Applied Biosystems, Foster City, CA) in a 15 µl volume. Arf4 mRNA expression was normalized to GAPDH mRNA abundance and compared between mutant and control animals with an unpaired Student t-test.
|
10.1371/journal.ppat.1007464 | Immunogenic particles with a broad antigenic spectrum stimulate cytolytic T cells and offer increased protection against EBV infection ex vivo and in mice | The ubiquitous Epstein-Barr virus (EBV) is the primary cause of infectious mononucleosis and is etiologically linked to the development of several malignancies and autoimmune diseases. EBV has a multifaceted life cycle that comprises virus lytic replication and latency programs. Considering EBV infection holistically, we rationalized that prophylactic EBV vaccines should ideally prime the immune system against lytic and latent proteins. To this end, we generated highly immunogenic particles that contain antigens from both these cycles. In addition to stimulating EBV-specific T cells that recognize lytic or latent proteins, we show that the immunogenic particles enable the ex vivo expansion of cytolytic EBV-specific T cells that efficiently control EBV-infected B cells, preventing their outgrowth. Lastly, we show that immunogenic particles containing the latent protein EBNA1 afford significant protection against wild-type EBV in a humanized mouse model. Vaccines that include antigens which predominate throughout the EBV life cycle are likely to enhance their ability to protect against EBV infection.
| Human herpesviruses are tremendously successful pathogens that establish lifelong infection in a substantial proportion of the population. The oncogenic γ-herpesvirus EBV, like other herpesviruses, expresses a plethora of open-reading frames throughout its multifaceted life cycle. We have developed a prophylactic vaccine candidate in the form of immunogenic particles that contain several EBV antigens. This is in stark contrast to the vast majority of EBV vaccines candidates that contain only one or two EBV antigens. Our immunogenic particles were shown capable of stimulating several EBV-specific T-cell clones in vitro. The immunogenic particles were also capable of expanding cytolytic EBV-specific T cells ex vivo and provided a protective benefit in vivo when used as a prophylactic vaccine.
| The Epstein-Barr virus (EBV) is a γ-herpesvirus that establishes asymptomatic infection in the majority of the human population. EBV infects both B cells and epithelial cells, but it is in the former in which EBV establishes latency and persists lifelong [1]. Despite being carried asymptomatically by most individuals, the global disease burden of EBV is substantial. EBV is the primary cause of infectious mononucleosis (IM), accounts for 200,000 new cancer cases annually [2] and is linked to the development of autoimmune diseases (e.g. multiple sclerosis) [3].
Shortly after the discovery of EBV, vaccination was touted as a possible means of controlling or eliminating EBV-associated diseases [4]. Despite EBV being the first human oncogenic virus to be discovered, and in spite of several decades of EBV vaccine research, no prophylactic EBV vaccine has made it onto the market. So far, the majority of prophylactic vaccine prototypes have focused on the major viral envelope glycoprotein gp350 [5]. One study, in which soluble gp350 was used for vaccination purposes, reported a decrease in the frequency of IM in vaccinated individuals over a study period of 18 months, but vaccination did not reduce the frequency of infection with the wild type virus [6]. Long-term information on the vaccinated cohort is not available.
Herpesviruses have complex life cycles and primary infection, latency and reactivation are achieved through the expression of a large number of open-reading frames [7–9]. Considering the number of antigens that are expressed during the EBV life cycle, it is not surprising that EBV infection is controlled in healthy individuals through humoral and cellular immune responses that target a variety of lytic and latent proteins [10]. Considering the breadth of EBV-specific immune responses in healthy individuals, it is surprising that EBV vaccine prototypes have only targeted a limited set of lytic [6,11,12] or latent proteins [13].
We previously generated a vaccine candidate in the form of EBV virus-like particles (VLPs) and light particles (LPs) [14]. Deletion of EBV proteins involved in DNA packaging produced particles that were DNA-free, non-infectious and highly immunogenic. Whilst EBV VLPs/LPs are likely to contain several dozen lytic proteins (viz. envelope, tegument and capsid proteins) [15], they are devoid of latent proteins. To address this shortcoming, we enlarged the antigenic spectrum of VLPs/LPs to include immunodominant latent proteins. We interrogated the antigenicity of VLPs/LPs containing latent antigens in vitro, ex vivo and in vivo.
To generate immunogenic particles with an enlarged antigenic spectrum, we aimed to introduce latent protein fragments into EBV VLPs/LPs. Since BNRF1 is abundant within virions [15], we rationalized that latent antigens could be introduced into VLPs/LPs by fusing them to BNRF1 (Fig 1A). However, since wild-type EBV (wtEBV) can be accurately and sensitively quantified through qPCR [14], we first modified BNRF1 of wtEBV to test this assumption. The BNRF1 of wtEBV was modified to contain a fragment from the highly antigenic latent protein EBNA3C [16] (S1 Fig). Bacterial artificial chromosome (BAC) DNA from wtEBV was modified to contain EBNA3C and then stably introduced into 293 cells to generate a virus producer cell line (293/EBV-E3C). The integrity of the EBV-E3C BAC DNA within producer cells was confirmed with restriction analysis (S1 Fig). Transfection of the lytic transactivator BZLF1 gene into 293/EBV-E3C and 293/wtEBV yielded a similar percentage of cells that expressed the late lytic protein gp350 (Fig 1B). This indicates that modification of BNRF1 to include a latent antigen fragment does not influence lytic replication. Next, we compared the antigenicity of EBV-E3C and wtEBV viruses in T-cell activation assays (Fig 1C). Autologous lymphoblastoid cell lines (LCLs) were pulsed with the two viruses or peptide controls and then cocultured with BNRF1- [17] or EBNA3C- [18] specific CD4+ T cells. This confirmed that modified virions, containing a BNRF1-EBNA3C fusion protein, were able to simulate EBNA3C- and BNRF1-specific CD4+ T cells. Conversely, wtEBV that contained unmodified BNRF1 was only able to stimulate the BNRF1-specific CD4+ T cells (Fig 1C). In all cases, the dose of the virus applied correlated to the response generated by the T cells, with as little as 1 x 104 virions (genome equivalents (geq)) being able to generate responses from the BNRF1- and EBNA3C-specific T cells. Importantly, BNRF1-specific CD4+ T cells recognized modified and unmodified EBV to the same extent. This indicates that BNRF1-latent antigen fusion proteins enlarged the antigenic spectrum of EBV without influencing the antigenicity of BNRF1. Next, we tested whether the enlarged antigenic spectrum of EBV-E3C was exclusively due to BNRF1-latent antigen fusions contained within virions. To this end, virus supernatants were pre-incubated with anti-gp350 neutralizing antibody [19] prior to being used in T-cell recognition assays (Fig 1D). This showed that the neutralizing antibody was able to abolish the antigenicity of EBV-E3C. Altogether, these results confirm that BNRF1-latent antigen fusion proteins are successfully packaged into virions and enlarge their antigenic spectrum.
Next, we confirmed the antigenicity of BNRF1-latent antigen fusion proteins in gp110-negative VLPs/LPs. Since gp110 has been shown to preclude viral and host membrane fusion [20], and abrogate toxicity [21], we exclusively used gp110-negative VLPs/LPs in the present study. We concurrently modified VLPs/LPs and wtEBV to encode EBNA3C and EBNA1 fragments, respectively generating 293/VLPs/LPs-E3C-E1 and 293/EBV-E3C-E1 producer cells (S2 Fig). EBNA1, like EBNA3C, is a highly immunogenic latent protein that is frequently recognized by the population [16,22]. The DNA-free VLPs/LPs-E3C-E1 were quantified using flow cytometry (S3 Fig) and then compared to an equivalent amount of EBV-E3C-E1 that was quantified with qPCR (Fig 2A). This confirmed that flow cytometry enabled the reliable quantification DNA-free VLPs/LPs. Next, VLPs/LPs-E3C-E1 were analysed in T-cell activation assays alongside EBV-E3C-E1 (Fig 2B). This showed that the VLPs/LPs, like EBV virions, were able to stimulate BNRF1- [17], gp350- [23], EBNA3C- [18] and EBNA1- [24] specific CD4+ T cells when they were modified to contain EBNA3C and EBNA1 fragments (Fig 2B). Furthermore, the modified VLPs/LPs stimulated the various lytic protein- and latent protein-specific T cells to the same extent as modified EBV. This confirmed that VLPs/LPs could be used as a platform to generate immunogenic particles that comprise lytic and latent antigens. Furthermore, the lack of gp110 does not negatively influence the antigenicity of the VLPs/LPs, indicating that their safety can be increased without compromising their antigenicity.
Since EBV-specific T cells play a crucial role in controlling EBV-infection [25,26], we tested whether modified VLPs/LPs could expand EBV-specific T cells with protective value. To this end, epitope-rich regions from EBNA1, arbitrarily named region I, region II and region I:II (Fig 3A), were used to generate VLPs/LPs producer cells that encode EBNA1 (S4 Fig). Analysis of the producer cells with western blot showed that the 293/VLP/LP-EBNA1RI:II producer cell line was unable to express the large BNRF1-EBNA1 fusion, whilst the 293/VLP/LP-EBNA1RI and 293/VLP/LP-EBNA1RII producer cells successfully expressed their BNRF1-EBNA1 fusions (Fig 3B). Hence VLPs/LPs-EBNA1RI:II were excluded from analysis. VLPs/LPs-EBNA1RI and VLPs/LPs-EBNA1RII were combined (VLPs/LPs-EBNA1RI+RII) and used to stimulate bulk PBMCs from unhaplotyped EBV-positive donors (Fig 3C). As a control, PBMCs from the same donors were also expanded with an antigen-armed antibody (AgAb) that contained the major EBV glycoprotein gp350. AgAbs were originally developed as a targeted therapy for B cell malignancies [18,27], but were repurposed in the present study to expand EBV-specific T cells of interest (S5 Fig). Stimulation of PBMCs from EBV-positive donors with VLPs/LPs-EBNA1RI+RII or gp350-AgAb expanded similar numbers of CD4+, CD8+ and total T cells from the PBMCs of EBV-positive donors (Fig 3C). Next, primary B cells from four donors were infected with highly infectious gp110high B95-8 [28] and then cocultured with VLPs/LPs-EBNA1RI+RII- or gp350-AgAb-expanded T cells. After 5 days, ex vivo cultures were analyzed by immunofluorescence for the presence of EBV-infected B cells (Fig 3D). This showed that nearly all the B cells in ex vivo cultures were EBNA2-positive, confirming that virtually all the B cells were successfully infected with EBV. However, there were noticeably fever EBV-infected B cells (CD20+EBNA2+) in the presence of VLPs/LPs-EBNA1RI+RII-specific T cells than gp350-specific T cells. This suggested that VLPs/LPs-EBNA1RI+RII-specific T cells were more efficient at controlling the EBV-infected B cells during the first five days of infection compared to gp350-specific T cells. Next, VLPs/LPs-EBNA1RI+RII- and gp350-AgAb-expanded T cells from eight donors were cocultured with infected B cells as before and quantitatively analysed with flow cytometry (Fig 3E and 3F). This confirmed that ex vivo cultures containing VLPs/LPs-EBNA1RI+RII-specific T cells had the lowest percentage of CD19+ cells. This indicates that VLPs/LPs-EBNA1RI+RII-specific T cells are more proficient than gp350-specific T cells at controlling EBV-infected B cells during the first five days of infection.
Having shown that VLP/LPs-EBNA1RI+RII-specific T cells efficiently control B95-8-infected B cells, we tested whether they could restrict the outgrowth of infected B cells over a longer period. Additionally, we tested whether VLP/LPs-EBNA1RI+RII-specific T cells could prevent the outgrowth of B cells infected with the prototypic B95-8 strain or the distantly related M81 strain [29] from Hong Kong. We stimulated PBMCs from eight EBV-positive donors as before (see Fig 3) and then cocultured them with autologous B cells that were infected with gp110high B95-8 and gp110high M81. As a positive and negative control, infected B cells were respectively cultured with CD19- PBMCs or in medium only. After 15 days, ex vivo cultures were analysed with flow cytometry to detect outgrowing B cells (Fig 4). Since proliferating B cells express CD23, outgrowing B cells were identified by detecting CD19+CD23+ double-positive cells [30,31]. EBV-infected B cells were found to consist of CD19+CD23-, CD19+CD23low and CD19+CD23high populations when they were cultured in medium only, with the majority of B cells being of the CD19+CD23high variety (Fig 4A). Comparatively, in the presence of CD19- PBMCs, gp350-specific T cells and VLPs/LPs-EBNA1RI+RII-specific T cells, the number of CD19+CD23+ cells were considerably reduced. This indicated that proliferating B cells were restricted in these cultures. Interestingly, whilst gp350-specific T cells were shown to be more efficient than CD19- PBMCs at controlling infected B cells during the early phase of infection (see Fig 3), the CD19- PBMCs of several donors were considerably more adept at restricting B-cell outgrowth than gp350-specific T cells over the longer 15 day period (Fig 4B). This suggest that the PBMCs from some donors contained EBV-specific T cells, other than gp350-specific T cells, that were able to restrict B-cell outgrowth. However, it is evident that proliferating B cells were restricted to a greater degree in ex vivo cultures that contained VLPs/LPs-EBNA1RI+RII-specific T cells. Moreover, this was observed for B95-8- and M81-infected B cells and for all donors (Fig 4B). This confirms that VLPs/LPs equipped with EBNA1 expand EBV-specific T cells that efficiently restrict B cells infected with B95-8 and M81 EBV.
Having shown that VLPs/LPs-EBNA1RI+RII-specific T cells control EBV-infected cells, it indicated that they were cytolytic in character. However, since VLPs/LPs-EBNA1RI+RII-stimulated PBMCs contained primarily CD4+ T cells (Fig 3C), it was unclear whether VLPs/LPs-EBNA1RI+RII are capable of stimulating EBV-specific CD8+ T cells. To this end, we tested the ability of LCLs pulsed with VLPs/LPs-EBNA1RI, containing the EBNA1 HPV CD8+ epitope, to stimulate an EBNA1 HPV-specific T-cell clone [18]. This showed that the EBNA1 HPV-specific CD8+ T-cell clone was unable to recognize LCLs pulsed with VLPs/LPs-EBNA1RI (S6A Fig). In contrast, LCLs pulsed with the VLPs/LPs-EBNA1RI were well recognized by the BNRF1 VSD-specific CD4+ T-cell clone. This suggests that the autologous LCLs were unable to cross-present the EBNA1 HPV epitope from VLPs/LPs-EBNA1RI to CD8+ T cells. We also assessed IFN-gamma secretion by CD4+ and CD8+ T cells after the stimulation of bulk PBMCs with VLPs/LPs-EBNA1RI+RII (S6B Fig). This revealed that both CD4+ and CD8+ IFN-gamma producing cells were detected upon stimulation with VLPs/LPs-EBNA1RI+RII (S6C Fig). However, it was evident that IFN-γ+ CD4+ T cells were more numerous than IFN-γ+ CD8+ T cells and suggests that the stimulation of PBMCs with VLPs/LPs-EBNA1RI+RII preferentially expands EBV-specific CD4+ T cells. Next, we expanded EBNA1- and gp350-specific CD4+ T cells from VLPs/LPs-EBNA1RI+RII-stimulated PBMCs and analyzed them for their cytotoxic potential. Bulk PBMCs from an unhaplotyped EBV-positive donor was stimulated with VLPs/LPs-EBNA1RI+RII for two rounds, after which gp350-AgAb (S5 Fig) or EBNA1-AgAb (S7 Fig) were used to expand gp350- and EBNA1-specific CD4+ T cells (Fig 5A). The expanded CD4+ T cells were confirmed to be specific for either EBNA1 or gp350 (Fig 5B). The ex vivo expanded CD4+ T cells specifically responded to EBNA1-AgAb or gp350-AgAb and to EBNA1 3G2 or gp350 1D6 epitope peptides. Next, we determined whether the EBNA1- and gp350-specific CD4+ T cells were capable of expressing CD107a, a surrogate marker for the release of cytolytic granules [32]. Autologous LCLs were pulsed with α-CD20, EBNA1-AgAb or gp350-AgAb then cocultured with the EBNA1- and gp350-specific CD4+ T cells. This showed that both CD4+ T-cell lines upregulated CD107a in response to the relevant antigen (Fig 5C). However, approximately 50% of gp350-specific CD4+ T cells expressed CD107a, whilst only 10% of EBNA1-specific CD4+ T cells expressed CD107a. Next, we tested the ability of EBNA1- and gp350-specific CD4+ T cells to release the mediator of cytolysis granzyme B (Fig 5D). Both the EBNA1- and gp350-specific CD4+ T cells released granzyme B in response to the relevant AgAb and epitope peptide. Lastly, we tested whether the EBNA1- and gp350-specific CD4+ T cells were capable of directly lysing autologous LCLs pulsed with antigen (Fig 5E). This showed that the both the EBNA1- and gp350-specific CD4+ T cells specifically lysed LCLs pulsed with epitope peptides and VLPs/LPs that contained EBNA1. Taken together, these results confirm that VLPs/LPs-EBNA1RI+RII have the ability to stimulate cytolytic CD4+ T cells specific for lytic and latent antigens. These results are consistent with previous studies that showed EBNA1- [30,33,34] and gp350- [35] specific T cells to be cytolytic.
Having shown that modified VLPs/LPs were antigenic in vitro and ex vivo, we assessed whether VLPs/LPs- EBNA1RI+RII had protective abilities in vivo. To this end, mice reconstituted with human immune system components, susceptible to EBV infection and capable of exerting EBV-specific immune control [36], were used to interrogate VLPs/LPs- EBNA1RI+RII. Humanized NSG-A2 (huNSG-A2) mice were randomly grouped and injected intraperitoneally with PBS, unmodified VLPs/LPs (1 x 106 particles) or VLPs/LPs-EBNA1RI+RII (1 x106 particles), using poly (I:C) as an adjuvant (Fig 6A). Four weeks later, mice were boosted using the same dose. Animals were challenged with gp110high B95-8 (1 x 105 GRUs) six weeks after the last boost and euthanized eight weeks later. From the literature we knew that this titer would enable infection without gross development of tumors [36]. The spleens of challenged animals were analysed by histology (Fig 6B). This showed that all animals contained human CD20- and CD3-positive cells in their spleens. However, there was no correlation between the abundance of CD20- and CD3-positive cells and the different treatments. In situ hybridization revealed the presence of interspersed cells that expressed EBV-encoded RNAs (EBERs) in the spleens of mice from the PBS and unmodified VLPs/LPs groups. In total, 60% of mice from the PBS group were found to contain EBER+ cells, while 37.5% of the mice from the VLPs/LPs group contained EBER+ cells (Fig 6C). Statistical analysis showed that this observation was statistically insignificant (P > 0.05). None of the spleen samples from the VLPs/LPs-EBNA1RI+RII group were found to contain EBER+ cells. This result was confirmed to be statistically significant from the PBS (P = 0,009) and VLPs/LPs (P = 0.035) groups. Next, qPCR was used to detect the presence of EBV in the peripheral blood of challenged animals (Fig 6D). This showed that 100% of mice from the PBS group contained EBV DNA in their peripheral blood, compared to 62% of the VLPs/LPs group and 14% of the VLPs/LPs-EBNA1RI+RII group. Once more, statistical analysis revealed that the observed difference between the PBS and VLPs/LPs group was not significant (P > 0.05). However, statistical analysis showed that the difference between the VLPs/LPs-EBNA1RI+RII group and the PBS (P = 0.0017) and VLPs/LPs (P = 0.0286) groups was significant. In summary, this indicates that vaccination with VLPs/LPs-EBNA1RI+RII afforded significantly better protection during the eight-week challenge period than vaccination with unmodified VLPs/LPs. However, whether vaccination with VLPs/LPs-EBNA1RI+RII would offer long-term protection against EBV infection in humans remains unknown. Nonetheless, our results bode well for the development of second generation VLPs/LPs that contain multiple latent antigens.
DNA-free EBV VLPs/LPs are structurally complex, composed of multiple lytic gene products, incapable of infection and highly immunogenic [14]. Considering the breadth of EBV-specific immune responses in healthy individuals [10], it is likely that prophylactic vaccination would benefit from an immunogen that contains EBV-antigens that are expressed during lytic replication and latency. To this end, we used EBV VLPs/LPs as a platform to generate immunogenic particles that comprise lytic and latent antigens. Considering the importance of T-cell responses in controlling of EBV infection [37], we extensively assessed the ability of modified VLPs/LPs to stimulate EBV-specific T cells.
Modification of EBV VLPs/LPs to contain latent antigens successfully enlarged their antigenic spectrum, enabling the stimulation of several lytic protein- and latent protein-specific CD4+ T-cell clones. The antigenicity of EBV VLPs/LPs containing EBNA1 were also analyzed in an ex vivo setting by stimulating bulk PBMCs from EBV-positive donors to yield EBV-specific T-cell lines. Ex vivo generated T-cell lines contained EBV-specific CD4+ and CD8+ T cells, but it was evident that CD4+ T cells were more numerous. The preferential expansion of CD4+ T cells from bulk PBMCs is supported by earlier studies which have shown that EBV structural proteins are weakly immunogenic towards CD8+ T cells [38,39] and are instead immunodominant targets of CD4+ T cells [26]. The ability of B cells to efficiently present several structural proteins from incoming virions to CD4+ T cells [35,40] suggests that structural protein-specific CD4+ T cells are likely to play a more dominant role than CD8+ T cells during the very early stages of infection. However, the recent identification of BcLF1-specific T-cell clones that can recognize LCLs pulsed with UV-inactivated EBV [41] suggests that structural protein-specific CD8+ T cells may also play an important role during the early stage of infection. In contrast, we found that LCLs pulsed with modified EBV VLPs were unable to stimulate an EBNA1-specific CD8+ T-cell clone, implying that cross-presentation might be limited to particular structural antigens. Future studies will certainly shed light on this newly discovered phenomenon.
We have shown that T cells expanded with modified VLPs are considerably more efficient at controlling recently infected B cells compared to those expanded with gp350. Since VLPs/LPs contain multiple EBV antigens, they likely expanded a polyclonal pool of EBV-specific T cells that simultaneously recognize different antigenic epitopes displayed by infected B cells. Indeed, recently infected B cells have been shown to display multiple structural proteins [35,40]. The inability of gp350-specific T cells to competently target recently infected B cells has profound implications for prophylactic vaccine design, since T-cell responses would be of paramount importance if a small number of EBV virions are not neutralized and manage to infect a subset of cells. The major envelope glycoprotein gp350, arguably the most intensively studied EBV antigen, has been investigated as a prophylactic vaccine for several decades and continues to garner attention [6,11,12,42,43]. The continued focus on gp350 is sensible considering gp350 is chiefly responsible for attachment to B cells during the infection process and is a frequent target of neutralizing antibodies [44]. Indeed, emerging gp350 vaccines have improved in their ability to generate neutralizing antibodies [12,43]. However, the ability of gp350-specific T cells to provide optimal protection against recently EBV-infected cells has previously not been seriously questioned. Our analyses suggest that gp350-specific T cells alone are suboptimal for targeting EBV-infected B cells during the early phase of infection. This implies that prophylactic vaccines are more likely to generate protective T-cell responses during the early phase of EBV-infection if they include multiple lytic antigens. Additionally, prophylactic vaccines composed exclusively of gp350 would not elicit antibody responses that protect against epithelial cell infection. EBV infects epithelial cells independently of gp350 [45], a process recently shown to rely on gH/gL [46,47]. From this perspective, a vaccine composed of gH/gL has an advantage over gp350 since it may prevent epithelial and B cell infection [48]. Since VLPs/LPs contain both gp350 and gH/gL, along with several other EBV glycoproteins, they are likely to generate antibody responses that protects against B-cell and epithelial cell infection. Whilst we did not focus on antibody responses in the present study, the structural similarity of VLPs/LPs to wtEBV suggests they would enable the generation of neutralizing antibodies to several EBV glycoproteins in their natural context. Indeed, UV-irradiated EBV has previously been shown to generate potent neutralizing antibodies [43]. From this perspective, prophylactic vaccines that contain numerous EBV antigens, including glycoproteins that mediate B-cell and epithelial cell infection, are likely to stimulate superior protective antibody and T-cell responses.
Humoral and cellular immune responses that recognize multiple structural proteins are likely to provide a protective benefit by targeting virions, recently infected cells and cells undergoing reactivation [35,49,50]. However, they are unlikely to offer protect against latently infected cells. The establishment of type I, II or III latency in just a handful of cells would enable the number of EBV-infected cells to increase through simple cell division [51]. The inability of vaccination to protect against latently infected cells is especially important to consider since it is latency transcription programs that predominate during IM, EBV-associated lymphoproliferative disease and EBV-positive malignancies [52]. The ability of EBV-specific T cells to control EBV-infected cells and prevent post-transplant lymphoproliferative disease (PTLD) strongly suggests that prophylactic vaccination should elicit latent protein-specific T cells in addition to lytic protein-specific T cells [53–55]. Since the mouse challenge model in the present study extended for eight weeks, the antigenic load of structural proteins are likely to have been markedly reduced relative to that of latent proteins. Vaccination of humanized mice with VLPs containing EBNA1 afforded more protection against EBV infection than vaccination with unmodified VLPs/LPs. Since humanized mice produce little or no EBV-specific antibodies [36,56], it likely that protection against EBV infection did not involve humoral immunity. Rather, it is more probable that vaccination with VLPs/LPs-EBNA1RI+RII elicited protective T cell responses. The recognition of structural proteins and EBNA1 possibly enabled the humanized mice to target EBV-infected cells during the whole EBV life cycle (except latency 0). Indeed, EBNA1 is the only latent protein that is expressed during all forms of EBV latency (except type 0) and in all EBV-positive malignancies [30]. Whether vaccination with lytic and latent EBV antigens can afford protection against EBV in humans remains to be determined.
Prototype EBV vaccine candidates have thus far not directly addressed the daunting task of affording protection against the plethora of EBV strains that exist worldwide. In recent years, the advent of high throughput sequencing has enabled the detailed analysis of various EBV strains [57]. This has revealed that EBV-encoded proteins can be highly polymorphic. Recently, T-cell epitopes from several EBV strains have been compared [58]. This revealed that epitopes differ considerably between different EBV strains, with almost 50% of CD4+ epitopes and almost 30% of CD8+ epitopes, including their flanking region, varying between the B95-8 and M81 strains. Whilst EBV strain heterogeneity presents a hurdle for T-cell immunity, our results show that polyclonal memory T cells stimulated with modified B95-8 VLPs/LPs can restrict B cells transformed with B95-8 and M81 at a relatively early stage of infection. However, it is unknown whether modified B95-8 VLPs/LPs would afford protection against M81 in vivo. Since latent genes are more polymorphic than lytic genes, it is possible that B95-8 latent antigens would not afford optimal protection against those encoded by other EBV strains. The identification of T-cell epitopes that are conserved amongst different EBV strains would certainly aid vaccine design efforts, enabling this hurdle to be overcome. Nonetheless, a vaccine consisting of multiple EBV proteins, possibly from different strains, is also likely to combat the problem of EBV strain heterogeneity.
EBV VLPs/LPs are likely to be further improved as a prophylactic vaccine through the inclusion of antigens in addition to EBNA1. Several interesting candidates, latent and lytic proteins, have recently been highlighted [40]. Modification of multiple tegument proteins will enable the incorporation of several antigens into EBV VLP/LPs. By incorporating the most immunodominant EBV antigens into VLPs/LPs it will enable them to prime the immune system against viral antigens that are expressed at all stages of infection and in all types of EBV-associated tumors, whilst enabling the generation of neutralizing antibodies against surface viral proteins. Such a multipronged approach is likely to increase the protection afforded by the EBV VLPs/LPs.
Peripheral blood mononuclear cells (PBMCs) were isolated from healthy donors that provided written informed consent (ethical approval granted by the Ethikkommission of the Medizinische Fakultät Heidelberg (S-603/2015)) or from anonymous buffy coats purchased from the Institut für Klinische Transfusionsmedizin und Zelltherapie (IKTZ) in Heidelberg and did not require ethical approval. Animal experiments were approved (approval number G156-12) by the federal veterinary office at the Regierungspräsidium Karlsruhe (Germany) and were performed in strict accordance with German animal protection law (TierSchG). Mice were handled in accordance with good animal practice, as defined by the Federation of European Laboratory Animal Science Associations (FELASA) and the Society for Laboratory Animal Science (GV-SOLAS), and were housed in the class II containment laboratory of the German Cancer Research Center.
Cell lines include EBV-positive Raji cells (ATCC CCL-86) [59], EBV-negative Elijah B cells (kindly provided by Prof. A.B. Rickinson), 293 cells (ATCC CRL-1573) [60], T cells specific for EBNA1 3E10, EBNA3C 5H11, gp350 1D6 and BNRF1 VSD epitopes (kindly provided by Prof. J. Mautner) and autologous LCLs (kindly provided by Prof. J. Mautner). Peripheral blood mononuclear cells (PBMCs) were isolated using Ficoll-Paque Plus and primary B cells were isolated using Dynabeads CD19 Pan B (Invitrogen) and DETACHaBEAD CD19 kit (Invitrogen). RPMI containing 10% fetal calf serum (F7524, Sigma) was used to culture 293, Raji and Elijah cells. T-cell clones and lines were cultured as previously described [17,35].
AgAbs were constructed using sequences coding for EBNA1 (390–622 aa) and gp350 (1–470 aa). Latent protein-coding sequences were PCR amplified and introduced downstream of an α-CD20 HC gene contained within the pRK5 expression vector [18]. The primers used to construct AgAbs are listed in S1 Table. The α-CD20 antibody and AgAbs were produced by transfecting the appropriate heavy chains and the α-CD20 light chain into 293 cells using polyethylenimine (PEI). The following day the PEI-containing medium was removed and replaced with serum-free FreeStyle 293 expression medium and cells were incubated for three days. Supernatants were centrifuged at 400 x g for 10 minutes and filtered through a 0,22 μm filter.
Recombinant BAC DNA was constructed using galK recombination [61]. In the present study, wtEBV (B95-8) [62] or VLPs/LPs (B95-8ΔBFLF1/BFRF1a/BALF4) [21] BAC DNA were modified to encode latent protein fragments. Only VLPs/LPs lacking BALF4, encoding the glycoprotein gp110, were utilized in the present study due to their enhanced safety [20,21]. The primers used for the construction of BAC mutants, as well as a description of all BAC mutants, are shown in S1 Table. The first step in galK recombination was the insertion of the galK cassette into the BNRF1 ORF of wtEBV or VLPs/LPs BAC DNA. Subsequently, the galK cassette was replaced with DNA fragments encoding latent protein moieties. Outgrowing colonies were analysed with restriction digestion and sequencing to confirm the integrity of BNRF1-latent protein fusions. Stable producer cells were generated with the recombinant BAC DNA as previously described [63].
An expression plasmid encoding BZLF1 (p509) was transfected into producer cells to induce virus or VLPs/LPs production [14]. For the production of EBV (B95-8 or M81) for ex vivo and in vitro studies, the pRA plasmid encoding gp110 was cotransfected with p509 for increased infectivity [28]. The liposome-based transfectant Metafectene (Biontex) was used to carry out transfections overnight. Subsequently, Metafectene-containing medium was removed and replaced with fresh medium. Transfected cells were incubated for three days before supernatants were harvested. Supernatants were centrifuged at 400 x g for 10 minutes and filtered through a 0.44 μm filter. VLPs/LPs used for ex vivo T-cell expansions and animal experiments were produced in serum-free FreeStyle 293 expression medium (Gibco). In all other cases, virus and VLPs/LPs were produced on RPMI supplemented with 10% FCS. Lastly, virus and VLPs/LPs used in animal experiments were concentrated at 18 000 x g for 3 hours and resuspended in PBS.
Virus titers were determined by real-time qPCR as previously described [14]. In brief, virus-containing supernatants were treated with DNase I (5 units) and proteinase K (1mg/mL). Next, real-time qPCR analysis was carried out using primers and probe specific for the EBV BALF5 gene. To determine the presence of EBV in peripheral blood, genomic DNA from vaccinated and challenged animals was compared to unchallenged animals.
wtEBV, previously quantified with real-time qPCR, was titrated (1, 0.75, 0.5, and 0.25 x 107 geq) and bound to Elijah B cells at 4°C. Cells were washed, stained with α-gp350 (clone 72A1) and α-mouse IgG-Cy3 antibodies and analysed with flow cytometry. MFI values were determined for different amounts of virus. A standard curve was generated for EBV genomes vs MFI. Concurrently, supernatants containing VLPs/LPs were incubated with Elijah B cells and stained as above. MFI values obtained for VLPs/LPs were extrapolated off the standard curve to quantify VLPs/LPs.
IFN-γ in cell culture supernatants was determined as previously described [18]. Autologous LCLs were pulsed overnight with antigen and cocultured for a minimum of 18 hours with T-cells at an E:T ratio of 1:1. Supernatants were analysed by ELISA (Mabtech). In blocking studies with neutralizing antibody (72A1 clone), virus containing supernatants were preincubated with antibody for 1 hour at 37°C before being used in T-cell activation assays.
Bulk PBMCs from EBV-positive donors were pulsed with VLPs/LPs-EBNA1 (1 x 106 particles) or gp350-AgAb (20 ng). After two days, cultures were supplemented with IL-2 (10 U/mL) and thereafter maintained in medium containing IL-2. Cells were restimulated 10 days later using IL-2 (10 U/mL) and the same amount of VLPs/LPs-EBNA1 or gp350-AgAb. One week later, cells were analysed for the presence of CD4, CD8 and CD3 expressing cells or where cocultured with primary B cells that were infected overnight with EBV.
Bulk PBMCs from four EBV-positive donors were stimulated for two rounds with VLPs/LPs-EBNA1 (1 x 106 particles) or gp350-AgAb (20 ng) in the presence of IL-2 (10 U/mL). Autologous primary B cells were infected overnight with B95.8 (MOI = 3) and then cocultured with the stimulated PBMCs, CD19- depleted PBMCs or medium only. Ex vivo cultures were analysed with flow cytometry and immunofluorescence 5 days post-infection to observe EBV-positive cells. Cells were stained with α-CD19-APC (HIB19 clone) prior to flow cytometry and α-CD20 (L26 clone), α-EBNA2 (PE2 clone) and DAPI prior to immunofluorescence.
Bulk PBMCs from eight EBV-positive donors were stimulated for two rounds with VLPs/LPs-EBNA1 (1 x 106 particles) or gp350-AgAb (20 ng) in the presence of IL-2 (10 U/mL). B cells were infected with B95-8 or M81, respectively using an MOI of 3 or 30 to account for their different transforming abilities [29]. Ex vivo cultures stained with α-CD19-APC (HIB19 clone) and α-CD23-PE-Cy7 (EBVCS2 clone) antibodies and analysed by flow cytometry.
PBMCs from an EBV-positive donor were stimulated for one round with VLPs/LPs-EBNA1RI+RII (1 x 106 particles) in the absence of IL-2. After two weeks, cells were restimulated using irradiated (40 Gy) autologous PBMCs, pulsed with the same dose of VLPs/LPs-EBNA1RI+RII, in the presence of IL-2. After another two weeks, EBNA1- or gp350-specific T cells were expanded by stimulating cells biweekly with AgAbs (10–50 ng) that contained EBNA1 or gp350. Autologous LCLs, generated using B95-8ΔZR, were used as antigen presenting cells after the fifth round of stimulation. T cells were maintained in AIM V medium supplemented with 10% pooled human serum, IL-2 (10 U/mL), 10 mM HEPES, 2 mM L-glutamine, 50 μg/mL gentamicin and 0.4 mg/ mL ciprofloxacin.
Bulk PBMCs (5 x 106 cells) from EBV-positive donors were pulsed with VLPs/LPs-EBNA1RI+RII (1 x 106 particles) and supplemented with IL-2 (10U/mL) two days later. After another six days, cultures were restimulated with medium, EBNA1 peptide (PepTivator, Miltenyi), gp350-AgAb (20 ng) or VLPs/LPs-EBNA1RI+RII (1 x 106 particles) in the presence of bredfeldin A (Biolegend) for 4,5 hours. Cells were stained with α-CD4-APC (RPA-T4 clone) and α-CD8-PE-Cy7 (RPA-T8), fixed/permeabilized (Fixation/Permeabilization solution kit, BD Biosciences), stained with α-IFN-γ-PE (B27 clone) and analysed by flow cytometry.
NSG-A2 mice (NOD.Cg-PrkdcscidIl2rgtm1WjlTg (HLA-A2.1) 1Enge/SzJ) were humanized with CD34+ hematopoietic progenitor cells (HPCs) that were isolated from human fetal liver tissue (Advanced Bioscience Resources, USA) [64]. Newborn mice were irradiated (1 Gy) and injected intrahepatically with CD34+ HPCs. After twelve weeks, the presence of human CD45+ cells in the peripheral blood of mice was determined to confirm successful humanization. In total, 20 humanized NSG-A2 (huNSG-A2) mice were randomly grouped according to similarity of humanization ratios and injected intraperitoneally in a single blind fashion with PBS, VLPs/LPs (1 x 106 particles) or VLPs/LPs-EBNA1RI+RII (1 x 106 particles). In all cases, 50 μg poly (I:C) was used as adjuvant. Animals were boosted one month later with the same treatments. One and a half months after the boost, animals were injected intraperitoneally with 1 x 105 GRUs of B95-8. Mice were sacrificed eight weeks post-infection and their blood and tissues analysed for evidence of EBV infection [64]. All the VLPs/LPs and virus used in animal experiments were obtained by centrifuging supernatants at 18 000 x g for 3 hours and resuspending in PBS.
|
10.1371/journal.ppat.0030059 | A Novel Linear Plasmid Mediates Flagellar Variation in Salmonella Typhi | Unlike the majority of Salmonella enterica serovars, Salmonella Typhi (S. Typhi), the etiological agent of human typhoid, is monophasic. S. Typhi normally harbours only the phase 1 flagellin gene (fliC), which encodes the H:d antigen. However, some S. Typhi strains found in Indonesia express an additional flagellin antigen termed H:z66. Molecular analysis of H:z66+ S. Typhi revealed that the H:z66 flagellin structural gene (fljBz66) is encoded on a linear plasmid that we have named pBSSB1. The DNA sequence of pBSSB1 was determined to be just over 27 kbp, and was predicted to encode 33 coding sequences. To our knowledge, pBSSB1 is the first non-bacteriophage–related linear plasmid to be described in the Enterobacteriaceae.
| Flagella are whip-like structures found on the surface of bacterial cells that mediate swimming. Flagella contain a protein called flagellin, which is recognised as a danger signal by the immune system. Salmonella Typhi, the bacteria that causes typhoid fever, normally have flagella called H:d, but some strains only from Indonesia express distinct flagella, called H:z66. In this study we have located and sequenced the genes responsible for expressing these alternative flagella. Remarkably, these genes are located on a linear plasmid, an extra-chromosomal element that we have named pBSSB1. The significance of this finding is that linear plasmids are relatively common in bacterial species such as Streptomyces and Borrelia. However, such a linear element has never previously been described in enteric bacteria such as Escherichia coli and Salmonella. The identification of this novel linear plasmid in genetically tractable bacteria will facilitate future studies on the biology of linear plasmids and the pathogenicity of both flagella and Salmonella Typhi.
| Flagella play a critical role in the lifestyle of many bacteria, and the flagellin subunit is an important target for pathogen recognition by the mammalian innate immune system through Toll-like receptor (TLR) 5 [1]. Antiserum against flagella (H antigen) and lipopolysaccharide (O antigen) are the cornerstone of Salmonella classification through the Kauffmann–White scheme [2], which divides Salmonella enterica into the various serovars. The majority of S. enterica serovars are biphasic, alternating expression between two flagellar antigens through a process called “phase variation” [3]. Only one of the two flagellin genes, fliC and fljB, which are located at distinct loci on the Salmonella chromosome, is expressed at any given time [4]. In contrast, S. enterica serovar Typhi (S. Typhi), the cause of the human systemic infection known as typhoid, is normally monophasic, harbouring only the phase 1 flagellin gene fliC, which encodes the H:d antigen, and lacking a fljB equivalent. However, some S. Typhi strains isolated in Indonesia express an alternative H antigen, known as H:j, and/or a second flagellin, called H:z66 [5]. Whilst H:j variants arise through a deletion within the fliC gene [6], the H:z66 antigen is encoded on an unlinked locus.
In 1981, Guinee et al. [7] described S. Typhi strains from Indonesia that were H:d and H:j negative but motile due to functional z66 flagella. Upon incubation with anti-z66 antiserum, these strains reverted to H:d or H:j, so the z66 antigen was presumed to be a phase 2 flagellum. Despite extensive screening of worldwide S. Typhi strain collections, the z66 antigen has only been detected in strains originating from Indonesia [5,8–10]. The z66 flagellin structural gene (fljBz66) (1,467 base pairs [bp]) has been previously isolated on a 3,325-bp DNA fragment cloned from a z66+ strain of S. Typhi [11]. The same DNA fragment also encoded a putative phase 1 flagellin repressor (fljA) downstream of fljBz66, but the upstream region had no similarity to the site-specific inversion region of hin, which is associated with phase switching [12]. The genetic location of the gene encoding the z66 antigen (fljBz66) was not identified, and the DNA sequences beyond the original 3,325-bp cloned fragment have not previously been described. Here, we demonstrate that the fljBz66 gene is located on a novel 27-kbp linear plasmid that has been isolated, sequenced, and shown to be capable of autonomous replication in Escherichia coli.
S. Typhi strains isolated in Indonesia were tested for H:z66 antigen expression and for motility in soft agar. Two highly motile H:z66+ S. Typhi strains, In20 and 404Ty, were selected for further investigation. Despite DNA encoding fljBz66 being readily cloned, it was not possible to map the gene onto the S. Typhi chromosome using Southern blotting following I-CeuI digestion (unpublished data). Subsequent analysis of DNA on agarose gels identified two candidate plasmids in S. Typhi In20 and 404Ty, one migrating at a similar speed to the 36-kbp circular plasmid marker, and another more diffuse band migrating with the 63-kbp marker (Figure 1A, lanes 2 and 3). DNA prepared from S. Typhi Ty2 yielded no plasmid DNA, whereas the multi-drug–resistant S. Typhi CT18 gave two plasmids corresponding to the previously characterised pHCM1 (218 kbp) and pHCM2 (106 kbp) [13,14]. To determine whether the novel candidate plasmids present in the z66+ S. Typhi strains encoded the fljBz66 gene, a DNA fragment of the fljBz66 gene was used to probe plasmid DNA preparations from the S. Typhi strains (Figure 1B). No hybridisation was detected with the DNA prepared from S. Typhi CT18 or Ty2 (Figure 1B, lanes 5 and 6). The z66+ strains, 404Ty (lane 2) and In20 (lane 3), yielded signals corresponding to both of the bands. The smaller band was identified as the native plasmid, named pBSSB1. The larger band may be an artefact of denaturation/renaturation during the alkaline lysis procedure because it is not visible with pulsed field gel electrophoresis (PFGE), which does not involve alkaline lysis. We suggest that it may be a complex renaturation product, rather than a simple linear molecule.
We tested the hypothesis that pBSSB1 was a conventional circular plasmid, but despite intensive cloning and sequencing efforts, were unable to generate a complete circular DNA sequence. Consequently, genomic DNA of S. Typhi 404Ty was sequenced by 454 Pyrosequencing, generating an 8-fold coverage of the entire genome. DNA sequence information derived from the plasmid was combined with sequences obtained by conventional cloning. Analysis of the subsequent linear plasmid sequence confirmed that it encoded the fljBz66 gene and identified identical inverted repeat sequences present at both termini (terminal inverted repeats [tirs]) (Figures 2 and S1). Tirs are a common feature of linear plasmids in Streptomyces and Borrelia [15,16]. In Streptomyces, the size of the tirs varies from short palindromic repeats in SLP2 of S. lividans [17] to 95 kbp in plasmid pPZG101 of S. rimosus [18]. pBSSB1 has 1,230-bp tirs, with no similarity to other tirs and no direct, tandem, or palindromic repeats. The guanine-cytosine (GC) content of the tir regions, at 41%, is higher than the non-repetitive sequence. The linearity of the pBSSB1 sequence and the correct assembly of the tirs were confirmed by PCR (Figure S1).
pBSSB1 is 27,037 bp in length, which is 7 kbp less than the 34 kbp predicted by circular plasmid sizing on agarose gels (Figure 1A). Only the 3,456 bp of the region containing the fljBz66 sequence [11] exhibited strong similarity to previously sequenced DNA in public databases. Annotation predicted 33 coding sequences (CDSs) (Figure 2; Table S1), only three of which (030, fljBz66, and fljA) have been previously described [11]. The predicted coding density of the plasmid is one gene per 1.257 kbp (85.4%); this is similar to the chromosome and plasmids pHCM1 and pHMC2 from S. Typhi CT18 (87.6%, 83.8%, and 87.1%, respectively) [19]. The GC content of pBSSB1 is 36.6% (Figure 2). This is substantially lower than the GC content of genomes of enteric bacteria (∼50%–52%), suggesting that the plasmid may originate from a non-enteric source. Additionally, the sequence is not significantly similar to any previously described plasmids or bacteriophage. At the least stringent BLAST E-value applicable (<0.01), the sequence demonstrates no similarity to any described bacteriophage proteins. Moreover, of the 30 novel coding regions, 22 do not demonstrate any similarity to any previously described DNA or amino acid sequence in public databases. The 22 CDSs with no similarity to previously sequenced DNA are coloured green in Figure 2. The previously sequenced region includes the gene encoding the putative fliC repressor (fljA), the gene encoding the z66 flagellin antigen (fljBz66), and a gene of unknown function (030) immediately upstream of fljBz66. This region in pBSSB1 is 99% identical to the sequence produced by Huang et al. [11]. 030, fljBz66, and fljA are the final three CDS on the forward strand, adjacent to the inverted repeat (coloured yellow in Figure 2), with a combined GC content of 43%, implying that they may be a more recent acquisition by the element. Further analyses and annotation of the CDSs encoded on pBSSB1 are included in Table S1.
A change in GC skew ((G−C)/(G+C)) suggests that the region immediately upstream of 017 may act as a bi-directional origin of replication for pBSSB1 (distinguished by an asterisk in Figure 2). Changes in GC skew are often associated with the origin of replication on plasmids and bacterial chromosomes [20] and have been used previously to predict the internal origin of bi-directional linear replication [21]. The direction of transcription of the majority of genes is consistent with this. It is also known that linear Streptomyces plasmids with tirs, including pSLA2, replicate divergently from a central origin towards the termini [22]. No short DNA repeats, commonly associated with replication origins, were found within this region in pBSSB1. However, CDS 017 contains an ATP-binding motif similar to those found in partition proteins from some plasmids, but otherwise has no overall similarity to these proteins.
A kanamycin resistance marker was inserted within pBSSB1 at position 1,295 bp (indicated by a dagger in Figure 2; Figure S1) to facilitate experimental analysis. The plasmid with the kanamycin resistance cassette was named pBSSB2 and the modified S. Typhi In20 was named SGB32. Plasmid pBSSB2 was isolated from SGB32 and yielded a plasmid pattern indistinguishable from that of S. Typhi 404Ty and S. Typhi In20, despite the insertion of the 1,432-bp cassette (Figure 1A, lane 4). E. coli TOP10 cells were electrotransformed with purified pBSSB2 DNA isolated from S. Typhi SGB32, and kanamycin-resistant colonies were obtained. One kanamycin-resistant transformant was designated E. coli SGB33. Plasmid DNA from E. coli SGB33 (Figure 1A and 1B, lane 7) was indistinguishable from that of the S. Typhi strains harbouring pBSSB1, and subsequent Southern blotting confirmed the presence of the fljBz66 gene. pBSSB2 was stably inherited by E. coli SGB33, even in the absence of antibiotic selection. Expression of the z66 antigen could not be detected in E. coli SGB33 using Western blotting (unpublished data). We hypothesised that undetectable z66 antigen expression in E. coli was due to differences in flagellar regulation between the different bacterial species. A z66− S. Typhi strain was transformed with pBSSB2 DNA isolated from E. coli SGB33, the plasmid was stably maintained, and the z66 antigen was dominantly expressed.
The linearity of pBSSB1 was confirmed experimentally by probing S. Typhi 404Ty genomic DNA cleaved with PmeI, SacI, SpeI, and XbaI with pBSSB1 (Figure 3A). Restriction endonucleases that were predicted to cut once (SpeI and XbaI) generated two DNA fragments, and those predicted to cut twice (PmeI and SacI) generated three DNA fragments (Figure 3A). The size of the observed restriction fragments corresponds to the sizes for linear pBSSB1 DNA predicted by the in silico digestion described in Figure 2. pBSSB2 DNA from E. coli SGB33 was also embedded in agarose plugs and incubated with S1 nuclease, which linearizes supercoiled circular DNA [23,24]. S1 nuclease activity was proven using purified pUC18 DNA. S1 nuclease failed to alter the mobility of pBSSB2 after incubation for 1 h (Figure 4, lane 4), consistent with linearity of the element. pBSSB2 DNA was entirely degraded by 3′-5′ exonuclease III (Figure 4, lane 5) but not by lambda exonuclease, which digests in a 5′-3′ direction (Figure 4, lane 6). Activity of both exonucleases was demonstrated on linearized pUC18 DNA. Additionally, omitting the proteinase from the preparation of plugs for PFGE prevented pBSSB2 mobilisation into the agarose gel (unpublished data). These data suggest that pBSSB1 may be similar to linear plasmids from Streptomyces in having protein covalently bound to the 5′ end of the DNA and not palindromic hairpin loops at the telomeres as found in other enteric linear elements, such as bacteriophage N15 from E. coli [25].
Cleaved genomic DNA from 404Ty was probed with a PCR amplicon of the tir sequence to determine if pBSSB1 was additionally inserted into the chromosome. If pBSSB1 is solely in an extra-chromosomal form, only two fragments corresponding to the tir would be detected. However, if pBSSB1 was also inserted in the chromosome, further DNA fragments would be expected. Figure 3B shows the Southern blotting analysis using the tir-generated probe against genomic 404Ty DNA cleaved with different restriction endonucleases. Only the two DNA fragments predicted to originate from linear pBSSB1 were detected, suggesting that the plasmid was not inserted into the chromosome at a detectable level.
Linear DNA replicons are extremely rare in enteric bacteria, and those that have been described, including PY54 of Yersinia enterocolitica [26], N15 of E. coli [25], and PKO2 of Klebsiella oxytoca [27], are linear hairpin-ended prophage. To our knowledge, pBSSB1 is the first linear element to be described in the Enterobacteriaceae that bears no detectable sequence homology to bacteriophage. The fact that a linear plasmid can exist and replicate in a pathogenic member of the Enterobacteriaceae and have an impact on the phenotype of the bacteria is a significant observation. The identification of pBSSB1 will facilitate future studies on the biology of such linear extra-chromosomal elements in other bacterial species. Despite pBSSB1 having no sequences common to previously described elements, it does share some structural features of known linear plasmids. Unlike other linear elements in enteric bacteria, pBSSB1 contains tirs, and our data demonstrate that the ends are capped with covalently bound protein, as found in Streptomyces linear plasmids [28,29], and not closed hairpin loops. The tir, GC skew, and coding bias suggest that pBSSB1 replicates from a central internal origin, as do all small and large Streptomyces linear plasmids, such as pSLA2 from S. rochei [28,29].
Global analysis of S. Typhi isolates suggests that genome variation is extremely limited in this pathogen [30,31]. Indeed, plasmids are relatively rare in this serovar and are generally restricted to members of the IncH1 family [14]. How and why S. Typhi acquired this element is open to speculation. Flagella play a critical role in the lifestyle of bacteria and are an important target of pathogen recognition by the mammalian innate immune system via the TLR5 pathway [1,32]. The incidence of typhoid in Indonesia is one of the highest in the world [33,34], and the fact that circulating S. Typhi strains have acquired and maintained an additional flagellin gene may be related to the population dynamics of typhoid infections in the region. H:z66+ strains have been identified only in this location, and although the H:d variant, H:j, has been isolated elsewhere, it is highly prevalent in Indonesia [35]. It is possible that there is significant immune selection ongoing within this S. Typhi population. This may be expected for a pathogen that causes systemic infection and has the potential to exist in a persistent state. Currently, we do not know if the presence of the z66 flagella or pBSSB1 impacts the pathogenicity of S. Typhi strains, nor is it readily possible to test this hypothesis in the laboratory, because S. Typhi only causes disease in humans.
The influence of gene gain via horizontal transfer on the pathogenesis of various Salmonella serovars has been well documented. The acquisition of novel DNA sequences by S. Typhi in Indonesia may have allowed them to adapt to a new niche, or may have increased their fitness within their prior niche.
S. Typhi In20 (H:z66+) and S. Typhi 404Ty (H:z66+) were isolated in Indonesia and were provided by Leon LeMinor (Salmonella Genetic Stock Centre, Calgary, Alberta, Canada). S. Typhi CT18 (H:z66−) and S. Typhi Ty2 (H:z66−), for which complete genome sequences exist, are from The Sanger Institute strain collection. S. Typhi In20 was transformed with pKD46 (S. Typhi SGB31) and the resulting kanamycin-resistant strain was named S. Typhi SGB32 (this study). High efficiency E. coli TOP10 (Invitrogen, http://www.invitrogen.com) were used to demonstrate the transferable nature of pBSSB2. Transformed E. coli TOP10 containing pBSSB2 was named E. coli SGB33 (this study). Transformed S. Typhi BRD948 containing pBSSB2 and expressing H:z66 was named S. Typhi SGB34 (this study). E. coli 39R861 was used for sizing plasmid extractions on agarose gels and contains plasmids of 7, 36, 63, and 147 kbp.
Plasmid DNA was prepared using an alkaline lysis method originally described by Kado and Liu [36]. The resulting plasmid DNA was separated by electrophoresis in 0.7% agarose gels made with 1x E buffer. Gels were run at 90 V for 3 h, stained with ethidium bromide, and photographed. High purity plasmid DNA was isolated for transformation using alkaline lysis and either AgarACE purification (Promega, http://www.promega.com) or ultra-centrifugation based upon a method described by Taghavi et al. [37].
Southern blotting was carried out using Hybond N+ nitrocellulose. Probes were prepared from purified PCR products (PCR purification kit; Qiagen, http://www.qiagen.com) amplified using primers outlined in Table S2, or from purified pBSSB1 DNA. Purified PCR products or plasmid DNA was labelled using the Gene Images CDP-Star and AlkPhos Direct Labeling kit (GE Healthcare, http://www.gehealthcare.com). Detection was performed with the Gene Images CDP-Star Detection kit. The sizes of restriction fragments were estimated by comparing migration distances against Hyperladder I (Bioline, http://www.bioline.com).
A kanamycin resistance gene was inserted into pBSSB1 using a modified version of the lambda red recombinase (one-step method) described by Datsenko and Wanner [38]. PCR products were amplified in ten 50-μl reactions with the primers described in Table S2 using pKD4 DNA as a template. PCR-amplified DNA was pooled, precipitated, and re-suspended in 10 μl of nuclease-free water. Re-suspended DNA was mixed with 50 μl of competent S. Typhi SGB31 cells (grown in LB broth, supplemented with 0.1 M arabinose, and harvested at 0.3 OD600) in 2-mm electroporation cuvettes (Invitrogen). Cells were electrotransformed (2.4 kV, 600 ohms, 25 μF; Bio-Rad Gene Pulser, http://www.bio-rad.com), allowed to recover for 2 h statically at 37 °C in 400 μl of SOC, and then plated onto LB medium supplemented with 25 μg/ml kanamycin.
An H:z66 cosmid was constructed by cloning the fljBz66 region into the BamHI site of vector cosmid p14B1 using a partial Sau3A digestion. The insert was shotgun sub-cloned into pUC18, sequenced, and annotated as previously described [39]. The sequence of pBSSB1 was completed by supplementing the cosmid insert sequence with draft data of S. Typhi 404Ty produced by 454 Pyrosequencing (454 Life Sciences, http://www.454.com) [40].
The linear nature of pBSSB2 was demonstrated using PFGE (CHEF DRII, Bio-Rad). Agarose plugs containing lysed bacterial cells were prepared using the CHEF Bacterial Genomic DNA Plug Kit (Bio-Rad) as recommended by the manufacturer. 1.2% agarose gels (0.5x TBE) were loaded with the genomic DNA plugs, and samples were electrophoresed for 16 h at 6 V/cm, 14 °C, 1–6 seconds switch time, in 0.5x TBE and stained with ethidium bromide. Band sizing was estimated by comparison to the migration of Hyperladder VI (Bioline). S1 nuclease, exonuclease III, and lambda exonuclease treatment was performed on pBSSB2 DNA in agarose plugs as previously described [24]. Activity of the enzymes was confirmed on linear and circular pUC18 DNA.
The pBSSB1 sequence and annotation is available from Genbank/EMBL (http://www.ncbi.nlm.nih.gov/Genbank) under accession number AM419040. The GenBank accession numbers for other sequences discussed in the manuscript are fljBz66 gene (AB108532), S. Typhi CT18 (NC_003198), and S. Typhi Ty2 (NC_004631).
|
10.1371/journal.ppat.1002720 | Rates of Viral Evolution Are Linked to Host Geography in Bat Rabies | Rates of evolution span orders of magnitude among RNA viruses with important implications for viral transmission and emergence. Although the tempo of viral evolution is often ascribed to viral features such as mutation rates and transmission mode, these factors alone cannot explain variation among closely related viruses, where host biology might operate more strongly on viral evolution. Here, we analyzed sequence data from hundreds of rabies viruses collected from bats throughout the Americas to describe dramatic variation in the speed of rabies virus evolution when circulating in ecologically distinct reservoir species. Integration of ecological and genetic data through a comparative Bayesian analysis revealed that viral evolutionary rates were labile following historical jumps between bat species and nearly four times faster in tropical and subtropical bats compared to temperate species. The association between geography and viral evolution could not be explained by host metabolism, phylogeny or variable selection pressures, and instead appeared to be a consequence of reduced seasonality in bat activity and virus transmission associated with climate. Our results demonstrate a key role for host ecology in shaping the tempo of evolution in multi-host viruses and highlight the power of comparative phylogenetic methods to identify the host and environmental features that influence transmission dynamics.
| Rapid evolution of RNA viruses is intimately linked to their success in overcoming the defenses of their hosts. Several studies have shown that rates of viral evolution can vary dramatically among distantly related viral families. Variability in the speed of evolution among closely related viruses has received less attention, but could be an important determinant of the geographic or host species origins of viral emergence if certain species or regions promote especially rapid evolution. Here, using a dataset of rabies virus sequences collected from bat species throughout the Americas, we test the role of inter-specific differences in reservoir host biology on the tempo of viral evolution. We show the annual rate of molecular evolution to be a malleable trait of viruses that is accelerated in subtropical and tropical bats compared to temperate species. The association between geography and the speed of evolution appears to reflect differences in the seasonality of rabies virus transmission in different climatic zones. Our results illustrate that the viral mechanisms that are commonly invoked to explain heterogeneous rates of evolution among viral families may be insufficient to explain evolution in multi-host viruses and indicate a role for host biology in shaping the speed of viral evolution.
| RNA viruses display exceptionally variable rates of molecular evolution, with up to 6 orders of magnitude in nucleotide substitution rates observed among viral species [1]. Because of the importance of genetic and phenotypic evolution for infecting new host species, evading immune responses and obstructing successful pharmaceutical development, understanding the factors that govern the speed of viral evolution is critical for mitigating viral emergence [2]. To date, explanations for evolutionary rate heterogeneity have been predominately virus-oriented. These have focused on features of genomic architecture that determine underlying mutation rates, aspects of the virus life cycle such as latency and transmission mode that can influence the replication rate within hosts and generation times between hosts, and diversification through positive selection [2]–[5]. Less well understood are the determinants of evolutionary rate variation among closely related viruses (e.g., species within genera) or among lineages of the same viral species circulating in different geographic regions or host species [6], [7]. Because viral genomic features and replication mechanisms are minimally variable at such shallow taxonomic levels, aspects of host biology that influence rates of transmission and replication may be more likely to control the tempo of viral evolution. Consistent with this hypothesis, several human viruses (e.g., HTLV, HIV and Chikungunya virus) that exploit multiple modes of transmission or experience variable immunological pressures within-hosts demonstrate accelerated molecular evolution in conditions associated with enhanced transmission and replication [6], [8], [9].
The propensity of many RNA viruses to ‘jump’ between host species presents an intriguing natural experiment to test whether viral evolutionary rates change according to traits of host species that influence viral replication and transmission or remain evolutionarily conserved along the ancestral history of the virus, reflecting intrinsic biological features of viruses [10], [11]. Moreover, knowledge of accelerated viral evolution in certain reservoir hosts might be useful for predicting the geographic or species origins of future host jumps if faster evolution enhances the genetic diversity on which natural selection may operate. Despite these implications for viral emergence and evolution, the relationship between host biology and viral evolution remains largely unexplored. In two recent analyses, influenza A viruses infecting wild and domesticated birds and infectious haematopoietic viruses of wild and farmed fish each showed some intra-specific variation in viral evolutionary rates. However, host species identity failed to explain these differences, perhaps because high rates of transmission between host groups in each system diluted the effects of any single species on virus evolution [7], [12]. Elucidating the influence of host biology on viral evolution therefore requires large datasets of closely related viruses from multiple ecologically distinct host species that are largely capable of independent viral maintenance.
Rabies virus (Lyssavirus, Rhaboviridae) is a globally distributed and lethal zoonotic agent that causes more than 50,000 human deaths annually [13]. Although most human rabies is attributed to dog bites in developing countries, rabies virus also naturally infects over 80 bat species from 4 chiropteran families, and bats represent an increasing source of human and domesticated animal rabies in the Americas [14]. The phylogeny of bat rabies virus reveals viral compartmentalization into many largely species-specific transmission cycles, which have arisen from repeated host shifts within the bat community [15], [16]. Coupled with the diverse behavioral and life history strategies of bats, rabies virus therefore provides a unique opportunity to explore the effects of host biology on virus evolution while explicitly accounting for the effects of the ancestral history of the virus on evolutionary rate by using the rabies virus phylogeny as a guide to past host shifts. Moreover, because many American bat species and genera are broadly distributed with distinct viral lineages in different parts of their geographic range, ecological effects that reflect geographic variation in host behavior can be distinguished from taxonomic effects that arise from the physiological similarity of closely related host species.
Bats represent an especially pertinent taxonomic group for exploring the effects of host biology on viral evolution because of growing interest in how bat ecology influences zoonotic agents such as SARS virus, Nipah virus and Ebola virus [17]. If the behavioral and ecological traits of bats that are hypothesized to influence the maintenance and emergence of pathogens affect either virus replication within hosts or the rate of transmission between hosts, they might also have consequences for viral evolution. For example, overwintering of temperate bats through hibernation or extended bouts of torpor might cause a seasonal pause in transmission and/or decelerated disease progression within hosts, perhaps due to metabolic down regulation of cellular processes or reduced contact rates while bats are inactive [18], [19]. These climate-mediated mechanisms might slow evolution in viruses associated with temperate bats compared to tropical species, where year-round food availability and milder temperatures extend bat and virus activity through all seasons. Next, high contact rates in colonial bats may promote infections with greater virulence and reduced incubation periods, increasing the number of viral generations per unit time and speeding viral evolution [19], [20]. Finally, long distance migration, a relatively common strategy in bats, may slow viral evolution by homogenizing viral populations or by reducing transmission if the physiological stress from migration removes infected hosts from the population, i.e., ‘migratory culling’ [21].
Here, we compile large datasets of bat rabies virus sequences to quantify variation in the evolutionary rate of rabies virus when associated with ecologically and behaviorally distinct reservoir species found in different geographic regions of the Americas. Further, we test whether the tempo of evolution undergoes episodic shifts when rabies virus establishes in new species, implicating host biology as a key driver of viral evolution, or evolves gradually along the ancestral history of the virus, reflecting the greater importance of conserved viral features in controlling evolutionary rates. Finally, we integrate ecological and genetic data through newly developed Bayesian hierarchical phylogenetic models to identify the traits of hosts and the environment that influence rates of viral evolution.
We employed maximum likelihood (ML) and Bayesian phylogenetic analyses to define 21 subspecies, species or genus specific lineages of rabies virus for comparative analyses of evolutionary rates (see Materials and Methods for analytical details and operational definitions of lineages). A relaxed molecular clock analysis indicated that viral lineages were relatively young, ranging in age from 83–305 years, with a most recent common ancestor of all bat lineages dating back to 1585 (95% Highest Posterior Density, HPD: 1493–1663; Table S1). To describe the evolutionary rate variation among viral lineages, we focused on substitution rates in the third codon position (CP3), as these predominately synonymous substitutions can indicate more clearly how viruses respond to processes affecting their replication rate and generation time between infections [22]. Average substitution rates estimated for each lineage independently (Independent Lineage Models, ILM) and by a hierarchical phylogenetic model (HPM) each spanned approximately one order of magnitude among viral lineages compartmentalized to different host species (ILM range: 8.31×10−5–2.08×10−3; HPM range: 2.16×10−4–1.07×10−3 substitutions/site/year). This indicates that select lineages exhibit approximately 5–22 fold acceleration of evolution relative to the slowest evolving viruses. The HPM substantially improved the precision of parameter estimates relative to the ILMs, with only negligible differences in point estimates for most lineages, as previously described in other host-virus systems [8], [23] (Figure 1). Notably, the more extreme values estimated by the ILMs, typically from viral lineages with less informative datasets, were drawn closer to the population mean, suggesting less susceptibility of the HPM to stochastic noise introduced by sampling error and potentially more accurate estimates.
If evolutionary rate is a relatively static trait of viruses, it should be conserved in novel environments and would be expected to reflect the ancestral history of the virus, with closely related viral lineages having similar rates, regardless of their contemporary host environment. We tested the degree of phylogenetic signal in the evolutionary rates of bat rabies virus lineages by quantifying values of Blomberg's K (a common statistic for diagnosing phylogenetic non-independence in comparative analysis) [24], [25]. In the context of viral host shifts, significant values of K would indicate that the evolutionary rate tended to remain similar after establishment in the recipient species, whereas weaker values of K would indicate greater shifts in evolutionary rates than expected between the donor and recipient host species. We detected very low and non-significant values of K for both rates estimated under the ILMs (K = 0.36, P = 0.58) and the HPM (K = 0.39, P = 0.40) and these estimates of K did not differ significantly from expected values given our phylogenetic tree and the observed rates under a null model with randomly distributed rates (Figure 2B). Indeed, lineages that shared a most recent common ancestor sometimes had disparate rates of evolution (e.g., LcV and LxV and PhV and MyV2 in Figure 2A), although other closely related virus pairs showed minimal differences in evolutionary rate (e.g., LnV and PsV). Notably, viral lineages maintained by bat species in the temperate zone frequently had slower rates of evolution than lineages from tropical or subtropical bats (Figure 2A). The plasticity of evolutionary rate was corroborated by our Bayesian phylogenetic analysis, which, accounting for uncertainty in the evolutionary history of bat rabies lineages, found no correlation in the evolutionary rates along consecutive branches in the bat rabies virus phylogeny (covariance = 0.005, 95% HPD: −0.051–0.058). These analyses demonstrated that rates of viral evolution may be altered or conserved following establishment in new host species and point to host biology rather than the ancestral history of the virus as the most likely candidate for controlling rabies virus evolution. Using a phylogeny of bat hosts from mitochondrial sequence data, we found that viral evolutionary rates were similarly unconstrained by host evolutionary relatedness (ILMs: K = 0.07, P = 0.18; HPM: K = 0.21, P = 0.08), such that viruses associated with closely related bat species or sub-species often had dissimilar evolutionary rates (Figure 1, Figure S1).
Because the evolutionary rates of rabies virus lineages could not be explained by the reservoir host or virus phylogeny alone, we tested whether physiological, ecological or environmental traits of hosts (Table S2) could instead determine viral evolutionary rates using a generalized linear model (GLM) comparison approach. The factors that we tested included both evolutionary conserved traits of bats (i.e., basal metabolic rate, coloniality, long-distance migration) and descriptors of sampling effort and climatic region of viral lineages that were largely independent of the bat phylogeny (Table S3). The GLM identified the climatic region of bat taxa as the single strongly supported predictor of viral evolution (Akaike importance weight = 1.0), alone explaining 66% of the variance in viral evolutionary rates (F1,19 = 37.2, R2 = 0.66, P<0.0001; Table S4). A phylogenetic generalized least squares regression approach, designed to control for any residual effects of the virus phylogeny on rates of viral evolution, yielded similar results (Table S5).
A weakness of the statistical methods described above was that they could not account for the often-substantial uncertainty in point estimates of evolutionary rate from the ILMs (Figure 1). We therefore incorporated the same categorical and continuous terms directly into a Bayesian HPM that allowed us to simultaneously quantify the posterior distribution of the rate of evolution for each viral lineage from the molecular sequence data, estimate regression model parameters and compare candidate models using Bayes factors (BF) while accounting for phylogenetic uncertainty. The Bayesian model echoed the strong support for accelerated viral evolution in the tropics and subtropics relative to viruses restricted to the temperate zone (log effect size, β = 1.24 [95% highest posterior density = 0.72–1.76]; BF = 466.54) with negligible support for all other predictors (BF≤1; Figure 3A). On average, rabies viruses found in tropical or subtropical bat species accumulated 9.44×10−4 (ILM: 1.15×10−3) substitutions per site per year (subs/site/year), while viruses in temperate bats accumulated only 2.53×10−4 (ILM: 2.92×10−4) subs/site/year (Figure 3B) – a nearly fourfold deceleration of viral evolution in temperate bats.
By applying an integrated ecological and genetic comparative approach to a unique dataset spanning hundreds of viruses isolated from many host species, our study demonstrated strong effects of host biology on the tempo of molecular evolution in an RNA virus. We observed approximately an order of magnitude of variation in rates of evolution among rabies viruses, indicating that certain lineages evolve up to 22 times faster than others, depending on the reservoir species (Figure 1). Such rate heterogeneity within a single virus species is exceptional given that similar variation is more commonly observed among different viral families and comparable to the variation in mitochondrial DNA divergence rates between vertebrate taxa genetically isolated for millions of years (e.g., whales versus rodents) [26]. Moreover, the young age of viral lineages in our analysis indicated that evolutionary rate could be altered within decades, consistent with rapid rate adjustment after shifts to new host species (Table S1). Since the genomic structure, transmission route and replication mechanisms are not known to vary among rabies virus lineages, the plasticity in evolutionary rate that we observed could only have arisen from ecological differences among reservoir bat species that influence transmission and/or replication.
Because the tempo of evolution shifted freely through the ancestral history of rabies virus, we sought to identify the traits of bat species that influenced viral evolution. Strikingly, the viral molecular clock ticked nearly four times slower in rabies viruses in temperate bat species compared to tropical and subtropical species (Figure 2A, Figure 3B). This pattern could not be explained by geographic structuring of bat diversity or evolutionary conserved aspects of bat physiology or behavior because several widely distributed bat species and genera supported disparate rates of evolution in viral lineages circulating in different climatic regions (Figure 1). This resulted in a weak phylogenetic signal of viral evolution in the bat phylogeny (Figure S1). Similarly, the geographic clustering of viral evolutionary rates in our analysis was unlikely to reflect contrasting patterns of natural selection among climatic regions because evolutionary rates were estimated exclusively from the third codon position of sequences, where most nucleotide substitutions are synonymous, and therefore most likely to be neutral.
Alternative explanations for the relationship between climatic region and the rate of viral evolution parallel previous work on latitudinal gradients of molecular evolution in free-living plants and animals. This work has suggested that accelerated evolution in the tropics might arise from shorter generation times and higher metabolic rates (potentially increasing mutations through greater production of free radicals) associated with warmer environmental temperatures [26], [27]. The latter effect of temperature on host metabolism is unlikely to influence the evolutionary rates of viruses found in heterothermic species such as bats, which also lack a strong relationship between latitude and basal metabolic rate [28]. In our study, the independence of the evolutionary rate of rabies virus from bat metabolic rates further argued against a metabolism-mediated relationship between environmental temperature and viral replication (Figure 3). In contrast, generation times between viral infections likely differed among climatic regions in ways that could produce the patterns of viral evolution that we observed. Specifically, year-round transmission and replication may increase the annual number of viral generations in tropical and subtropical bats relative to seasonal pulses of transmission in temperate species, thereby speeding evolution. Indeed, when we conditioned our Bayesian analysis on 231 iterations of the HPM that lacked the climatic region term, seasonal inactivity was the only predictor that gained strong statistical support (BF = 36) with significantly faster viral evolution in bat species that remain active year-round relative to species that hibernate or use prolonged torpor during winter (β = 1.01 [95% highest posterior density = 0.39–1.52]). In our statistical models, the selection of climatic region (a surrogate of seasonal activity) rather than records of activity collected from the literature may be explained by poor understanding of the occurrence and duration of seasonal inactivity and torpor for many bat species [29]. Because assignment of overwintering records often required generalization of a few observations to an entire species range, climatic region may have been a more accurate descriptor of seasonal activity, especially for species that demonstrate geographically variable overwintering behaviors [30]. Still, the effects of variable transmission dynamics on viral evolution that we suggest should be confirmed in other host-virus systems with natural variation in seasonality or through experimental manipulation of virus transmission.
Rapid evolution can enable the cross-species emergence of RNA viruses by increasing the genetic and phenotypic variation available to natural selection [2]. However, whether accelerated viral evolution increases the likelihood of emergence depends on the underlying forces of selection in the reservoir host, whether faster evolution increases variability in the genomic regions that are key to adaptation, and the strength of other ecological and physiological barriers to infecting new host species. In the case of bat rabies, faster viral evolution seemed to arise through an epidemiological mechanism: a greater number of viral generations per year. Enhanced transmission could therefore increase the likelihood of viral emergence in the tropics/subtropics by allowing more ecological opportunities for cross-species transmission. However, whether escaping seasonal transmission bottlenecks also provides an evolutionary advantage for host shifting requires understanding how the evolutionary rates that we estimated relate to standing diversity in the genomic regions that mediate viral adaptation to new host species. Although previous work in plant RNA viruses has demonstrated effects of host species on viral genetic diversity, the role of evolutionary rate in generating these effects and their impacts on cross-species emergence remain unknown [31]. Therefore, identifying the genomic regions that enable rabies virus host shifts and the ecological and evolutionary factors that may contribute to their diversity should be a key goal for predicting future rabies virus emergence.
Beyond bat rabies, accelerated molecular evolution in tropical environments is a topic of general interest for understanding the maintenance and emergence of viral infections that occur across geographic regions or experience altered transmission dynamics as a result of anthropogenic environmental change. For example, lineages of Chikungunya virus evolve more slowly in seasonal African environments where mosquito populations and transmission dynamics are more variable relative to urban transmission cycles in Asia, where consistently large human and mosquito populations may shorten times between infections and support epidemic maintenance over multiple years [9]. Similarly, viruses such as influenza show reduced seasonality in the tropics relative to temperate zones [32]. Our results would predict that this sustained transmission might accelerate evolution in tropical viral lineages relative to their temperate counterparts if each is maintained independently. We therefore emphasize the need to consider not only functional traits of viruses, but also the seasonality and epidemiological dynamics of the host-virus interaction for a more complete understanding of the tempo of viral evolution.
In conclusion, our study demonstrated a relationship between climate and the speed of viral evolution, which reinforces similar geographic structuring of molecular evolution as observed in free-living plants and animals [33]. This speeding up of evolution appeared to be driven by changes in the generation time between infected hosts, but not host genetic relatedness, indirect effects of temperature on host metabolism or differences in selective pressures. The broad geographic and host range of many rapidly evolving viruses, together with the increasing availability of molecular sequence data, makes them an ideal, real-time system to examine the epidemiology and evolution of host-pathogen ensembles. Our study revealed the complex interplay between ecological and evolutionary dynamics in multi-host viruses and highlighted an equally integrated framework for dissecting those interactions by combining ecological and genetic data.
Viral sequences were generated from tissue samples from naturally infected bats that were collected by state public health laboratories following human or domesticated animal exposures. Total RNA was extracted directly from bat brains without passage using Trizol (Invitrogen, Carlsbad, CA) according to the manufacturer's instructions. A 903 bp fragment comprising the last 687 bp of the N gene, a non-coding region following the 3′ end of N and a small fragment of the phosphoprotein gene was amplified by reverse transcription-polymerase chain reaction and sequenced using oligonucleotide primers 550F and 304R, as described previously [16]. To enable comparison with existing sequences in GenBank, only the coding region of N was used in subsequent analyses. Sequences generated herein have been deposited into GenBank under accession numbers JN594500–JN594503 and DQ445318–DQ445330, DQ445352 (updated sequences). An additional 650 complete or partial rabies virus nucleoprotein gene sequences that were associated with bats from North and South America and contained information on sampling date to year were downloaded from GenBank.
We collected information on the overwintering activity patterns, migratory behavior, roosting behavior and metabolic rates (basal and during seasonal torpor) of the bat species that served as reservoir hosts for the rabies viruses included here from the primary literature and existing databases (Table S2). Long distance migration was defined as seasonal movement of individual bats of at least 1000 km [34]. For 4 species for which basal metabolic rate (BMR) data were unavailable, we borrowed values from species within the same genus or family that had similar body mass. Notably, body mass explains >92% of variation in BMR in bats and phylogeny explains much of the residual variation [29]. When torpid metabolic rate (TMR) estimates spanned a range of temperatures or spatial locations, rates were selected to match the conditions that bats are likely to experience in northern latitudes of their range where hibernation/torpor is most important. To calculate TMR for species that lacked values in the literature, we estimated the relationship between BMR and TMR for the 9 species in our dataset for which both values were available. This relationship was remarkably consistent across species (TMR = 2.2–3.2% of BMR, mean = 2.8%), with the exception of Tadarida brasiliensis, for which TMR was 12.6% of BMR. Because the reported estimate of TMR for that species (and for the other subtropical and tropical species in our study) likely represented a daily torpor rather than longer-duration, seasonal torpor, it was excluded from the calculation of the average mentioned above [35]. Overwintering activity was challenging to classify because the frequency and duration of bat activity during winter are poorly understood for many temperate bat species and can vary substantially throughout their geographic range [30], [36]. Therefore, we classified species as inactive during winter if extended bouts of seasonal torpor or hibernation were reported in any part of their geographic range, recognizing that this classification may have been overly conservative. As a potentially more geographically sensitive proxy of year-round activity, the climatic region (tropical, subtropical, temperate) of the center of the geographic range of each viral lineage was also recorded. North American lineages circulating between 35° and 23.5° latitude and South American lineages circulating south of −23.5° latitude that support mean winter temperatures of ≥10°C were considered subtropical, and lineages found towards and away from the equator relative to these latitudes were classified as tropical and temperate, respectively [37].
To define phylogenetic lineages to be included in subsequent analyses, ML and Bayesian phylogenetic analyses were performed using Garli v.0.96b and BEAST v.1.6.1, respectively [38], [39]. The ML analysis used the General Time Reversible (GTR) model of nucleotide substitution with invariant sites (I) and Γ distributed rate variation among sites as suggested by Akaike's information criterion corrected for small samples size (AICc) in jModeltest [40]. The ML tree was estimated by 5 independent searches with random starting trees, followed by 5 additional searches using the best tree from the previous set of searches as the starting tree. For the Bayesian analysis, we linked substitution rates for the first and second codon positions (CP12) and allowed independent rates in CP3. Separate substitution models were selected for CP12 and CP3 in jModeltest using AICc after partitioning aligned sequences by codon position. The BEAST analysis therefore applied the TIM1ef+I+Γ substitution model to CP12 and the TVM+Γ substitution model to CP3. We used the Bayesian skyride model as a flexible demographic prior for viral effective population size and an uncorrelated lognormal relaxed molecular clock to accommodate rate variation among lineages. Five independent Markov Chain Monte Carlo (MCMC) analyses were run for 50 million generations each, with samples from the posterior drawn every 50,000 generations following variable burn-in periods based on convergence of likelihood values and model parameters. The results from the five runs were combined to generate a maximum clade credibility tree and divergence time summaries. Lineages for subsequent analyses of substitution rates included those that (i) contained at least 8 sequences (mean = 30.9 sequences), (ii) were supported by Bayesian posterior probabilities of >0.9 and (iii) were sampled over a minimum time span of 4 years (mean = 19.4 years). These conditions aimed to achieve a compromise between precision in estimates and hypothesis testing ability. To ensure that sparsely sampled viral lineages did not bias our central findings, statistical analyses were conducted with covariates designed to identify effects of sampling heterogeneity or using hierarchical phylogenetic modeling to incorporate uncertainty in rate estimates. Of the 28 viral lineages identified in the initial phylogenetic analyses, 21 fit our criteria for inclusion in subsequent analyses, amounting to a final dataset of 648 sequences collected between 1972 and 2009 from 21 bat species or sub-species.
Sequence alignments were constructed for each viral lineage and nucleotide substitution models were selected for CP12 and CP3 as described above. For each lineage, the substitution rate was estimated in BEAST assuming an uncorrelated lognormal relaxed molecular clock to accommodate rate variation along branches and the Bayesian skyline model as a flexible demographic prior that could be applied to all viral lineages. The evolutionary rate in CP3 was calculated by multiplying the mean substitution rate by the relative rate parameter for that partition. Each simulation was run for at least 100 million generations, with parameters sampled every 5,000–10,000 generations. The first 10% of each run was discarded prior to the construction of the posterior probability distributions of parameters. Each analysis was run sufficiently long that effective sample sizes for parameters were >200 and results of several independent runs were combined. Analyses of evolutionary rates focused on substitution rates in CP3 since these largely synonymous substitutions reflect differences in evolution associated with generation time [22]. However, rates in CP12 were closely correlated with CP3 rates (r = 0.87, P<0.0001). Rate estimates for all codon partitions are shown in Table S6 and Table S7.
The separate estimation of substitution rates for each rabies virus lineage assumes complete independence of parameters across viral lineages, but this is unlikely the case given the close evolutionary relationships among lineages and biological similarities of the processes of infection and replication among lineages. Because the quantity of data varied among lineages (the number and temporal range of sequences), independent estimation in sparsely sampled lineages may lack power, causing imprecise estimates. HPMs have been proposed to improve the precision of parameter estimates for partially independent datasets such as these (e.g., populations of HIV within different patients) by assuming that individual lineage parameters vary around a shared unknown, but estimable population mean [23]. More recently, tools have been developed within BEAST to incorporate fixed effects into HPMs and to select among candidate models via Bayes factors [8]. These models take the general form of:(1)where θ is the evolutionary response variable of interest (here, the rate of molecular evolution in CP3), β0 is an unknown grand mean, δ is a binary indicator that tracks the posterior probability of the inclusion of predictor, P, in the model and β is the estimated effect size of predictor P. The use of binary indicator variables (δ) within the MCMC search allows for a Bayesian stochastic search variable selection approach that simultaneously estimates the posterior probabilities of parameters for all possible combinations of predictors and allows for calculation of the Bayes factor support for individual predictors as the ratio of the posterior odds to the prior odds of each predictor in the model.
We constructed a HPM for the 21 bat rabies virus lineages that assigned separate strict molecular clocks to CP12 and CP3 of each viral lineage and included fixed effect predictors of the evolutionary rate in CP3. For each CP, the evolutionary rate parameter of the molecular clock, the parameters of the GTR substitution model and the shape parameter of the discrete Γ distribution were modeled hierarchically across lineages, with all other parameters varying independently across data partitions. Results were robust to simpler substitution models lacking Γ heterogeneity within each codon partition and to statistical analyses using the rate of evolution averaged across all three codon positions. The fixed effects in the full model included climatic region (temperate vs. tropics/subtropics), mass-independent BMR, mass-independent TMR, coloniality (solitary vs. colonial), seasonal activity (non seasonal vs. hibernation/periodic seasonal torpor) and long-distance migration (migrants vs. non-migrants) (Table S2). Climatic region was condensed to two categories in the HPM based on exploratory analyses that demonstrated no difference in evolutionary rate between tropical and subtropical viral lineages. All continuous variables were log transformed. The HPM was implemented in BEAST using four independent MCMC searches of 150 million generations each, with the posterior sampled every 5,000 generations. Results from the four runs were combined after discarding the first 10% of each. Effect sizes of predictors reported from the HPM were calculated conditionally on the portion of the posterior distribution for which the respective effects were included in the model (i.e., βi|δEffect i = 1). Similarly, evolutionary rate estimates from the HPM were calculated conditionally on samples of the posterior in which the statistically supported predictor was included in the model (107,773/108,004 samples). Source files for the BEAST analysis and R scripts for conditional effect size and parameter estimation are available from the corresponding author upon request.
In addition to the Bayesian hierarchical hypothesis testing framework described above, we conducted a more traditional GLM analysis of host predictors of mean viral evolutionary rates and a phylogenetic least squares (PGLS) regression analysis. These analyses contained the factors included in the HPM above, but because these methods do not account for uncertainty in the estimation of evolutionary rates (Figure 1), we also included several factors to identify the influence of estimation error: the number of years spanned and the number of sequences that comprised each dataset. For the GLM analysis, an initial model containing all terms was simplified using an exhaustive search of possible models using AICc in the glmulti package of R [41], [42]. Models with Akaike weights within 10% of the highest were retained in the confidence set shown in Table S4. The PGLS regression was conducted in the caper package of R, using the Pagel's λ statistic to account for phylogenetic non-independence of viral evolutionary rates [43]. Because λ for climatic zone was low in the virus phylogeny (λ = 4.5e-5), we had little power to reconstruct ancestral climatic states. This precluded testing whether viral jumps between climatic zones were correlated with directional changes in viral evolution.
Blomberg's K measures the degree of phylogenetic non-independence of species traits, with values ranging from 0 to infinity [24], [25]. Values of K<1 indicate less phylogenetic signal (more trait lability) than expected under a Brownian motion model of evolution and K>1 indicate more correlation with phylogeny than expected. The K statistic was calculated for the ILM and HPM sets of rate estimates using the topology of the rabies virus phylogeny in Figure 2A in the picante package of R [42], [44]. The statistical significance of estimates of K was tested by comparing the expected distribution of values on our phylogenetic tree to 5,000 randomizations of observed rates along the tips of the tree. A similar analysis using the ML phylogeny of bats estimated from published mitochondrial cytochrome oxidase I sequences assessed whether the ecological and physiological similarity of closely related host species promotes evolution towards similar rates of viral evolution (Table S3, Figure S1).
|
10.1371/journal.pgen.1000150 | Mitochondrial Targeting Adaptation of the Hominoid-Specific Glutamate Dehydrogenase Driven by Positive Darwinian Selection | Many new gene copies emerged by gene duplication in hominoids, but little is known with respect to their functional evolution. Glutamate dehydrogenase (GLUD) is an enzyme central to the glutamate and energy metabolism of the cell. In addition to the single, GLUD-encoding gene present in all mammals (GLUD1), humans and apes acquired a second GLUD gene (GLUD2) through retroduplication of GLUD1, which codes for an enzyme with unique, potentially brain-adapted properties. Here we show that whereas the GLUD1 parental protein localizes to mitochondria and the cytoplasm, GLUD2 is specifically targeted to mitochondria. Using evolutionary analysis and resurrected ancestral protein variants, we demonstrate that the enhanced mitochondrial targeting specificity of GLUD2 is due to a single positively selected glutamic acid-to-lysine substitution, which was fixed in the N-terminal mitochondrial targeting sequence (MTS) of GLUD2 soon after the duplication event in the hominoid ancestor ∼18–25 million years ago. This MTS substitution arose in parallel with two crucial adaptive amino acid changes in the enzyme and likely contributed to the functional adaptation of GLUD2 to the glutamate metabolism of the hominoid brain and other tissues. We suggest that rapid, selectively driven subcellular adaptation, as exemplified by GLUD2, represents a common route underlying the emergence of new gene functions.
| Little is known about the functional evolution of new hominoid genes. In this study, we utilized a combination of evolutionary analyses and cell biology experiments to unveil a novel mode by which the human- and ape-specific glutamate dehydrogenase enzyme (GLUD2) functionally adapted. We find that whereas the GLUD1 parental protein (present in all mammals) localizes to mitochondria and also to the cytoplasm, GLUD2 is specifically targeted to mitochondria. Using resurrected ancestral proteins and site-directed mutagenesis, we show that the optimized mitochondrial targeting capacity of GLUD2 is due to a single positively selected amino acid substitution in its N-terminal targeting sequence, which occurred soon after the duplication event in the ape ancestor 18–25 million years ago. The specialization in mitochondrial localization is probably linked to the function of GLUD2 in the glutamate metabolism of the brain (recycling of glutamate in astrocytes), but is likely also of functional relevance in other tissues in which GLUD2 is expressed. We suggest that in addition to the traditionally considered modes of functional adaptation (changes in gene expression and/or the biochemical function of the protein), rapid and selectively driven subcellular adaptation to specific ancestral compartments may represent a common yet previously little-considered mechanism for the origin of new gene functions.
| The process of gene duplication is the major mechanism underlying the origin of new gene functions and has thus significantly contributed to the emergence of adaptive evolutionary novelties during evolution [1],[2]. DNA-based gene duplication—the duplication of chromosomal segments containing genes—has been prevalent during hominoid evolution [3]. Similarly, the process of retroposition (or “retroduplication”), a mechanism generating intronless gene copies (retrocopies) via the LINE retrotransposon-mediated reverse transcription of mRNAs from “parental” sources genes [4],[5], has resulted in a large number of gene copies in apes [6]. A small number of functional ape-specific duplicates created by these mechanisms have been identified (e.g. refs. [6]–[11]). However, although several of these genes revealed signatures of positive Darwinian selection (e.g. [9]), suggestive of adaptive protein sequence evolution, the evolutionary fate and functional protein evolution of new ape genes remains poorly understood.
GLUD2 is one of the few hominoid-specific proteins for which positively selected amino acid substitutions could be related to functional change and adaptation. It is encoded by the intronless GLUD2 gene, which emerged via the reverse transcription of a messenger RNA from its parental gene GLUD1 in the hominoid ancestor 18–25 million years ago (Mya) [7]. The GLUD genes encode two distinct isoforms of glutamate dehydrogenase (GLUD, also termed GDH), an enzyme catalyzing the oxidative deamination of glutamate to α-ketoglutarate (generating ATP through the Krebs cycle) and ammonia, a reversible reaction that takes place in mitochondria [12]. Previous work showed that the GLUD2-encoded enzyme evolved unique, potentially brain-specific functional properties soon after the duplication event by virtue of two key amino acid substitutions that were fixed as a result of positive selection [7],[13]. Due to these substitutions, GLUD2 shows higher activity at neutral pH than GLUD1, is less sensitive to GTP inhibition, and—unlike GLUD1—requires high ADP levels for its allosteric activation [14]. It was suggested that these properties reflect the functional adaptation of GLUD2 to the metabolism of neurotransmitter glutamate in the brain [13],[15].
Here we have further investigated the functional adaptation of the ape-specific glutamate dehydrogenase. We show that whereas GLUD1 localizes to the mitochondria as well as the cytoplasm, GLUD2 is specifically targeted to mitochondria, due to a single key amino acid substitution in its signal peptide, which emerged in the common hominoid ancestor and appears to have been fixed under the influence of positive selection. The enhanced mitochondrial targeting capacity of GLUD2 probably reflects further selectively driven optimization of this enzyme to the glutamate/energy metabolism of the brain and other tissues. More generally, the evolution of GLUD2 represents a remarkable example of rapid, selectively driven subcellular adaptation and thus reveals a novel mode for the functional adaptation of new duplicate genes.
We previously identified selectively driven substitutions in the mature GLUD2 protein that led to altered enzymatic properties [7]. When further investigating the evolution of the GLUD coding sequences in apes, we noticed an overall significantly higher nonsynonymous to synonymous substitution rate on the GLUD2 branches (dN/dS∼2) compared to those of GLUD1 (dN/dS∼0.2; P<10−3) after the duplication event, when restricting the analysis to the 5′-end (first 159 nucleotides) of the sequences (Figure 1A, see also legend and Materials and Methods for details). Notably, this region codes for a mitochondrial targeting sequence (MTS) of 53 amino acids (Figure 2A), which is required to direct the GLUD1 enzyme to mitochondria [16].
Prompted by this observation, we sought to assess whether the accelerated change of the MTS of GLUD2 reflects the action of positive selection (rather than relaxation of selective constraint) and might therefore be of functional relevance. To this end, we traced the evolutionary history of the full-length GLUD2 coding sequence (including the MTS-encoding sequence), focusing on the first two internal branches after the duplication event (Figure 1A). These branches were previously shown to represent an adaptive phase during the evolution of the mature GLUD2 protein [7]. A maximum likelihood procedure that tests for selection at certain sites (see Materials and Methods for details) revealed two amino acid substitutions (position 7 and 25) under positive selection in the MTS region (P>0.95, Figure 2A). We thus hypothesized that GLUD2 might have functionally adapted by evolving altered subcellular targeting.
To explore this hypothesis, we first used an in silico method ([17], Materials and Methods) to predict subcellular localization of reconstructed ancestral GLUD2 MTS variants, representing sequences before (sequence upon duplication event, node A) and after (great ape ancestor, node C) the period of positive selection (Figure 1A). Interestingly, this analysis suggested a substantially higher mitochondrial targeting probability for the node C targeting sequence (0.92) than for that of node A (0.28; Figure 1A).
To experimentally test these predictions, we synthesized the reconstructed MTSs for the node A and node C variants and fused them to a fluorescent (GFP) reporter. As GLUD2 is thought to have particularly adapted to a function in degrading neurotransmitter glutamate in astrocytes [13],[14], we transfected a human astrocyte-derived cell line (LN229, glioblastoma) with a vector encoding these recombinant proteins. These experiments revealed a striking pattern. We found that the node A MTS-GFP fusion protein localized to mitochondria, as expected, but that it could also be detected in the cytoplasm in most cells (Figure 1B, 1D, and Figure S1). In contrast, the node C MTS protein localized specifically to mitochondria in the vast majority of cells (Figure 1C, 1D, and Figure S1). Thus, consistent with the in silico predictions, our experimental analysis strongly suggests that the MTS of GLUD2 evolved the capacity to more specifically target the GLUD2 enzyme to mitochondria during the period of positive selection.
Further localization experiments showed that—similarly to the node A protein (Figure 1B, 1D, and Figure S1)—the human GLUD1 MTS-GFP fusion protein generally localizes to both mitochondria and the cytoplasm (Figure 1D and Figure S1). This suggests that GLUD1 preserved the ancestral localization pattern since the time of the duplication event (node A), which is consistent with the paucity of amino acid substitutions during GLUD1 evolution and its low mitochondrial targeting prediction value (0.30, Figure 1A).
These experiments also showed that the human GLUD2 MTS retained the increased mitochondrial targeting capacity (Figure 1D and Figure S1), in agreement with the high mitochondrial localization probability (0.92) estimated in silico (Figure 1A). Thus, the enhanced mitochondrial targeting specificity of the GLUD2 MTS was preserved after the period of positive selection on the lineage leading to humans. We obtained similar results for two other cell lines (human HeLa cells and COS7 from African green monkeys), further supporting the notion of a subcellular targeting shift of GLUD2 during its early evolution (Figure S2).
To more precisely date the shift of the GLUD2 targeting specificity, we assessed the subcellular localization of GLUD2 from the last common hominoid ancestor (node B). We found that the resurrected node B protein localized specifically to mitochondria in the majority of cells (Figure 1D and Figure S1), consistent with the high mitochondrial prediction value (0.91, Figure 1A). This suggests that GLUD2 had already evolved an increased mitochondrial localization specificity in the common hominoid ancestor ∼18–25 million years ago.
To assess the subcellular targeting capacities of the GLUD MTS sequences in the context of their physiologically targeted proteins, we performed similar experiments using full-length GLUD-fluorescent protein fusions. These experiments confirm the results obtained using the MTS-GFP fusions for the human GLUD1 and GLUD2 proteins (Figure 3A–C and Figure S3). We also analyzed the subcellular localization of extant GLUD2 proteins from the other apes. Indeed, GLUD2 from all apes localizes predominantly to mitochondria (Figure 3A and Figure S3). Thus, the enhanced mitochondrial targeting specificity of GLUD2 was conserved throughout hominoid evolution.
Which substitutions in the GLUD2 MTS contributed to the increased mitochondrial targeting capacity? A typical MTS contains several positively charged residues, such as lysines or arginines, and hydrophobic residues, generating an amphipathic helix [18],[19]. Due to the electrical potential across the inner membrane of mitochondria (the mitochondrial matrix being negatively charged), positive charges within the MTS are assumed to electrophoretically promote transfer of proteins across this membrane [18],[19].
One of the two positively selected amino acid changes in the GLUD2 MTS involves a glutamic acid (E) to lysine (K) substitution at position 7 of the sequence (Figure 2A). Notably, this substitution–which occurred in the common hominoid ancestor during the time of the switch in targeting specificity–introduces a positive charge to the MTS by replacing a negatively charged residue (Figure 2A). The second amino acid substitution under positive selection (D25H) replaces a negatively charged residue (aspartate, D) and at the same time introduces a partially positively charged amino acid (histidine, H) at position 25 of the MTS.
A helical wheel representation of the secondary structure of the helix formed by the GLUD2 MTS illustrates its modified properties (Figure 2B). The E7K and D25H substitutions introduce additional positively charged amino acids at one side of the α-helix within a previously weakly positively charged patch. Opposite to this charged patch, an ancestral patch with hydrophobic amino acids is found, which has remained largely unchanged during the evolution of GLUD2. Thus, the two positively selected substitutions are predicted to promote the formation of an amphipathic helix, which may function as an optimized MTS.
Consistently, changing these residues in the different GLUD sequences alters the in silico predictions of the GLUD mitochondrial targeting capacities. In particular, the E7K substitution dramatically alters these predictions. For example, introducing this substitution into the human GLUD1 sequence leads to an increase of the mitochondrial targeting probability from 30% to 90%, whereas the D25H substitution increases the GLUD1 wild type value to only 35%. The predominant effect of E7K is expected, given that this substitution represents the more radical substitution, as it replaces a negatively charged residue with a fully positively charged amino acid (see above, Figure 2B). Thus, we hypothesized that the E7K substitution was the key contributor to the evolution of optimized mitochondrial targeting of GLUD2.
To test this hypothesis, we introduced the E7K substitution into the MTS of human GLUD1 using site-directed mutagenesis. Remarkably, the mutant GLUD1E7K MTS shows a dramatic increase in mitochondrial localization capacity relative to the wild type variant (Figure 4), which is indistinguishable from those observed for extant or ancestral GLUD2 variants from node B and C (Figure 1 and Figure S1). We obtained similar results when introducing the E7K substitution into the full-length GLUD1 protein (Figure 5). Thus, in accord with our prediction, the subcellular adaptation of GLUD2 appears to have been mainly driven by this one key substitution that occurred soon after the retroduplication event in the common hominoid ancestor.
In support of the notion that E7K substitution was key to the increased mitochondrial targeting capacity of GLUD2, we find that reverting this substitution back to the ancestral glutamic acid in the GLUD2 sequence reduces its mitochondrial targeting specificity to a level that is indistinguishable from that of the parental GLUD1 protein (Figure 5). In conclusion, while the D25H substitution and potentially other substitutions that occurred during the period of positive selection might have contributed to enhanced or altered mitochondrial targeting in vivo, the key substitution rendering GLUD2 specific to mitochondria was E7K. Notably, this residue is conserved (as glutamic acid) in GLUD targeting sequences from other mammals, including mouse (Figure 2A) and opossum (not shown), a marsupial that diverged from primates around 180 Mya [20]. Generally, our results lend striking experimental support to a hypothesis suggesting that subcellular localization changes of duplicate proteins could occur by key substitutions in protein targeting sequences [21].
We finally note that the GLUD2 MTS seems to have lost some of the enhanced mitochondrial targeting specificity on the gibbon lineage (Figure 3), consistent with the lower in silico prediction value for the gibbon GLUD2 MTS (0.74, Figure 1A). This is presumably mainly due to a substitution at the third position of the gibbon GLUD2 MTS–a change from the ancestral arginine residue in the positively charged patch of the MTS helix to a non-charged cysteine residue (Figure 2A)–which reduces the net positive charge of the protein and leads to a reduction of the in silico-predicted mitochondrial localization probability (from 0.92 to 0.74).
Based on previous work, it was suggested that the emergence of GLUD2 in hominoids may have permitted an increased astrocyte metabolism of glutamate [7],[13]. GLUD2 evolved its unique enzymatic properties soon after the duplication event in the common hominoid ancestor (∼18–25 Mya), on the basis of two positively selected amino acid substitutions in the mature protein (see introductory paragraph and refs. [7],[13],[14]).
Here we have identified an additional mechanism through which GLUD2 appears to have functionally adapted. We show that GLUD2 evolved an enhanced mitochondrial targeting specificity, mainly by virtue of a single amino acid change in its MTS, which also appeared during the period of positive selection in the common hominoid ancestor. Thus, while its parental protein GLUD1 localizes to mitochondria (as previously reported, ref. [16]) but also to the cytoplasm, the subcellular localization of GLUD2 is largely restricted to mitochondria.
What was the selective benefit of the enhanced mitochondrial targeting capacity of GLUD2? We propose two—not mutually exclusive—scenarios that may explain this observation. First, in addition to its mitochondrial function, GLUD1-encoded GDH may have an–as yet–unknown function in the cytoplasm, akin to other mitochondrial enzymes (e.g., fumarase, ref. [22]). Due to the changes in its targeting sequence upon the retroduplication event, the ape-specific GLUD2 protein evolved a specific function in one of these ancestral compartments–the mitochondrion. This subcellular adaptation might have been particularly relevant with respect to the presumed function of GLUD2 in astrocytes (see above), where it is thought to be involved in the degradation/metabolism of neurotransmitter glutamate—a process taking place in mitochondria. We note, however, that recent work revealed that GLUD2 (similarly to GLUD1) is transcribed in many or most human tissues (Bryk et al., unpublished). This finding is in contrast to a previous study, which suggested that GLUD2 is rather specifically expressed in brain, retina, and testis [23]. Consequently, the subcellular adaptation of GLUD2 is likely of functional significance for hominoid tissues in general.
A second possibility that might explain the more specific mitochondrial targeting of GLUD2 involves the rather large variability of mitochondrial membrane potentials, which depend on the tissue and cell type [24],[25]. While mitochondria in tissues such as heart and muscle have high membrane potentials (i.e., they are more negatively charged inside the mitochondrial matrix than mitochondria in cells from other tissues), glial cells—such as astrocytes—have lower membrane potentials. Thus, GLUD2 may have evolved a more positively charged targeting sequence to compensate for the low membrane potential of mitochondria in astrocytes, thus ensuring efficient import of GLUD2-encoded GDH into mitochondria in these glial brain cells.
Further work is required to more precisely understand the physiological implications of the enhanced mitochondrial localization specificity of this recently emerged hominoid protein. In any event, our results suggest that the shift in subcellular targeting specificity of GLUD2 was beneficial to the evolution of the glutamate/energy metabolism of the hominoid brain and/or other tissues, as it appears to have been driven by positive selection.
More generally, our study provides a remarkable example of a novel mode for the origin of new gene functions [21], [26]–[28]. It has long been known that paralogous protein family members may localize differently in the cell (e.g., ref. [29]). Indeed, recent work using yeast as a model system suggests that subcellular adaptation represents a rather common mechanism through which duplicate genes may functionally diversify [30]. Interestingly, a hominoid-specific protein was recently shown to have completely changed its subcellular localization during its evolution due to positive selection [31], thus representing a case of “neolocalization” [30]. Here we have shown that newly emerged proteins such as GLUD2 may rapidly adapt to specific ancestral compartments (a process termed “sublocalization”; ref. [30]) under the influence of positive selection at key sites. We thus suggest that in addition to changes in gene expression and/or the biochemical function of the protein, rapid and selectively driven subcellular adaptation (by either neo- or sublocalization) is likely to represent a common mechanism underlying the emergence of new gene function.
The phylogenetic tree of the GLUD1/GLUD2 sequences coding for the mitochondrial targeting peptide was based on the known GLUD topology (ref. [7], which also corresponds the commonly accepted hominoid species phylogeny). dN/dS ratios and the number of synonymous and nonsynonymous changes in the phylogenetic tree were estimated using the codeml free-ratio model as implemented in the PAML4 package [32]. To assess whether the dN/dS ratio of the GLUD2 MTS is significantly elevated relative to that of GLUD1, we first compared a one-ratio codeml model (which assumes an equal dN/dS ratio for all the branches in the phylogeny) to a two-ratio model, where an additional dN/dS ratio is allowed on the GLUD2 lineages. Differences between these two models as well as the null and alternative models described in the following were compared using a likelihood-ratio test [33]. We note that the dN/dS rate of GLUD1 after the duplication event is not significantly different from that in the remaining GLUD1 branches in the tree (P<0.49), which suggests that the selective constraint on the coding sequence of GLUD1 has not changed after the emergence of GLUD2.
To assess whether the GLUD2 coding sequence (including its MTS) has evolved under the influence of positive selection, we used a conservative branch-site test [34]. We compared the likelihood of a model, which allows for dN/dS>1 at a subset of sites (i.e., dN/dS is estimated from the data) on the two internal branches after the duplication event, to that of a null model where dN/dS of this site class was fixed to 1. The dN/dS ratio was found to be significantly larger than 1 (P<0.02), consistent with a previous analysis focusing on the sequence encoding the mature protein [7].
Specific sites under positive selection were predicted using a Bayesian approach [35] as implemented in codeml. The ancestral sequences for nodes A, B, and C, were reconstructed using a one-ratio model (M0) as implemented in codeml. The posterior probabilities for reconstructed codons at all nodes were high (>0.95). Only the ancestral sequences for the two codons at positions 24 and 25 could not be unambiguously determined at nodes B and C, as these positions overlap with the deletion of 9 nucleotides in gibbon and two substitutions occurred at these positions on the branches between nodes A and C. The substitutions were assigned to branch A–B (Figure 1A and 2A), as determined by codeml, but could equally be assigned to branch B–C.
To analyze the mitochondrial targeting sequences of GLUD1 and GLUD2 and to assess subcellular localization, we used the PREDOTAR software ([17], http://urgi.versailles.inra.fr/predotar/french.html). We note that other target sequence analysis/subcellular prediction tools provided similar results (not shown).
To analyze the structure and property changes of the GLUD1/GLUD2 mitochondrial targeting sequences, we used a helical wheel prediction tool (http://rzlab.ucr.edu/scripts/wheel/wheel.cgi).
GLUD1 and GLUD2 coding sequences were obtained by PCR (primers sequences available upon request) using the following primate genomic DNA samples from the ECACC repository (Wiltshire, UK): Human “Caucasian”, chimpanzee (Pan troglodytes), gorilla (Gorilla gorilla), orangutan (Pongo pygmaeus), gibbon/siamang (Symphalangus syndactylus), and African green monkey (Cercopithecus aethiops sabaeus). The reconstructed GLUD sequences (see above, section Evolutionary Analysis) were synthesized by GenScript and cloned. GLUD targeting sequence mutants were obtained through site-directed mutagenesis by introducing the substitutions E7K and K7E in the GLUD1 and the GLUD2 sequences, respectively (all primers and restriction enzymes used are available upon request). All sequences were cloned into pEGFP-N1 (Clontech) vectors using standard procedures.
GLUD sequences that were not already available (GLUD1 MTS coding sequences from apes and African green monkey) were determined using standard sequencing procedures (sequences were run on an ABI 3730 automated sequencer) and the samples described above. These sequences were deposited in Genbank (see below for accession numbers).
HeLa, LN229 and COS7 cells were cultivated under standard conditions. Cells grown on MatTek Glass Bottom Culture Dishes (MatTek) for 24 hours were transiently transfected with the different GLUD constructs using Lipofectamine Plus (Invitrogen) according to the protocol of the supplier. 23.5 hours after transfection, mitochondria were stained with MitoTracker Red CMXRos (Invitrogen). Living cells were analyzed using a Confocal Microscope Zeiss LSM 510 Meta INVERTED by using a 63-fold oil objective. We used LSM for image analysis. In order to quantify the number of transfected cells that express GLUD proteins specifically in mitochondria, or in both the cytoplasm and mitochondria, we assigned a code to each dish with the respect to the construct used for transfection. We then proceeded with blind counts of the cellular phenotypes for each experiment. Specifically, the percentage of cells with GFP signals only in mitochondria was assessed by examining 10–50 transfected cells at 40-fold magnification over ten arbitrarily chosen areas on the dish. Each experiment was repeated five times. Differences between treatment groups were evaluated using ANOVA followed by a Post Hoc (Tukey HSD Test), with significance set at P<0.01.
The Genbank (http://www.ncbi.nlm.nih.gov/Genbank/) accession numbers for the previously unpublished GLUD1 MTS coding sequences are: EU828516 (chimpanzee, Pan troglodytes), EU828520 (gorilla, Gorilla gorilla), EU828517 (orang-utan, Pongo Pygmaeus), EU828518 (Siamang, Symphalangus syndactylus), and EU828519 (African green monkey, Cercopithecus aethiops sabaeus).
|
10.1371/journal.pcbi.1002984 | Stability and Responsiveness in a Self-Organized Living Architecture | Robustness and adaptability are central to the functioning of biological systems, from gene networks to animal societies. Yet the mechanisms by which living organisms achieve both stability to perturbations and sensitivity to input are poorly understood. Here, we present an integrated study of a living architecture in which army ants interconnect their bodies to span gaps. We demonstrate that these self-assembled bridges are a highly effective means of maintaining traffic flow over unpredictable terrain. The individual-level rules responsible depend only on locally-estimated traffic intensity and the number of neighbours to which ants are attached within the structure. We employ a parameterized computational model to reveal that bridges are tuned to be maximally stable in the face of regular, periodic fluctuations in traffic. However analysis of the model also suggests that interactions among ants give rise to feedback processes that result in bridges being highly responsive to sudden interruptions in traffic. Subsequent field experiments confirm this prediction and thus the dual nature of stability and flexibility in living bridges. Our study demonstrates the importance of robust and adaptive modular architecture to efficient traffic organisation and reveals general principles regarding the regulation of form in biological self-assemblies.
| While migrating, the nomadic army ant Eciton burchellii forms long trails of workers that can extend over hundreds of meters in the rain forest. To facilitate the movement of sometimes millions of individuals on uneven and unpredictable terrains, part of the ant workers link together their legs and bodies to form temporary bridges over gaps along the trails. In this work we showed that these bridges were formed readily when the flow of ants hit an unspanned gap and were dismantled very quickly after traffic has ceased on the trail. However, we also observed that the bridges were formed and remained stable under a large spectrum of the traffic intensities on the trail. Using field experiments and computer simulations we discovered the construction rules used by the ants to create these living structures that are capable of enduring variations of the traffic while remaining highly responsive to its interruption. These results offer important insights about the mechanisms that regulate biological self-assemblies and they have potential applications in swarm robotics and swarm intelligence.
| The functional complexity of many collective biological organisations, including aggregates of cells [1], tissues [2], [3] and organisms [4]–[6], derives from the integration and processing of information between system components. Tissue and organ functions emerge from mechanical, chemical and/or electrical communication among cells, and many collective activities of group-living animals result from relatively locally-mediated interactions [4]–[8]. Collective dynamics are also prevalent in human societies; networks of social interactions influence a wide range of social processes such as adoption of health behaviour [9], the spread of opinions [10], [11], gossip [12] and the spontaneous formation of lanes, jamming and oscillatory flows in pedestrian crowds [13]–[16].
Effective traffic organisation is not only central to the success of many modern human societies – it is a crucial feature of coordinated action in many social insects [17]–[20]. For example under crowded conditions, the black garden ant Lasius niger maintains an optimal rate of food return by either spontaneously dividing the traffic on their foraging trails between different routes [18], or by alternating clusters of inbound and outbound ants when no other route is available [19]. Near-blind army ants of the genus Eciton offer some of the most spectacular examples of behavioural and physical adaptations to minimise congestion, and to regulate traffic flow. They live in large colonies (on average 700,000 ants [21]–[23]), and live their lives at a high tempo; Eciton burchellii workers, despite being approximately 1 cm long, form vast dendritic trail networks that can extend over 100 m with individual ants maintaining speeds of 13 cm s−1 [17] (8 cm s−1 when carrying prey items [24]).
Their high rates of traffic flow must be maintained over highly unpredictable and irregular terrain in order to maintain the prey-capture rate required for colony survival [25]. They thus face challenging traffic coordination problems. An important part of their strategy is their capacity to modify their environment using their own bodies as structural components to fill holes, span gaps, pass overhanging banks and direct flow on sharply-angled turns [21]–[26]. These self-assemblages form by ants linking together legs and bodies with special tarsal claws (see Fig. 1a), a morphological adaptation which also allows them to form temporary nests (“bivouacs”) [26].
To date, it is neither clear how these structures form or whether, and if so how, they are regulated. However the response of individual ants to environmental heterogeneities has been investigated [24]. On uneven terrain, individual workers can use their bodies to plug ‘potholes’, remaining still as long as a consistent flow of traffic is maintained over their body. While offering important insights regarding the rules of thumb adopted by ants in filling holes, this study did not consider the spanning of larger gaps that would require the coordination of multiple ants. Recent experiments with fire ants have investigated collective formation of living rafts to survive flooding events [27]. Ants reaching the raft edge, and perceiving water, exhibit a small probability to attach to the edge instead of walking back. This simple rule is sufficient to generate quickly growing rafts during initial moments of flooding that can remain stable for its duration.
Army ant bridges must be consistently produced and remain coherent and functional within widely varying habitats and under variable traffic conditions. Here we employ an integrated experimental and theoretical approach to reveal how these ants, each of which has access only to relatively local sensory information, coordinate such robust, yet flexible, architectures. We identify the individual-level rules employed by ants when constructing and maintaining bridge structures and reveal how these relate to the functional properties of bridges as a collective structure including exceptional versatility, resilience and fault-tolerance.
Army ant bridges are highly dynamic entities, reforming very quickly following removal (see Movie S1). We analysed the dynamics of bridge construction using the following procedure: a natural bridge was located on a trail and filmed for a 10 minute interval following which all ants in the bridge were removed with tweezers (the duration of removal being <1 s); consequently the bridge (typically) reforms, a process we filmed for 10 mins before all ants in the bridge were again removed. This process of removal and reconstruction was repeated twice more, such that each experiment contained a baseline period (during which the “natural” bridge was filmed) and a total of three 10 minutes replicates in which the bridge was successively removed and re-construction was filmed. Thirteen of twenty bridges experimented upon in this way were analysed. Of the seven excluded bridges, it was not possible to satisfactorily observe the ants in four cases, and in the three other cases the ants altered the course of flow after the removal of the first bridge, voiding subsequent trials.
The number of ants participating in the bridge structure, as well as the outward and return flow of ants across the bridge, was recorded for 120 seconds, in 10 second intervals, prior to, and after, each removal of the bridge. Flow rate was measured as the number of individuals that fully crossed the gap. Therefore, individuals that entered the bridge structure were not incorporated into the flow rate, while ants voluntarily exiting the bridge were counted in the flow rate. In addition the size of each gap spanned by a bridge was measured as the longest distance between the objects (leaves or branches) to which the ants are attached on each side of the gap (see Fig. 1a). Gap lengths ranged from 7 mm to 26 mm, with an average length of 14.92 mm. We also measured the bridge width as the longest distance between ants along a line perpendicular to the length of the gap. Bridge width ranged from 8 mm to 55 mm, with all but 2 bridges between 8 mm and 18 mm wide. Note that bridge width was never found to have a significant effect in any of the subsequent analysis and therefore will not be considered in the rest of this study.
The traffic during the two minutes before and after the removal of the bridge is shown in Fig. 1b. Since it was stable prior to removal (GLMM, traffic at −120 vs traffic at −120 to −10 sec: N = 936, p>0.05 in every case) we aggregated traffic occurring over a two minute period preceding the manipulation to serve as a baseline with which to compare to traffic flow following bridge removal. After bridge removal traffic recovers quickly, reaching the baseline value after only 30 seconds (GLMM, traffic at 30 sec vs. baseline traffic: N = 936, z = 0.838, p = 0.402). A small but significant excess of traffic occurs from 40 to 100 seconds after bridge removal (GLMM, traffic at 40 to 100 sec vs. baseline traffic: N = 936, z = 4.847, p<0.0001), likely due to the accumulation of ants on each side of the gap. After 110 seconds, traffic resumes its baseline value (GLMM, traffic at 110 to 120 sec vs. baseline traffic: N = 936, z = 0.664, p = 0.507).
The corresponding number of ants forming the bridge both before and after bridge removal is shown in Fig. 1c. As with traffic dynamics, bridge size is stable during the two minutes prior to the removal and thus can serve as a baseline value with which to compare the reconstruction process (GLMM, bridge size at −120 vs bridge size at −120 to −10 sec: N = 936, p>0.05 in every case). The number of ants participating in the structure increases with time after bridge removal, but does not reach the baseline value (GLMM, traffic at each 10-second period vs. baseline traffic: N = 936, p = <0.0001 in every case); after two minutes, the number of ants participating in the structure represents on average 71% of the baseline value. After 30 seconds, which corresponds to the time after which the traffic over the gap has fully recovered, the number of ants participating in the structure represents on average only 43% of the baseline value. This suggests that only a minority of ants in the bridge are actually necessary to handle the traffic. We also tested the effect of three covariates on the number of ants forming the bridge: the global traffic flow over the gap; the proportion of ants returning toward the bivouac; and the gap length. Only the gap length has a significant effect (GLMM: N = 936, z = 2.409, p = 0.016), with an increase of the number of ants forming the bridge with the gap length.
In order to relate individual behaviours to the dynamic properties of bridges we recorded the times of arrival and departure, and the incoming direction of the ant (from the bivouac or raiding front) and the caste, for all ants that participated in bridge construction during the 2 minutes after each bridge removal. For all ants that did not participate in bridge construction, we recorded the times of arrival and the incoming direction only, the caste being difficult to assess in moving ants due to their high speed and the low shutter speed of the camera. Individual ants participating in the structure of the bridge were classified into four separate castes, using a visual sorting method that Franks [25] developed and demonstrated to be highly accurate.
In addition we calculated the localised flow rate (recorded as the number of passing ants exhibiting observable tactile contact with the body or antennae of that ant) for 50 separate individual ants (10 individuals randomly selected from 5 separate experiments) that left a bridge voluntarily. These individuals were selected among those that joined the bridge after its removal and they were followed until they left the bridge (for a maximum interval of 10 minutes after which a new bridge removal was performed). To ensure that individuals were really within the bridge structure prior to determining flow rate, only ants that were immobile for at least 10 seconds in the bridge were included in this sample. As much as possible, these ants represent the full range of different durations of time spent in the bridge. The position of the observed ants (edge or centre of the bridge), and the number of surrounding ants to which the focal ant appeared connected (i.e., we could observe a direct contact between the legs of the focal individual and the surrounding ants), was also recorded. We considered ants to be in the centre of the bridge if they were surrounded by at least one other ant on each of their right and left side. Otherwise ants were considered to be on the edge of the bridge.
Ants are more likely to decide to participate in bridge formation if the traffic on the trail (GLMM: N = 15216, z = 2.767, p<0.01) and if the packing density of the bridge (the number of ants participating in the structure divided by the length of the gap) at the instant when the focal ant joins the structure (GLMM: N = 15216, z = −2.671, p<0.01) are lower. The sooner ants participate in bridge formation, the longer they remain as part of the structure (CPHM: N = 449, z = 4.002, p<0.0001). This duration also increases with the packing density of the bridge at the instant when the focal ant joins the structure (CPHM: N = 449, z = −5.337, p<0.0001; see Fig. S1, inset) and is affected by the caste of the ant (see Fig. 2a), minor ants spending more time as part of the structure than media (CPHM: N = 449, z = 2.141, p = 0.0323) and sub-major/major ants (CPHM: N = 449, z = 2.476, p = 0.0133). This suggests a specialization of minor ants in the building of self-assemblage structures, an assumption supported by the over representation of this caste in bridges and bivouacs (see Fig. 2b): the percentage of minor ants in this structures - resp. 31.8% (293/920) and 33.7% (193/573) - is considerably higher than the one found in raiding trails - 22% (729/3314; data for bivouacs and raiding trails from [25]). Note that an opposite tendency is found for media ants (resp. 63.2% - 582/920 -, 62.1% - 356/573 - and 74.5% - 2471/3314). The incoming direction of the ant (from the bivouac or from the raiding front), and the total volume of traffic over the gap, however, do not affect significantly the time spent by ants in the structure of the bridge (CPHM: N = 449, p>0.05 in every case).
A survival analysis of the time spent by ants as part of the bridge confirms the existence of a stable core of ants, with about 50% of the ants joining a bridge structure remaining part of it after two minutes. The shape of the survival curve suggests that these ants might actually stay in the bridge for much longer periods of time (see Fig. S1), and extended observations of a subset of 50 ants showed that some ants stayed in the bridge for up to 500 seconds (which corresponds to the maximum observation time, see Fig. 3a).
Unlike our previous analysis of the influence of the total volume of traffic, detailed analysis of the aforementioned subset of 50 ants demonstrates that local traffic intensity strongly (and predictably) influences the time an individual remains as part of a bridge (Fig. 3a; GLMM: N = 50, p<0.0001). Furthermore - as suggested by the packing density - as the number of connected ants increases so does the time spent in the bridge (Fig. 3b; GLMM: N = 50, p = 0.016). These variables (local traffic intensity and number of surrounding ants) are significantly correlated (Pearson's product-moment correlation: t = 3.3927, df = 48, p = 0.0014), but removing one or the other variable from the GLMM results in a lower goodness of fit as evaluated with the Akaike Information Criterion and the Bayesian Information Criterion. This suggests that the two effects may be additive. One explanation for the existence of this correlation comes from the spatial organisation of the bridge; ants in the middle part of a bridge experience a significantly higher overhead traffic (see Fig. S2a; W = 137.5, p = 0.0005) and are connected, on average, to a larger number of neighbours (see Fig. S2b; W = 196.5, p = 0.0164).
To formalise the relationship between individual-level behaviour and the dynamics of bridge self-assembly we develop, and analyse, a data-driven, pseudo-spatial individual-based model of the ants' behavior (IBM, see Material and Methods section below for a detailed description of the model). This allows us to validate our understanding and to explore further the stability of the bridge structure under varying traffic conditions, thus allowing us to make explicit and testable predictions.
Unlike inert constructions, self-assembling structures have the capacity to rapidly and autonomously adapt their form to changing conditions [33]–[35]. We demonstrate that army ant bridges are versatile and fault-tolerant living architectures whose construction relies on simple rules of thumb. Bridges spontaneously adapt to the physical environment in which the colony operates and exhibit the property of being maximally robust for an oscillation period of approximately 3 seconds (Fig. 5a), whatever the traffic oscillation intensity, a property that stabilises bridges under natural oscillatory traffic fluctuations (Fig. 4a&b). However bridges are simultaneously highly sensitive and responsive in the face of sudden alteration of traffic on the trail. Thus this living architecture filters appropriately variations in traffic, discriminating between normal operating conditions and sudden, large perturbations of the foraging activity. This allows E. burchellii colonies to spontaneously create living bridges when and where they are needed, and appropriate to the size and type of gap encountered along the trail. We show that these properties contribute considerably to the efficient operation of E. burchellii raiding trails, as do features such as lane formation [17], pothole plugging [24], specialist porter castes [25] and the division of labour among teams of carriers [36], [37].
A detailed analysis of the behaviour of ants participating in bridge formation reveals the individual rules responsible for the above properties. The time spent by an individual in a bridge depends on its connectivity in the structure; highly connected ants being less likely to leave. These data support the view proposed by Schneirla [22] that interconnections among ants induce immobilisation. Highly connected ants are also more likely to occupy a central position in a bridge, where the majority of the traffic occurs thus increasing the stability of the structure and buffering the bridge to moderate fluctuations in flow. Our data suggest that not all the ants forming a bridge are absolutely necessary to ensure efficient traffic flow over a gap. It is plausible that these additional ants may give the bridge capacity to handle sudden increases in traffic.
The time spent by an individual in the bridge was also highly correlated with the caste to which this ant belongs. Ants of larger castes remained in the bridge for significantly shorter periods of time than ants of smaller castes. As a consequence the proportion of ants of each caste found in the bridge was different from the relative fraction of each caste found in the raiding trails [25]. In particular, minims were present in a higher frequency than expected, and medias were found at a lower frequency than expected (submajors and majors were found too rarely in bridges to provide accurate estimations). The relative percentages of each caste most closely approximated that found in E. burchellii bivouacs [25], which suggests that the construction principles in bridges and bivouacs are related. The amount of time each caste spent in the bridge appears to correspond with the broader context of their role in the colony. Both the specialist soldier caste (majors [22], [25], [32]) and the specialist porter caste (submajors [25]) spent the least amount of time in bridges. The generalist medium caste [25] stayed in the bridge longer than the specialist castes, though for less time than the minims. The role of the latters in raiding trails is not documented and our results suggest that they could be considered as bridge specialists. However it is not possible to decide whether this specialization is the result of an evolved behaviour. Perhaps, the most parsimonious explanation is that smaller castes have a higher probability of encountering gaps where they are forced to stretch their back legs and trigger their immobilizing mechanism than larger castes. This is supported by recent observations of pothole plugging in the same species showing that larger ants plug larger holes than smaller ants [24]. Therefore, minims' apparent specialization could be a simple side effect of their body length. Regardless of the exact mechanism that leads ants of smaller castes to stay in bridges longer than those of larger castes, minims function like a living pavement that fills in gaps, so that specialist nestmates may efficiently complete their transport and defence tasks.
Army ants within bridges exhibit a strong exponential relationship between the time they remain in the structure and the flow rate over them. The implications of this exponential relationship are considerable for the stability of bridges. For example, an increase in perceived flow rate of approximately 0.25 ants/sec results in doubling the time spent in the bridge. As a consequence, the stability of the bridges does not increase linearly with the intensity of the traffic on the trail. Instead, the transition between unstable and stable bridges as a function of the flow rate is closer to a step function. If the relationship between the flow over an individual and the time that individual spends in a bridge were linear, bridges would be comparatively less stable and less likely to form at high flow rate. The exponential relationship ensures that bridges are rapidly constructed, and are stable under high traffic conditions when bridges are most important to facilitate efficient movement and transportation over the trail network. At low flow rate however, when the maintaining of bridges is more costly relative to the benefits of foraging, this exponential relationship facilitates the disbandment of the unnecessary structure. Therefore, E. burchellii can considerably alter the stability of their bridges according to the conditions on the raiding trail.
The highly effective mix of stability and flexibility in this living architecture could provide valuable inspiration for technological solutions in collective robotics or swarm intelligence [38]–[41]. For example the “behavioural algorithm” implemented may be applied to develop self-reconfiguring aggregates of interconnected, or modular robots, or algorithms for dynamic allocation of connection slots in communication networks. Our work also offers insights about the adaptive value and functional design principles of biological self-assemblages [26], [33], [34].
All analysis and statistical tests were performed using R 2.13.0 [42]. Unless otherwise stated, comparisons between two independent samples were performed using the Mann-Withney test and correlations between variables were assessed using the Pearson's product-moment correlation test.
Bridges were filmed on the principal raiding trails of an E. burchellii colony during its nomadic phase in Soberania National Park, Panama. Bridges were located within 1–24 meters of the bivouac site and recorded over eight days of the dry season (12th March–20th March, 2001) using a digital video camcorder (Sony DVCAM) between 0900 and 1500 hours.
Analysis of flow over the bridge and the dynamics ants in the bridge were tested with generalised linear mixed effect models (GLMM) with a Poisson error distribution using the “lme4” package (version 0.999375-39). Data were nested as 10-second period per replicate per bridge.
The probability that an ant passing over the gap participate in bridge construction was analyzed with a GLMM with binomial distribution. The time spent by ants as part of the bridge during the 120 seconds after removal were analyzed using the “survival” analysis package (version 2.36-9). A Cox proportional hazards model (CPHM) was fitted to the data, and data were stratified by bridge. The proportional hazards assumption was tested using the method described in [43] and no significant difference from it was found both at a global level and at the level of the different tested factors.
Effects of the localised flow rate and the number of surrounding ants on the behaviour of the subset of 50 ants were tested with a GLMM with a Gaussian error distribution. The validity of the resulting model was verified graphically for each tested factor. P values were obtained using the “pvals.fnc” function from the “languageR” package (version 1.2).
Trails were filmed during stationery and nomadic phases in Sobernia National Park from June 14 to August 10 2009, between 0900 and 1600. At this time of day the flow rate of ants is usually bi-directional. The trails were all located within 50 m from the bivouac and filmed using a high definition camera (Panasonic HDC H300) for periods lasting from 24 to 1370 seconds.
Particle image velocimetry (PIV) [28], an optical flow technique, was used to quantify the flow of ants from a 256×256 pixels window centred on the main axis of trail (the axis being obtained by computing the average variation in intensity for all pixels in the image and fitting a straight line using a weighted regression). The temporal derivative was calculated by differentiating two successive frames, while the two spatial derivatives were calculated using 45 by 45 Sobel filters, and the products of these three derivatives were calculated and convolved with a 100 by 100 Gaussian filter. This resulted in the generation of a ‘velocity field’, i.e. an estimate of the direction and speed of motion for each pixel within the image. By projecting these velocity vectors on the main axis of the trail, and summing the norms of these projections, we could reliably estimate the relative variations of the traffic although due to inherently highly variable lighting and contrast under the rainforest canopy it is not possible to obtain with this method an accurate estimate of the absolute value of the traffic on the trail.
We analysed our traffic flow data with Lomb-Scargle periodograms over the frequency range 1/30 to 1 Hz (as described in [44]); statistically significant peaks of the corresponding periodogram for each dataset were averaged following weighting by their spectral power density to reveal the distribution of dominant frequencies.
At each time step (1 second, corresponding to the unit of time we used to analyse the individual behaviours of the ants), a random number of ants, drawn from a Poisson distribution with parameter λ, cross a gap of a given size. Each ant has a probability of stopping to form/join the bridge structure. A GLMM on our experimental measurements showed that should decrease with the traffic on the trail and the packing density of the bridge, roughly following a sigmoidal decrease in both cases. We chose to model the effect of traffic and packing density on as a 3 dimensional sigmoidal function (see Fig. 6) of the form: , where Dt represents the packing density of ants filling the gap at time t (in ants mm−1). α, β, γ and θ are control parameters, their values estimated by fitting the previous equation to data of success/failure to join the bridge during the two minutes after the bridge removal (least squares fit: α = 0.02; β = 144.5; γ = 3.258; θ = 12.97).
After an ant i joins the bridge structure, it exhibits a probability to leave it at time t. This probability decreases exponentially with the traffic over its body at each time step (i.e., the inverse of the data represented in Fig. 3a): . ρ and σ were estimated by fitting the previous equation to the inverse of the time spent by an ant as part of the bridge as a function of the traffic over its body (Least squares fit: ρ = 1.959; σ = 2.789).
Finally, the value of the traffic over the body of an ant at time t is obtained with the following equation: . The first part accounts for the proportion of the total traffic over the bridge that passes over the body of the ant i and depends on 3 control parameters φ, ψ and ω (see below for the estimation of their respective values), and on the packing density of ants in the bridge structure at the time where the ant i joined the bridge. Since they are more likely to occupy a central position in the bridge, ants that join when the packing density is low will tend to experience a larger proportion of the future traffic than the ants that join when the packing density is already high. The second part of the equation accounts for the ability of ants to integrate the traffic flow over their body for the last m seconds (m = 5, as suggested in [24]).
Only three parameters, φ, ψ and ω, could not be obtained directly from our experimental data. Their values were estimated by minimising the difference between the average dynamics of the number of ants forming the bridge in our experiments and in simulations of the model under similar conditions (bridge lengths and traffic intensity). The best fit was obtained for φ∼−0.52, ψ∼3 and ω∼2, with an average square difference between experimental and simulation data points of only 0.12 ants.
|
10.1371/journal.ppat.1002102 | Borrelia burgdorferi Requires Glycerol for Maximum Fitness During The Tick Phase of the Enzootic Cycle | Borrelia burgdorferi, the spirochetal agent of Lyme disease, is a vector-borne pathogen that cycles between a mammalian host and tick vector. This complex life cycle requires that the spirochete modulate its gene expression program to facilitate growth and maintenance in these diverse milieus. B. burgdorferi contains an operon that is predicted to encode proteins that would mediate the uptake and conversion of glycerol to dihydroxyacetone phosphate. Previous studies indicated that expression of the operon is elevated at 23°C and is repressed in the presence of the alternative sigma factor RpoS, suggesting that glycerol utilization may play an important role during the tick phase. This possibility was further explored in the current study by expression analysis and mutagenesis of glpD, a gene predicted to encode glycerol 3-phosphate dehydrogenase. Transcript levels for glpD were significantly lower in mouse joints relative to their levels in ticks. Expression of GlpD protein was repressed in an RpoS-dependent manner during growth of spirochetes within dialysis membrane chambers implanted in rat peritoneal cavities. In medium supplemented with glycerol as the principal carbohydrate, wild-type B. burgdorferi grew to a significantly higher cell density than glpD mutant spirochetes during growth in vitro at 25°C. glpD mutant spirochetes were fully infectious in mice by either needle or tick inoculation. In contrast, glpD mutants grew to significantly lower densities than wild-type B. burgdorferi in nymphal ticks and displayed a replication defect in feeding nymphs. The findings suggest that B. burgdorferi undergoes a switch in carbohydrate utilization during the mammal to tick transition. Further, the results demonstrate that the ability to utilize glycerol as a carbohydrate source for glycolysis during the tick phase of the infectious cycle is critical for maximal B. burgdorferi fitness.
| Borrelia burgdorferi is the vector-borne pathogen that causes Lyme disease. It has a complex life cycle that involves growth in a tick vector and a mammalian host — two diverse environments that present B. burgdorferi with alternative carbohydrate sources for support of growth. Previous studies suggested that glycerol may be an important nutrient in the tick vector. Here we show that genes predicted to be involved in glycerol metabolism have significantly elevated expression during all tick stages. Repression of expression in the mammalian host is dependent on the alternative sigma factor, RpoS. A mutant that cannot convert glycerol into dihydroxyacetone phosphate to support glycolysis was able to infect mice. In contrast, the mutant was present at significantly lower levels in nymphal ticks, its replication was delayed during nymphal feeding and longer feeding times were required for transmission from nymph to mouse. The results demonstrate that the ability to utilize glycerol as a carbohydrate source for glycolysis during the tick phase of the infectious cycle is critical for maximal B. burgdorferi fitness.
| Borrelia burgdorferi is the spirochetal agent of Lyme disease, the most frequently reported vector-borne disease in the United States [1]. In the Northeastern United States, B. burgdorferi is transmitted between mammalian hosts by the bite of the black legged deer tick, Ixodes scapularis, with the white-footed mouse (Peromyscus leucopus) serving as the primary reservoir host [2], [3]. The transmission cycle is as intricate as the life of the tick itself. B. burgdorferi are acquired by uninfected larvae feeding on an infected small mammal [4]. This is essential for the continued maintenance of B. burgdorferi in nature, since there is no transovarial transmission in Ixodes spp. [5], [6]. The bacteria remain in the midgut of engorged larval ticks through the molt. The infected nymph will take a blood meal on a mammal, at which point B. burgdorferi multiply and begin their migration from the tick midgut to the salivary glands from which they are transmitted to a mammalian host [7]–[9], thereby completing the enzootic cycle.
B. burgdorferi must adjust its gene expression program in response to the different physiological cues encountered during the natural enzootic cycle. In bacteria, regulation of gene expression in response to environmental cues is often mediated by two-component systems (TCS) and/or alternative sigma factors [10], [11]. The B. burgdorferi genome encodes only two alternative sigma factors and two TCS [12], [13]. Thus, B. burgdorferi must orchestrate its complex expression programs with a limited repertoire of known transcriptional regulators. Studies by Norgard and co-workers demonstrated a link between one TCS, Hk2-Rrp2, and the alternative sigma factors RpoN and RpoS [14], [15]. The expression of several virulence genes, including ospC, dbpA and bbk32, are dependent on RpoS [14], [16], [17]. RpoS is also essential for repression of genes whose expression is required during the tick phase, but not in the mammalian host [17], [18]. BB0647 (BosR, Fur) has also been shown to play a role in RpoN-dependent expression of rpoS [19]–[21]. Less is currently known regarding the second TCS, consisting of Hk1 and Rrp1, but recent studies have begun to elucidate the processes that are regulated by this TCS [22]–[24]. In particular, Rrp1 has been shown to be responsible for production of bis-(3′-5′)-cyclic dimeric guanosine monophosphate (c-di-GMP) and mutagenesis of Rrp1 results in alteration of expression for a substantial number of genes, including those involved in uptake and dissimilation of glycerol [22], [23], [25].
Different carbohydrates are selectively available to B. burgdorferi during its enzootic cycle. Glucose is the primary carbohydrate constituent in mammalian blood [26], [27] and B. burgdorferi can use glucose to support growth [28]. Ticks rely on a high concentration of carbohydrates and other nutrients available in the blood meal for molting and successful oogenesis [29], [30]. During feeding, ticks create a peritrophic matrix above the epithelial cell layer, which serves both as a compartment to trap the blood meal and as a barrier to prevent invasion by microorganisms that accompany the blood meal. However, the peritrophic matrix is permeable to hexose sugars [29], [31]. Once hexoses permeate across the matrix, they are sequestered by midgut epithelial cells during larval feeding [29], [31]. Consequently, nutrients present in the blood meal are rapidly depleted during larval feeding and are likely non-existent in an unfed nymphal midgut. Therefore, spirochetes resident in the midgut must identify and utilize alternative carbohydrates until the unfed nymph takes its next blood meal.
Glycerol, a diffusible carbohydrate, is a readily available nutrient in the tick. Glycerol is produced by Ixodes spp. and serves as a colligative antifreeze for tick survival during the winter [32]–[34]. B. burgdorferi encodes a putative glycerol utilization operon consisting of three genes. glpF (bb0240) encodes a putative transmembrane facilitator protein that mediates the entry of free glycerol into the cell. glpK (bb0241) encodes a putative kinase that would produce glycerol 3-phosphate (G3P) which would be the substrate for glycerol 3-phosphate dehydrogenase (G3PDH), an enzyme putatively encoded by the third gene in the operon, bb0243. The resulting product, dihydroxyacetone phosphate, can enter glycolysis through the action of triose phosphate isomerase and ultimately result in the net production of one ATP molecule per original glycerol molecule [12], [28]. Alternatively, G3P may be converted to phosphatidic acid through the action of two enzymes, BB0327 (G3P acyltransferase) and BB0037 (Lysophosphatidic acid acyltransferase); this pathway is required for phospholipid biosynthesis and production of new cell membrane [12] (Figure 1).
Three lines of evidence suggest that glycerol utilization may be important during the vector phase of the enzootic cycle. Ojaimi et al. reported that all genes of the glycerol utilization operon are more highly expressed during in vitro growth in BSK-II medium at 23°C as compared to growth at 35°C [35]. Caimano et al. demonstrated that repression of glp operon expression is dependent on RpoS within the mammalian host [17]. Moreover, constitutive expression of the glp operon partially restores the ability of Rrp1-deficient B. burgdorferi to survive within feeding ticks [23].
In order to elucidate the role of glycerol uptake and utilization by B. burgdorferi during its natural life cycle and the regulatory events that govern glp operon expression, the gene predicted to encode G3PDH (bb0243) was disrupted and the effects of mutagenesis were evaluated in vitro and during infection of ticks or mice. The results demonstrate that the ability to utilize glycerol as a carbohydrate for use in the glycolytic pathway during the tick phase of the infectious cycle is critical for maximal B. burgdorferi fitness.
The B. burgdorferi G3PDH genomic sequence has been annotated to putatively encode the anaerobic form of the enzyme based on its similarity to the anaerobic G3PDH ortholog of Haemophilus influenzae strain Rd (glpA) [12]. Other organisms containing an anaerobic GlpA (such as E. coli) contain two additional subunits as part of the functional G3PDH enzyme; GlpB, a subunit involved in FMN binding [36] and GlpC, a small membrane anchoring subunit [37]. Together, the individual protein molecules form a functional GlpABC heterotrimer [36]. Whole genome sequencing of B. burgdorferi failed to reveal putative genes with homology to any known glpB or glpC orthologs [12]. Further, BLASTP analysis revealed no orthologs in B. burgdorferi with similarity to either E. coli strain K12 or H. influenzae strain Rd GlpB or GlpC.
The tertiary structure of B. burgdorferi G3PDH was predicted using the SWISS-MODEL server [38]–[40] by comparison to E. coli K12 aerobic GlpD and anaerobic GlpA, as well as H. influenzae strain Rd anaerobic GlpA. B. burgdorferi G3PDH autoaligned with E. coli aerobic GlpD (PDB 2QCU), but not with E. coli GlpA [41] (Figure 2). E. coli anaerobic GlpA and H. influenzae GlpA auto-aligned to the anaerobic GlpA of Bacillus halodurans (PDB 3DA1) [42] (Figure 2). The modeling predicts that B. burgdorferi G3PDH and E. coli aerobic GlpD share similar tertiary structures. Yeh et al. have described 14 amino acid residues that participate in the E. coli GlpD active site based on a 1.75 Å structural model [41]. B. burgdorferi G3PDH contains conserved residues at 12/14 positions, in contrast to the E. coli and H. influenzae GlpA proteins (9/14). Taken together, the bioinformatic analyses suggest that B. burgdorferi G3PDH has greater similarity to aerobic forms of the enzyme. We propose that annotation of bb0243 should be changed to indicate that it putatively encodes an aerobic GlpD and B. burgdorferi G3PDH is referred to as GlpD in the remainder of this report.
The physical linkage of bb0240, bb0241, and bb0243 in the B. burgdorferi chromosome suggests that these genes comprise an operon [35]. RT-PCR analysis using RNA extracted from B. burgdorferi strain B31-A3 revealed that these genes are transcribed as a single operon (Figure 3). Ojaimi et al. reported that the B. burgdorferi glycerol operon is more highly transcribed at 23°C relative to transcript levels in cells grown at 35°C [35]. In order to explore if this increased transcript level is reflected in protein, strain B31-A3 whole cell lysate was tested to determine the protein expression levels at these two temperatures. Immunoblot analysis revealed that when B. burgdorferi strain B31-A3 was grown at 25°C, 7-fold more GlpD was generated compared to the level in cells grown at 37°C (Figure 4).
To explore the possibility that expression may be differentially regulated in vivo, glp transcript levels were measured in infected ticks or mouse joints by real time RT-PCR; transcription of ospA and ospC was monitored as a control. Expression of the latter genes followed the expected pattern; ospA was expressed exclusively in ticks and ospC transcript was detected only in feeding nymphs and mouse joints (Figure 5). glpD expression was substantially higher during all tick stages (fed larvae, 3.68±2.72 copies/10 copies of flaB; unfed nymphs, 5.31±4.42; fed nymphs, 4.97±0.74) than in mouse joints (1.45±1.98 copies/10 copies of flaB). A similar expression pattern was observed for glpF, the first gene in the operon (fed larvae, 4.42±1.07 copies/10 copies of flaB; unfed nymphs, 1.48±0.61; fed nymphs, 3.70±1.89; mouse joint, 0.13±0.23) (Figure 5). Caimano et al. showed that transcription of glp operon genes is subject to RpoS-dependent repression [17]. This was confirmed at the protein level for GlpD as shown in Figure 6. Wild-type or RpoS mutant cells were grown in vitro at either 23°C or 37°C or in dialysis membrane chambers (DMCs) implanted in rat peritoneal cavities. Induction of OspC expression and repression of OspA expression in DMCs confirmed that B. burgdorferi attained the host-adapted state and abrogation of these changes in expression in the RpoS mutant showed that these alterations were dependent on RpoS, as expected. GlpD expression was virtually abolished in wild-type B. burgdorferi grown in DMC and this effect was not observed in the RpoS mutant cells (figure 6).
To study the role of glycerol utilization in B. burgdorferi, glpD, the distal ORF in the glycerol operon was inactivated in strain B31-A3 by disruption with a flgB-aadA cassette inserted at residue K149 (Figure 7). Three mutants, two with the flgB-aadA cassette in the same orientation as the operon (CP176, CP177) and one with the insert in the opposite orientation (CP257), were isolated (Figure 8A). Southern blot analysis confirmed a disruption in glpD and showed that recombination occurred by a double crossover event in all three mutants (Figure 8B). Western blot analysis revealed that GlpD was absent in the mutants (Figure 9). Analysis of plasmid content of the wild type by PCR revealed that it lacked lp5 and cp9 and contained all other B31 linear plasmids, including those essential for murine infectivity. GlpD mutants had the same plasmid profile as the parental strain (data not shown).
Repeated attempts to isolate a complemented mutant strain were unsuccessful. Most experiments described below were carried out with all three isolated glpD mutants. Mutants CP176 and CP177 were isolated from one transformation and CP257 was obtained independently, thereby mitigating the concern that the observed mutant phenotypes were the result of a second site mutation.
To begin to characterize the role of GlpD in B. burgdorferi physiology, growth of glpD mutants was compared to that of wild-type B31-A3 in BSK-II, an undefined, enriched medium that contains glucose as the principal carbohydrate source [43]. No differences in final cell density were detected between wild-type and glpD mutants at either 25°C (1.3×109 and 1.1×109, respectively) or 37°C (1.4×109 and 1.3×109, respectively). In addition, no difference in growth characteristics was observed (data not shown).
Previous studies have shown that N-acetyl glucosamine (GlcNAc) is required for growth of B. burgdorferi in vitro [44]–[46]. There were no differences in growth characteristics between B31-A3 and CP176 grown in a modified BSK-II medium that did not contain glucose but had GlcNAc as the carbohydrate source (BSK-lite [28]) (Figure 10A). In contrast to growth in BSK-II with GlcNAc only or medium supplemented with glucose (data not shown), there was a significant difference in growth between wild type and glpD mutants when glycerol was supplied as the principal carbohydrate source. B31-A3 reached a significantly higher cell density in BSK-glycerol medium compared to CP176 when grown at 25°C (6.4×108 and 1.1×108, respectively; P<0.001) (Figure 10B). Interestingly, this effect was observed only at 25°C; when cultivated at 37°C, the growth characteristics of wild type and CP176 were indistinguishable. Indeed, B31-A3 cultures achieved significantly higher cell densities at 25°C as compared to 37°C in BSK-glycerol medium (6.4×108 and 9.1×107, respectively; P<0.001). These experiments were repeated with the other two independent glpD mutants (CP177, CP257) with essentially identical results (data not shown). These findings suggest that B. burgdorferi can utilize glycerol to support enhanced growth at the lower temperature. This observation would be consistent with the elevated expression of GlpD at 25°C (Figure 4).
In order to determine whether the absence of GlpD affects the pathogenic properties of B. burgdorferi, C3H/HeJ mice were needle inoculated with 1×104 cells of either wild type or glpD mutant. All mice in both the wild type and glpD mutant groups were infected and were seropositive by four weeks post-inoculation (Table 1). Unfed larvae were allowed to feed to repletion on these infected mice to allow spirochete acquisition by ticks. Infected fed larvae that molted to nymphs were fed to repletion on naïve C3H/HeJ mice. Viable spirochetes were recovered from all mice that were fed on by either wild type- or glpD mutant-infected ticks and seroconverted by 4 weeks post-feeding (Table 1). These results demonstrate that GlpD is not required for murine infection by B. burgdorferi. Further, GlpD-deficient spirochetes were acquired by larvae fed on infected mice, persisted through the molt and were transmitted to naïve mice by infected nymphs.
The enhanced growth of B. burgdorferi at lower ambient temperature in vitro when glycerol is the principal carbohydrate source, elevated expression of GlpD at the lower temperature and its RpoS-dependent repression in DMCs suggested that glycerol may be an important nutrient for B. burgdorferi during the tick phase of its life cycle. Therefore, the effect of glpD disruption was explored more extensively in infected ticks. Naïve, unfed larvae were placed on mice infected with either B31-A3 or CP176, allowed to feed until repletion and molt to the nymphal stage. Spirochete loads in infected ticks were measured by qPCR (Table 2). Larvae infected with either B31-A3 or CP176 had similar spirochete loads (approximately 700 spirochetes/larvae).
Spirochete numbers in wild type-infected ticks increased slightly after larval molting to nymphs but decreased in nymphs infected with any of the glpD mutant strains. This resulted in a significant five-fold decrease in CP176 density in unfed nymphs compared to the wild-type (Table 2). Further, in independent experiments, CP177 and CP257 had an identical phenotype to that observed for CP176, i.e. spirochete densities were significantly lower after molting as compared to spirochete loads in B31-A3-infected ticks (data not shown).
Infected nymphs were fed on naïve mice and spirochete loads were measured in the resulting fed nymphs. As expected, spirochete numbers increased substantially during nymphal feeding in both wild type- and CP176-infected nymphs, although the spirochete burdens in the glpD-infected engorged nymphs were significantly lower than in nymphs infected with the parental strain (P<.02) (Table 2). These results suggest a role for glycerol utilization by B. burgdorferi as an important factor for spirochete maintenance during transtadial transition.
As previously described, wild type- and glpD mutant-infected nymphs are equally capable of transmitting B. burgdorferi to, and causing infection in, mice (Table 1). However, those studies were conducted by allowing nymphs to feed to repletion. A feeding nymph will attach and feed on a host for 72 hours or longer [47]. During this time, replicating B. burgdorferi surround midgut epithelial cells, penetrate the midgut basement membrane, and enter the hemocoel and salivary glands from which they are ultimately transmitted to the mammal [8], [9]. Therefore, replication is a critical step in spirochete transmission from the vector to the mammalian host. Since the density of glpD mutant spirochetes decreases as a result of molting and was five-fold lower than in wild type-infected unfed nymphs, we reasoned that nymphs harboring glpD mutant spirochetes would require a longer feeding period before transmission to the host due to its delayed exit from the tick midgut.
To explore this possibility, B31-A3- and CP176-infected nymphs were placed on naïve mice, allowed to begin feeding, but forcibly removed at different time points post-attachment. Mice were then monitored for evidence of infection. In a pilot experiment, nymphs were fed on naïve mice for 65 hours; 2/2 mice fed on by B31-A3-infected nymphs became infected, whereas 0/3 mice fed on by CP176-infected nymphs acquired infection. Based on this pilot study, B31-A3- or CP176-infected unfed nymphs were placed on the outer ear of naïve C3H/HeJ mice and allowed to feed for either 24, 48, 55, 62 or 72 hours or collected at drop off (>72 hours). Ticks were removed at each time point and spirochete load was determined by qPCR. Results presented in Figure 11 demonstrate that CP176 experienced a lag in replication and achieved significantly lower spirochete loads at times beyond 48 hours of feeding (P<0.001). At these points, spirochete loads per tick were 3.5–5 fold lower in CP176-infected ticks than in those infected with B31-A3 (e.g., at 55 hours of feeding spirochete loads were 42,633 and 8,581 for B31-A3 and CP176-infected nymphs, respectively) (Figure 11). Mice were also monitored for infection. Results demonstrate that mice fed on by wild type-infected nymphs were infected by 62 hours of feeding. In contrast, CP176-infected nymphs produced infection in mice only after at least 72 hours of feeding (Table 3). These data suggest that wild type-infected nymphs are more readily able to infect naïve mice due to a more rapid increase in spirochete density induced on commencement of tick feeding. Further, disruption of glycerol utilization results in reduced fitness of the spirochete during the tick phase and spirochetes that are unable to utilize glycerol are at a disadvantage for transmission to a mammalian host.
Chitobiose is a di-GlcNAc molecule that is a component of the peritrophic matrix and tick chitin [45], [46], [48], [49]. The B. burgdorferi genome contains open reading frames that encode gene products that can mediate chitobiose transport and metabolism [12], [46], [48], [49]. These include the three subunits of a chitobiose transporter (BBB04-BBB06), a putative chitobiase (BB0002) and NagA and NagB (BB0151 and BB0152). In combination, the actions of these gene products would result in production of glucose 6-phosphate that could enter the glycolytic pathway (Figure 1). Both chitobiose and chitin can support B. burgdorferi growth in vitro [45], [46], [48], [49]. Therefore, the transcript levels for chbC (bbb04), which encodes subunit C of the chitobiose transporter, were also measured to determine whether chitobiose utilization by B. burgdorferi may be important during the tick phase of the enzootic cycle. Substantially higher expression of chbC occurred during the various tick stages (fed larvae, 5.10±7.00 copies/10 copies of flaB; unfed nymphs, 4.42±2.33; fed nymphs, 2.87±0.36) as compared to expression in mouse joints (0.59±0.71 copies/10 copies of flaB) (figure 5).
On acquisition by feeding larvae from an infected mammal, B. burgdorferi must initially adapt to the new host (i.e. tick) environment. The spirochete must then survive the tick molting process and endure a substantial period in a nutrient-poor milieu (unfed nymph). This is not a period of metabolic dormancy since several studies, including our own, demonstrate that B. burgdorferi gene expression is modulated in different tick developmental stages and that expression of some genes is higher in unfed nymphs than in fed nymphs [17], [50], [51]. During the subsequent nymphal blood meal, B. burgdorferi enter a rapid replication phase, experiencing a significant increase in density within a 48 hr period [8], [9] and must prepare for transmission back to a mammal. How does B. burgdorferi generate the energy required to withstand this harsh environment?
Glycerol and its metabolites play important roles in cellular biochemistry [52] and glycerol is a readily available carbohydrate in Ixodes ticks [32]–[34]. Most bacteria have the ability to acquire glycerol from the surrounding milieu or to re-utilize it from its own metabolites [52], [53]. G3P is a crucial intermediate for energy metabolism (via its conversion to dihydoxyacetone phosphate and entry into the glycolytic pathway) and for phospholipid biosynthesis (via its conversion to phosphatidic acid [Figure 1]). The B. burgdorferi genome putatively encodes all the enzymes required for both processes [12], [54]. The possibility that glycerol uptake and utilization may play an important role during the tick phase of the enzootic cycle was suggested by previous studies showing that genes comprising the glp operon had elevated expression during growth in vitro at 23°C and were subject to RpoS-dependent repression within the mammalian host [17], [35]. A number of findings from the current study confirm that this is the case. First, in medium supplemented with glycerol as the principal carbohydrate source, wild-type B. burgdorferi grew to a significantly higher cell density compared to a glpD mutant during growth at 25°C (Figure 10B). This difference was not observed during growth at 37°C or when glucose was employed as the principal carbohydrate source. Second, transcript levels for glpF and glpD were significantly lower in mouse joints relative to their levels in ticks (Figure 5). Third, GlpD protein was not produced during growth in DMCs and its repression was dependent on the presence of RpoS (Figure 6). Finally, the glpD mutant was fully infectious in mice when introduced by either needle or tick inoculation (Table 1), but had a replication defect in ticks (Table 3 and Figure 11).
Absence of GlpD results in reduced spirochete fitness in the tick. This defect is manifested at two distinct points during this phase of the enzootic cycle. Spirochete loads are reduced five-fold after the larval molt in the mutant relative to the wild type. Spirochete loads were measured in larvae that had fed to repletion and in unfed nymphs two weeks after the molt. Therefore, it is not clear whether the reduction in B. burgdorferi density occurred during the molt or during the initial period in the unfed nymph. We favor the latter possibility. At the onset of nymphal feeding, the glpD mutants display a lag prior to beginning replication; as a result, they replicate more slowly than wild-type spirochetes and fail to achieve the same final spirochete densities (Figure 11). As a consequence, there is delayed transmission of glpD mutant spirochetes to mice during feeding. Whereas ticks infected with wild-type B. burgdorferi caused infection by 62 hours of feeding, those infected with mutant spirochetes required at least 72 hours of feeding before productive transmission occurred. Dunham-Ems et al. have demonstrated that B. burgdorferi migration from the midgut to the salivary glands for transmission to a mammal proceeds in two phases. In the initial step, replicating spirochetes form non-motile networks that advance toward the basolateral surface of the gut epithelium. The non-motile spirochetes then transition to motile organisms that penetrate the basement membrane into the hemocoel and migrate to the salivary gland [9]. This model of B. burgdorferi dissemination provides an explanation for the delayed transmission phenotype of the glpD mutant. As dissemination of B. burgdorferi in the tick during the first phase of feeding does not depend on motility, but instead is replication driven, the reduced replication rate of the mutant would result in delayed dissemination to the hemocoel. As a result, the mutant would require additional time for successful tick to mammal transmission.
Spirochete loads of wild type and glpD mutant were identical in fed larvae whereas those of the mutant were reduced approximately five-fold after the larval molt; the reduced level of mutant persisted throughout the subsequent stages of the tick cycle (Table 2). In a very recent study, He et al. showed that a glpF polar deletion mutant, which does not express any of the glp operon genes, had a phenotype both in vitro and in vivo very similar to that described here for a glpD mutant (i.e. the mutant failed to reach the same cell density as the wild type when cultured in medium with glycerol as the principal carbohydrate and had reduced spirochete loads in infected nymphs) [23]. Interestingly, the reduction in mutant spirochete levels in nymphs was much more severe (>2 logs) in their study than was observed here for the glpD mutant. Presumably, this difference is due to the fact that the glpD mutant will only have an effect on glycerol utilization for glycolysis, whereas the glp operon mutant will also affect phospholipid biosynthesis (Figure 1). Thus, study of the glpD mutant is important in allowing evaluation of the contribution of glycerol utilization for energy metabolism without any confounding from effects on other metabolic pathways.
Why isn't the glpD mutation lethal rather than being simply growth inhibitory? Clearly, there must be an alternative carbohydrate source that can be metabolized via glycolysis to produce the required ATP. We propose that this alternative energy source is chitobiose. Tilly et al. reported that chbC transcript is elevated at 23°C relative to 34°C [45] and we have found that chbC expression is significantly higher in ticks than in mouse joints (Figure 5). Taken together, these data suggest that chitobiose utilization by B. burgdorferi is important during the tick phase of the cycle. Chitobiose would be available to the spirochete during tick feeding, when it is shed from the forming peritrophic matrix, as well as during molting when the tick cuticle is being re-modeled for growth [31], [49]. It has been suggested that chitobiose utilization would be essential for B. burgdorferi during the tick phase of the enzootic cycle for spirochete glycolysis and cell wall synthesis. However, chbC mutants, which cannot take up exogenous chitobiose or utilize chitin to support growth, successfully complete the mouse-tick-mouse infectious cycle [48]. It is possible that glycerol availability may be partially responsible for rescue of the chbC mutant. This would suggest that B. burgdorferi maintains spirochete fitness in the nutrient deplete environment of the tick midgut by utilizing either glycerol and/or chitobiose as glycolytic precursors. It would be of interest to determine whether a glpD-chbC double mutant would be capable of completing the natural infectious cycle.
As described earlier, signals that lead to phosphorylation of Rrp2 result in activation of RpoN which, in turn, initiates transcription of rpoS [14], [15]. RpoS is expressed only in feeding nymphs and mammals and therefore, is thought to be responsible for the regulon that is required for mammalian infection [17], [55]. This is consistent with the fact that Rrp2, RpoN and RpoS mutants cannot establish infection in mice [16], [56], [57]. The RpoS regulon includes genes that are absolutely dependent on RpoS for their transcription (e.g. ospC), as well as genes subject to RpoS-dependent repression [17], [18]. The glp operon is in the latter category. Several recent studies have begun to reveal the role of the Hk1/Rrp1 TCS in B. burgdorferi [22]–[24]. Lack of Hk1 and Rrp1 has no effect on infectivity in mice. However, Hk1 and Rrp1 mutants are killed within the tick midgut during feeding [23], [24]. Hk1 and Rrp1 appear to be expressed during all stages of the B. burgdorferi life cycle [22], [24], but the important consideration is whether Rrp1 is phosphorylated leading to the production of c-di-GMP. Interestingly, Caimano et al. have recently shown that Hk1 mutants are killed during the larval and nymphal blood meals, indicating that c-di-GMP is required during both tick feeding stages [24]. It is reasonable to conclude that the Rrp2/RpoN/RpoS pathway governs the expression of B. burgdorferi genes required in the mammalian host, whereas Hk1/Rrp1 controls a subset of borrelial gene products that is critical for survival in the tick vector. A number of genes that are induced by Rrp1 (i.e. c-di-GMP) are subject to RpoS-dependent repression; these include the glp operon genes and bba74 [22], [23].
The current study demonstrates that glp operon expression is modulated by nutrient availability. Several reports have established that this operon is regulated in a reciprocal manner by RpoS and Rrp1 [17], [22], [23]. The glp genes are the first borrelial gene products linked to spirochete metabolism whose expression is subject to regulation by both B. burgdorferi TCSs. As such, these genes represent a valuable paradigm for elucidating the interplay between these two regulatory pathways. A model that integrates both carbohydrate availability and presence/absence of transcriptional regulators is presented in Figure 12. It is presumed that early in larval feeding RpoS is still present, but must be degraded in order to allow for expression of tick phase genes; the precise timing of RpoS disappearance is not currently known. Glucose should be available at this stage in quantities sufficient to support B. burgdorferi growth. Later in the blood meal hexose sugars and other serum constituents crossing the peritrophic matrix are sequestered by tick midgut epithelial cells. This creates a nutrient-poor environment in which B. burgdorferi must rely on glycerol and rapidly depleting chitobiose to support glycolysis. The turnover of RpoS during larval feeding would result in the de-repression of glycerol pathway enzymes and presence of c-di-GMP will activate their expression, ensuring that spirochetes can rapidly switch from a glucose-based to a glycerol-based metabolism. Once spirochete infection is established in the midgut and the larva molts to an unfed nymph, B. burgdorferi remains a metabolically active spirochete that must rely on glycolysis for maintenance of cellular integrity. Glycerol is presumed to be the primary carbohydrate at this stage and absence of RpoS would allow expression of the glp operon. Early in nymphal feeding while blood constituents are scarce, B. burgdorferi must actively replicate and migrate through the epithelial midgut lumen to begin its migration to the salivary glands. At this point glycerol would be a primary energy source for support of cellular replication, as chitobiose will not be available until the eventual breakdown of the peritrophic matrix. Inability to utilize glycerol, as in the glpD mutant, would result in delayed and reduced spirochete replication that could impact transmission of B. burgdorferi to the mammalian host. Presence of c-di-GMP would result in sustained expression of glp operon genes. As RpoS levels increase during the nymphal blood meal, presence of c-di-GMP will counteract the repressive effects of RpoS on glp gene expression, ensuring that glycerol utilization can continue until the spirochetes are transmitted to a mammalian host.
The model presented in Figure 12 accounts for carbohydrate source availability and presence of regulatory molecules throughout the tick-mouse enzootic cycle and highlights the interdependence of these two parameters. Concentrations of glucose and glycerol in mouse plasma are approximately 150 and 2.8 mg/100 mL, respectively [58]. Glycerol is abundantly present during all tick stages. On this basis, the model assumes that B. burgdorferi utilizes glucose as the preferred nutrient source when it is available in the mammal or at certain stages during the tick blood meal, but switches to utilizing glycerol, especially in the unfed nymph when glucose is not present. This may represent the B. burgdorferi version of carbon catabolite repression (CCR), which is defined as a regulatory mechanism by which the expression and enzymatic activities of enzymes involved in the use of secondary carbohydrates are reduced in the presence of sufficient levels of the preferred carbohydrate [59]. The mechanisms underlying CCR in most bacteria involve a glucose-specific phosphotranferase subunit (EIIA) that can be reversibly phosphorylated based on the phosphoenolpyruvate to pyruvate ratios. B. burgdorferi encodes a putative EIIA subunit (BB0559), but does not appear to contain other major components that modulate CCR in other bacteria [12]. It is reasonable to assume that B. burgdorferi possesses sensing mechanisms that monitor the relative levels of glucose and glycerol in the environment. It is tempting to speculate that modulation of Hk1 kinase activity is one outcome of the fluctuating nutrient ratios. When the glycerol/glucose ratio increases Hk1 would phosphorylate Rrp1 leading to the production of c-di-GMP. The specific molecular signal that is recognized by Hk1 is not currently known. Likewise, the precise timing of the transcriptional activation/repression of RpoS and the possible reciprocal modulation of c-di-GMP levels is not known. Studies designed to elucidate the molecular events underlying the proposed model are warranted.
All animal experimentation was conducted in strict accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were approved by the Institutional Animal Care and Use Committee of New York Medical College (Approval number 31-1-0310H).
B. burgdorferi strains B31-A3 [60], 297 (c162) and a strain 297-based RpoS mutant (c174) [17] were employed in this study. Spirochetes were grown in modified Barbour-Stoenner-Kelley-II medium [43], [61] supplemented with 6% heat inactivated rabbit serum (Sigma, St. Louis, MO) (BSK-II). BSK-lite medium was based on the formulation of Barbour [43] with modifications as previously described [28].
B. burgdorferi were grown to late log phase (5–10×107 cells/ml) in BSK-II medium at 25°C. For BSK-lite experiments, spirochetes were diluted 100 fold in BSK-lite medium to remove BSK-II medium constituents. 5×104 spirochetes in 40 ml of BSK-lite medium with a specific carbohydrate (either glucose or glycerol) were aliquoted into eight 5 ml tubes. Four tubes of each sample were placed at either 25°C or 37°C and observed for up to 60 days. Individual tubes were counted daily for cultures grown at 37°C and every two days for those incubated at 25°C. Spirochete density was enumerated by dark field microscopy as previously described [62]. Student's two-tailed, unpaired t-tests were performed on data collected during exponential phase and stationary phases of cell growth. Significance was defined as a P<0.01.
Cultivation of c162 and c174 in DMCs was carried out as described [63].
The strategy for disruption of B. burgdorferi is presented in figure 7. A 2519 bp region of B. burgdorferi chromosomal DNA containing bb0243 was amplified by PCR using primers bb0243F/R (Table 4) ligated into the pGEM-T vector (Promega, Madison, WI), transformed into E. coli DH5α and cells containing recombinant plasmids were selected by blue-white screening. Selected transformants were purified to single colonies and plasmids were confirmed to contain bb0243 by DNA sequence analysis (Davis Sequencing, Davis, CA). The pGEM-T-bb0243 construct was digested with DraII (corresponding to position 248,983 in the B. burgdorferi chromosome). A DNA fragment containing flgB-aadA (a spectinomycin/streptomycin resistance cassette driven by the B. burgdorferi flgB promoter) was amplified from vector pKFSS1 as previously described [64] and inserted into the DraII site within PGEM-T-bb0243 by blunt-end ligation. The construct was transformed into E. coli DH5α and selected by growth on LB agar plates supplemented with 100 µg/ml of spectinomycin. Transformants harboring plasmids containing bb0243 disrupted by the flgB-aadA cassette were isolated, purified to single colonies and plasmid inserts were confirmed by PCR and DNA sequence analysis. flgB-aadA cassette orientation was determined by restriction enzyme digestion. A plasmid construct designated pCP100 had the flgB-aadA cassette in the same orientation as bb0243 and a plasmid construct designated pCP200 had the flgB-aadA cassette in the reverse orientation. The ampicillin resistance cassette (bla) located in pGEM-T was disrupted in both constructs as previously described [65] yielding a plasmid designated pCP101 from pCP100 and pCP201 from pCP200. Spectinomycin-resistant, ampicillin-sensitive colonies of each construct were selected by growth on LB agar containing either 100 µg/ml ampicillin or 100 µg/ml spectinomycin.
pCP101 and pCP201 were isolated and transformed into B. burgdorferi B31-A3 competent cells by electroporation as described [66]. Transformants were screened by growth in a 96 well plate in the presence of streptomycin (100 µg/ml). Selected transformants were cloned by limiting dilution in BSK-II medium containing streptomycin (100 µg/ml). The glpD disruption in selected transformants was confirmed by Southern blot and Western blot analyses (Figures 8 and 9). Plasmid content for selected mutants was determined as previously described [67] to ensure that all plasmids essential for murine infectivity were present.
Southern blot analysis and generation of a digoxygenin-labeled bb0243 probe were performed as previously described [67] with the following modifications. A 196 base pair fragment of bb0243 was generated from strain B31-A3 by PCR using primers 243probeF/R (Table 4). B. burgdorferi DNA was fragmented by incubation with 4.5 units of BamHI in 1× buffer B (Fermentas, Glen Burnie, MD) or EcoRI in buffer EcoRI (Fermentas) overnight at 37°C.
C3H/HeJ mice (Jackson Laboratories, Bar Harbor, ME) were infected with either wild-type or glpD mutant B. burgdorferi by needle inoculation as previously described [68], [69]. Once infection was established as determined by culture of ear biopsy, mice were anesthetized with ketamine and 100–300 naïve, unfed larvae were placed in and around the ear canal. Mice were placed individually into cages with approximately 1 cm water at the bottom. A metal grid of the same length and width as the cage, and standing 1.5 cm high, was placed in the cage. Larvae were allowed to feed until repletion. Following drop off, larvae were collected, rinsed in water, pooled into groups of 30 in 5 ml tubes with a porous cover and maintained in a desiccator at 21°C, >95% relative humidity with a 16 hour∶8 hour light: dark cycle. Larvae molted to unfed nymphs in approximately 5–6 weeks. At four weeks post molt, three unfed nymphs were placed on three-week old uninfected C3H/HeJ mice (Jackson) and allowed to feed until repletion. Fed nymphs were collected as described above.
For interrupted feeding experiments, 3 unfed nymphs were allowed to feed on naïve C3H/HeJ mice for 24, 48, 55, 62, and 72 hours or to repletion. Ticks were carefully removed from mice by forceps at the indicated time point. Ticks were processed for DNA isolation as described below. Mice were tested for infection as previously described [68], [69].
DNA isolation from 5×108 B. burgdorferi was performed using the Puregene DNA isolation kit as per manufacturer's instructions (Qiagen, Valencia CA). DNA pellets were resuspended in 30 µl of nuclease free water. DNA concentration was measured by spectrophotometric analysis at 260 nm.
DNA was isolated from pools of 10 fed larvae. DNA was obtained from unfed nymphs that were processed either individually, in groups of 5 or in groups of 10. DNA was isolated from individually processed fed nymphs. Ticks were surface sterilized by washing successively with 800 µl of sterile H2O, 0.5% sodium hypochlorite, 3% hydrogen peroxide (Sigma), 70% ethanol (Fischer Scientific, Pittsburgh, PA) and sterile H2O each for 1 minute. DNA extraction was performed as adapted from the Qiagen DNeasy blood and tissue kit as described by Beati et al. [70] with the following modifications. Ticks were homogenized with an 18.5 gauge needle. Samples were lysed with 220 µl animal lysis buffer and 0.45 mg recombinant proteinase K (Roche, Mannheim, Germany) per reaction overnight in a 56°C incubator. Following all wash steps, mixtures were centrifuged at 10,000 rpm. DNA was eluted twice with 25 µl of PCR-grade H2O pre-warmed to 72°C.
qPCR reaction mixtures (25 µl total volume) contained 2 µl of sample DNA, 3 µl of nuclease free water, 20 pmol each of primers FL-571F/FL-677R, 5 pmol of flaB- specific Taqman probe (flaBFAM) (table 4), and 12.5 µl Taqman PCR mastermix (Roche). DNA copy number was determined on an ABI prism 7900HT thermocycler with an amplification profile of 50°C for 2 minutes, 95°C for 10 minutes, followed by 40 cycles of 95°C for 15 seconds and 60°C for 1 minute. Samples were run in duplicate and each plate contained two samples lacking DNA as negative controls. Ct values were obtained using the SDS2.1 software program (Applied Biosystems, Carlsbad, CA). To assess spirochete density per sample, standard curves were generated for flaB, a constitutively expressed gene, in log increments (10–104).
Copy numbers were compared by a two-tailed, unpaired t-test for each condition (fed larvae, unfed nymph, fed nymph), where significance was defined as P≤0.05.
Ticks infected with B. burgdorferi were processed in pools of 50 for fed larvae, 100 for unfed nymphs and 35 for fed nymphs. Ticks were homogenized in 1 ml TRIzol reagent (Invitrogen, Carlsbad, CA) for 5 minutes. For in vitro experiments, 50 ml of cell culture (approximately 2.5×109 cells) was centrifuged at 12,000 rpm for 10 minutes and 1 ml TRIzol was added to the cell pellet. RNA was recovered as per manufacturer's instructions (Invitrogen). The RNA pellet was resuspended in 30 µl nuclease-free water and DNase treated twice with the Ambion DNA free kit per manufacturer's instructions (Ambion, Austin, TX).
Mammalian hind limb joints were surgically removed from euthanized C3H/HeJ mice, snap frozen in liquid nitrogen and pulverized with mortar and pestle. The powdered tissue was transferred to a glass homogenizer and homogenized with 0.5 ml denaturation solution and the supernatant containing total RNA was isolated from samples as per manufacturer's instructions (ToTALLY RNA, Ambion). RNA was rehydrated in 30 µl of nuclease free water and DNase treated as described above. Mouse RNA was removed by MICROBEnrich as per manufacturer's instructions (Ambion). The recovered RNA pellet was resuspended in 15 µl of sodium citrate buffer (Ambion) and 15 µl of nuclease free water.
cDNA was generated from RNA samples by addition of 2 µg of purified RNA to a mixture containing 4 µl of 5× reverse transcriptase buffer (Promega), 0.02 mM dNTPs (Roche), 0.5 µg random hexamer (Promega), 2 units of RNase inhibitor (Ambion), 5 units of AMV reverse transcriptase enzyme (Promega) and nuclease free water in 20 µl total volume. The reaction mixture was incubated at 42°C for 2 hours. Reverse transcriptase enzyme was heat inactivated at 95°C for 5 minutes and cDNA was stored at −20°C until further use.
For generation of standard curves, specific gene fragments for bb0240, bb0243 and bbb04 were amplified by PCR using primers pairs bb0240qRTPCRF/bb0240qRTPCRR, bb0243qRTPCRF/bb0243qRTPCRR, bbb04qRT-PCRF/bbb04qRT-PCRR, ospA-288F/ospA-369R and ospC-B31FTq/ospC-B31RTq, respectively (table 4). PCR reaction mixtures contained 100 ng of B31-A3 DNA, 0.25 µl Taq polymerase (Roche), 0.5 µl dNTPs (Roche), and 1× Taq polymerase buffer (Roche) in a total volume of 25 µl. Amplification conditions were 95°C for 5 minutes, followed by 36 cycles of 95°C for 30 seconds, 55°C for 30 seconds and 72°C for 30 seconds and a final incubation at 72°C for 10 minutes. Production of the expected product was confirmed by gel electrophoresis and the PCR products were ligated into the TOPO 2.1 cloning vector, the vector was transformed into E. coli Mach1 cells and recombinant clones were selected as per manufacturer's instructions (Invitrogen). Clonal isolates were grown in 10 ml of LB broth supplemented with 100 µg/ml ampicillin and the plasmids were extracted as described above. PCR confirmed the presence of the desired gene fragments. Plasmid concentration was determined by spectrophotometric analysis at 260 nm, followed by mathematical computation of copy number (http://www.uri.edu/research/gsc/resources/cndna.html).
Transcript levels for bb0240 (glpF), bb0243 (glpD), bbb04 (chbC), ospA, ospC and flaB were determined by performing qRT-PCR as previously described [65] using the primer pairs listed in table 4 on an ABI Prism 7900HT thermocycler followed by analysis using the SDS2.1 software program (Applied Biosystems). For each experimental run, standard curves for these genes were generated using known quantities (10–104 in log increments) of gene specific plasmids for calculation of absolute copy number.
One-way analysis of variance was performed on qRT-PCR results. To determine significance, a Kruskal-Wallis multiple comparison Z-value test (Dunn's test) was performed, where significance was defined as P≤0.05.
bb0243 was amplified using primers NdeI243F/XhoI243R (Table 4), cloned into the TOPO 2.1 vector, transformed into E. coli Top10 cells and recombinant clones were selected per manufacturer's instructions (Invitrogen). The bb0243 insert was excised from the TOPO 2.1 plasmid by double digestion with NdeI and XhoI (Fermentas) and the insert was purified after separation by gel electrophoresis using the Wizard SV genomic gel purification kit according to manufacturer's instructions (Promega). pET-15b (Novagen, Gibbstown, NJ) was digested with NdeI and XhoI and the gel-purified bb0243 insert was ligated with the NdeI/XhoI-cut pET-15b at a 2∶1 ratio with 10 units of T4 ligase (New England Biolabs, Ipswich, MA). The recombinant plasmid was transformed into E. coli DH5α and clones were selected on LB agar plates containing 100 µg/ml ampicillin. Recombinant pET-15b carrying bb0243 was transformed into E. coli BL21-DE3 and grown on LB agar plates containing 100 µg/ml ampicillin. A clone containing bb0243 was selected and subjected to DNA sequencing. This sequence contained two single nucleotide changes relative to the reported sequence in strain B31-MI [12]. Nucleotide 107 had a T to C change that would result in a predicted amino acid change of I359T and nucleotide 591 had an A to T substitution that would result in an amino acid change of E197D.
The selected clone was grown at 37°C in 250 ml Luria broth containing 100 µg/ml ampicillin and 1 mM IPTG with agitation for 4 hours. Cells were recovered by centrifugation at 8000 RPM for 10 minutes and rGlpD was isolated from the cells using Ni-NTA His Bind Resin (Novagen) according to the manufacturer's instructions. Fractions containing rGlpD, as determined by SDS-PAGE, were pooled and loaded into an Amicon Ultra 50 kDa molecular weight cut off spin column (Millipore, Billerica, MA). The protein sample was centrifuged at 7500× g for approximately 8 minutes to a volume of 800 µl. The protein solution was dialyzed against 2 liters of 1× PBS, 6 M urea (pH 7.4) with stirring overnight at 4°C. Identity of the protein as B. burgdorferi rGlpD was confirmed by LC-MS/MS analysis (Keck Biotechnology Resource Laboratory,New Haven, CT). The yield of purified rGlpD was 1.3 mg.
100 µg of purified rGlpD in 1×PBS, 6 M urea (pH 7.4) was inoculated along with Freund's adjuvant into two Sprague-Dawley rats by Harlan Laboratories (Madison, WI). The rats received a boost at day 28 and day 56 post-inoculation and were sacrificed and bled on day 70 post inoculation. GlpD antiserum was tested by ELISA and confirmed to be specific for GlpD by immunoblot analysis.
B. burgdorferi cells grown in vitro or in DMCs were lysed with Bugbuster HT (Novagen) and 1 µg/ml of lysozyme (Sigma) according to manufacturer's instructions. 2 µg of whole cell lysate was subjected to 12.5% SDS-PAGE and separated proteins were visualized by silver staining as described [71]. For immunoblotting, separated proteins were transferred to PVDF membrane. Membranes were exposed to protein-specific primary rat antiserum (GlpD, 1∶400 dilution; FlaB, 1∶2500 dilution) followed by alkaline phosphatase-linked anti-rat secondary antibody (1∶500) (KPL, Gaithersburg, MD). The membrane was washed three times for 10 minutes with 1× TBS/0.05% Tween 20 and developed with BCIP/NBT phosphatase substrate (KPL) until band development (approximately 2–4 minutes).
To determine seroconversion in mouse infection studies, mouse serum was added to 1× TBS with 0.5% dry milk at 1∶200 dilution and incubated with whole B. burgdorferi lysate Marblot strips (MarDX, Jamestown, NY) for 1 hour at room temperature. The remaining procedure is as described above, with anti-mouse secondary antibody (1∶5000 dilution).
|
10.1371/journal.pntd.0006143 | Understanding the legal trade of cattle and camels and the derived risk of Rift Valley Fever introduction into and transmission within Egypt | Rift Valley Fever (RVF) is a mosquito-borne zoonosis, which may cause significant losses for the livestock sector and have serious public health implications. Egypt has been repeatedly affected by RVF epidemics, mainly associated to the importation of animals from sub-Saharan countries, where the disease is endemic. The objective of our study was the improvement of the surveillance and control strategies implemented in Egypt. In order to do that, first we evaluated the legal trade of live animals into and within Egypt. Then, we assessed the risk of Rift Valley Fever virus (RVFV) transmission within the country using a multi-criteria evaluation approach. Finally, we combined the animal trade and the risk of RVFV transmission data to identify those areas and periods in which the introduction of RVFV is more likely. Our results indicate that the main risk of RVFV introduction is posed by the continuous flow of large number of camels coming from Sudan. The risk of RVFV transmission by vectors is restricted to the areas surrounding the Nile river, and does not vary significantly throughout the year. Imported camels are taken to quarantines, where the risk of RVFV transmission by vectors is generally low. Then, they are taken to animal markets or slaughterhouses, many located in populated areas, where the risk of RVFV transmission to animals or humans is much higher. The measures currently implemented (quarantines, vaccination or testing) seem to have a limited effect in reducing the risk of RVFV introduction, and therefore other (risk-based) surveillance strategies are proposed.
| Rift Valley Fever (RVF) is a mosquito-transmitted disease, which may affect animals (ruminants and camels) and cause severe disease. Egypt has been affected by RVF in the past, and it is considered at risk because of the importation of animals from endemic countries. We developed a map of the risk of RVFV transmission by vectors using current knowledge on the disease, and then combined it with animal trade data to identify the areas at the highest risk for RVFV introduction. Our results indicate that the main risk is posed by camels imported from Sudan, when they are transported to animal markets or slaughterhouses, which are generally located in areas where the risk of RVFV transmission is much higher. In fact many of them are located in the proximities or within the main cities, resulting in an increased risk of RVFV transmission to humans.
| Rift Valley Fever virus (RVFV) is an arbovirus which belongs to the Phlebovirus genus, within the family Phenuiviridae (order Bunyavirales). It is the causative agent of Rift Valley Fever (RVF) that may affect both humans and animals, mainly ruminants, but also camels. In livestock RVF causes large number of abortions (abortion storms) and high mortalities in young animals. In humans, RVFV infection generally causes a self-limiting, acute and febrile illness, although in some cases disease progresses to more severe forms, with neurological disorders, blindness, hemorrhagic fever or thrombosis [1]. RVFV infection of animals may occur by the bite of an infected mosquito (mainly of the Culex or Aedes genera) or through direct contact with infected animal tissues or fluids [2]. Infection via mosquito bites is considered the primary mode of transmission in the first stages of the epidemic, while contact with contaminated material is more relevant during the amplification stage. RVFV within infected tissues is quite resistant to inactivation and may remain infectious for a few days [2]. In fact, the majority of human cases are likely to occur in farmers, veterinarians or slaughterhouse employees due to contact with infected material [3].
RVF may have a dramatic consequence on producers and livestock industries, including the impact on international trade [4]. The public health impact of RVF may also be severe, as demonstrated by the 200,000 people infected and 600 fatal cases reported in the 1977 epidemic in Egypt [5]. Besides that epidemic, Egypt has been affected by RVF in 1993, 1994, 1997, and finally in 2003 [6]. In fact, Mroz and collaborators [7] have recently reported the low level circulation of RVFV in some areas of Egypt.
The continuous importation of viraemic ruminants or camels, mainly from Sudan, was considered the main source of introduction of RVF into Egypt [6]. Furthermore, those importations may also result in the introduction of other exotic diseases such as the Middle East Respiratory Syndrome coronavirus (MERS-CoV) [8]. RVF is considered to be endemic in sub-Saharan African countries (including Sudan), with major outbreaks associated with periods of heavy rainfall and flooding [9]. Socio-economic changes, with human population growth and the associated increased demand for meat, or changes in the meat market prices could result in greater controlled and uncontrolled transboundary movements of livestock. A consequence of all those changes would be an increase of the risk of introduction of RVFV into areas of the Mediterranean basin and the Middle East [3]. While surveillance and diagnostic methods for RVFV are available, RVFV control is hindered by the difficulties of vector control, and the fact that vaccines are only available for ruminants, with either inactivated vaccines (with a limited efficacy) or live attenuated vaccines (with a residual pathogenic effect) [3,10]. Therefore, the best strategy to protect countries at risk of RVFV introduction is the implementation of regional monitoring and control programmes in endemic countries, as well as the establishment of early warning systems in the countries at risk. The Egyptian authorities implement both measures to prevent the introduction of RVFV (vaccination of imported animals, quarantines and testing of a proportion of imported animals) and measures to prevent the spread (vaccination of the Egyptian susceptible population). If those measures were effective, there would be no risk of RVFV introduction and spread, however, their efficacy has not been evaluated, and some deficiencies have been identified in the past [6]. Given the limitation of resources, risk-based methods (i.e. focusing on those areas and/or practices that pose the highest risk) may be considered as a cost-effective alternative for the surveillance of RVF in Egypt. Many of those risk-based methods are data-driven, which rely on comprehensive information about disease-associated events. In a data-scarce context, knowledge-driven methods, which rely on the previous information on the factors associated with disease occurrence, offer an alternative approach to identify areas at risk [11].
The main objectives of this study were:
That will allow the improvement of the surveillance and control measures currently being implemented by the Egyptian Authorities, and make a better use of the resources available. Improvement of RVFV surveillance should allow the early detection of the disease, while the improvement of control measures should help prevent cases of RVFV infection in both animals and humans.
The spatial risk of RVFV transmission within Egypt was estimated using a GIS-based multi-criteria evaluation (MCE) approach [12], which allows to combine the (spatial) data on the different factors that may influence the risk of a given disease. In the case of RVF it is essential to account for the distribution of both the susceptible hosts and the vectors of RVFV [13]. All layers were transformed into a raster format, with 1 km × 1 km spatial resolution and the common UTM 35N projection. R software [14] was used for both the analyses and generation of maps.
Besides the favorability for transmission in a given area, the actual risk of RVF occurring in that area is determined by the probability of the virus being introduced into that location. Given that in Egypt imported animals are quarantined on specific premises on arrival to the country, and that transmission within those quarantine stations and to animals in their surroundings may occur, those will be the risk locations of primary interest. No measures to protect animals from mosquitos are taken on those quarantines.
After quarantine, imported camels may be taken to markets or to slaughterhouses, where transmission within the premises and to animals in their surroundings may also occur, and therefore those will be the risk locations of secondary interest.
Studies on the dispersal of Cx. pipiens in urban habitats found maximum flight ranges of 1.98 km [29] and 2.48 km [30]. However, given the low probability of recapturing a marked mosquito at large distances from the point of release, we used a conservative radius of up to 5km for the buffer area. We also assessed the effect of changing the size of the radius on the risk of RVF transmission.
Animals imported into Egypt are not randomly distributed throughout the country, instead they are transported to specific locations where the risk of RVFV transmission by vectors will be quite variable. First, they are taken directly to specific quarantine premises depending on the species and origins (Fig 10A). After quarantine, imported camels also follow an established movement pattern: they may be either be taken to slaughterhouses (Fig 10B) or to animal markets (Fig 9B). In contrast, all cattle are slaughtered within the quarantine facilities.
The summary estimates of the risks of RVFV transmission by vectors at those main locations after the importation of animals, i.e. markets, quarantines and camel slaughterhouses, considering buffers of either 5000 or 2000 meters, are shown in Table 3.
Of the quarantines (Fig 3 and Fig 10A), Abu Simbel, where a significant proportion of the camels and all the cattle imported from Sudan arrived, seemed to have a low risk of RVFV transmission by vectors (between 0.04 and 0.08 depending on the buffer considered). For Shelateen quarantine, where camels imported from eastern Sudan arrived, the risk was 0. For Sahl Hasheesh, where cattle from Ethiopia arrived, the risk was extremely low (between 0 and 0.01 depending on the buffer). For Adabiya, where camels from Ethiopia arrived, the risk was also relatively low (between 0.07 and 0.14). In contrast, for Al Qata, where also camels from Ethiopia arrived, the risk of RVFV transmission by vectors was quite high (between 0.70 and 0.83).
Of the two markets with the highest risk-ranking score (Fig 5), Aswan had a risk of RVFV transmission by vectors relatively low (between 0.15 and 0.16), while Birqash market had risk of RVFV transmission by vectors quite high (between 0.70 and 0.74) (Fig 9B). Of the remaining high-risk markets there are 10 with risk of RVFV transmission by vectors estimates above 0.7. Of them, 7 are located in risk-area 1 in Fig 8, in the governorates of Gharbia (3), Qalyubia (3) and Monofia (1). The other 3 are located in Minya governorate (risk-area 2). In risk areas 3 and 4, there is only one high-risk market in each with risk of RVFV transmission by vectors above 0.5, in Qena and Aswan governorates, respectively.
The camel slaughterhouses (Fig 6A and Fig 10A) are mostly located in the proximities of Cairo, where conditions are favourable for RVFV transmission and therefore their risk are relatively high (Table 3). Of the three slaughterhouses that killed the most animals, Elbasateen had a risk 0.49 regardless of the buffer distance, while Elwarak had a risk between 0.47 and 0.50, and Kerdassa between 0.57 and 0.69.
Camel meat is an important source of cheap protein for the Egyptian people. However, as the resident camel population in Egypt is really small, 66,233 animals in 2015 (according to OIE data) and located mainly in remote areas, large number of camels need to be regularly imported into Egypt, the majority of which come from Sudan (86% of the total in 2015) and the rest from Ethiopia. The flow of camels is continouos throughout the whole year and the number of animals imported per year seem to have an increasing trend. In 2015, an average of 18,000 camels were imported per month.
During the first RVF epidemic in Egypt in 1977, mortality and abortions were reported in camels in Aswan governorate (in the south of the country), and RVFV was isolated from this species [32,33]. Therefore, even though the source of the outbreak could not be determined, the possibility of introduction of the disease from Sudan by camels was considered the most likely explanation. In the following epidemics (1993, 1994, 1997 and 2003), importation of camels, mainly from Sudan, was also considered as a possible source of introduction, although no proper epidemiological investigations were carried out, and alternative mechanisms were considered [6,26]. As a result, the role of camels in the epidemiology of RVFV remains unclear. In fact, it has been suggested that camels may be essential for RVFV transmission in some ecosystems where their density is high, which would allow viral amplification and transmission to alternative hosts, but not in others [34]. Furthermore, data on factors such as duration of viraemia, infectiousness to mosquitos, or the risk of infection of humans from contaminated products remain unknown, and further studies on camels would be required. Despite those uncertainties, camels are considered susceptible to RVFV, and the virus has been isolated from blood from healthy, naturally infected camels in Egypt and Sudan [32,35]. While RVFV infections in camels are usually mild or asymptomatic, abortion waves were observed in RVF outbreaks in Kenya and Egypt [34]. In the 2010 epidemic in Mauritania, clinical signs other than abortions were observed in camels, either a peracute form with sudden death within 24 hours; or an acute form with fever and systemic lesions, that resulted in death when haemorrhagic signs occurred [36]. RVFV infections in camels have been recorded in most sub-Saharan African countries, with prevalences ranging from 3.0 to 51.9 percent [34]. In fact, RVF is thought to be endemic in sub-Saharan African countries (including Sudan), with large epidemics occurring after heavy rainfall and flooding [9]. Phylogenetic analyses carried out in Sudan with RVFV strains from the 2007 and 2010 epidemics, indicated that they were the result of multiple introductions from eastern Africa [37]. Since then, there were evidences of circulation of RVFV during the inter-epidemic period: in 2014–2015, 9.6% of camels in Khartoum State (Sudan) had RVFV antibodies, including at least one positive born after the last reported epidemic [38].
Besides, camels may be involved in the introduction of other diseases such as MERS-CoV, which may cause severe disease in humans. MERS-CoV was detected in Egypt in camels imported from Sudan and Ethiopia [8], and seems to still be present in Ethiopian camels [39].
Camels provide food (milk and meat), fibre (wool and leather) and draft power (for transportation and cultivation) to human communities located in dry environments [34]. As a result, in some societies they are essential from a cultural and socio-economic point of view. The world camel population has steadily increased in recent years to reach almost 28 million heads in 2014 (according to FAO). Besides, camel farming practices are changing, with a significant increase of intensive production in peri-urban areas, which results in a closer contact with the human population [34]. As camels are used for trade allowing the connection distant populations, in particular in arid and semi-arid ecosystems, they may play a role in the large-scale dissemination of pathogens.
The number of cattle legally imported into Egypt, 4074 on average per month in 2015, is small as compared to camels. The flow of cattle became more regular by the second half of the 4-years period, coinciding with the intensification of the trade with Sudan. In contrast to camels, the data revealed a more seasonal pattern, with an increase of importations in the months of July, August and September, just before the major Islamic feast of Eid al-Adha (“Sacrifice Feast") in those years. Importation of cattle also poses a threat for Egypt, in fact, the epidemics of foot and mouth disease (serotype A) and of lumpy skin disease that affected Egypt in 2006 were both linked to the importation of animals from Ethiopia [40]. Increase of importations of cattle and camels into Egypt may be explained by the growing demand of meat. While between 2006 and 2015 the human population in Egypt grew a 19.9% [41], the population of ruminants and camels in the country only grew a 3.2% [42]. According to the World Bank projections, the human population in Egypt is expected to continue growing at a rate of almost 2% per year in the following decade, and therefore legal (and potentially illegal) trade of animals into Egypt is expected to increase.
After quarantine, camels may be transported to markets or slaughterhouses throughout Egypt. Those movements within Egypt would contribute the spread of RVFV in the event of introduction of the virus, as evidenced in other countries [43,44]. The EVS characterized the network of animal markets by collecting information on the location, but also on several characteristics of 273 animal markets distributed throughout Egypt. That combined with the slaughterhouse data allows having a basic picture of the animal movement patterns within the country, which offers an alternative approach in countries without the resources to have a comprehensive recording of animal movements. That information is very useful for planning risk-based surveillance strategies (i.e. target those markets where detection of a particular disease is more likely), with the advantage the criteria may be adapted to each specific disease. The collection of more comprehensive information on for example the number of animals by species or the origins and destinations of the animals that went across those markets, would allow a much clearer picture of the animal movement network, and is therefore recommended The animal market network in Egypt is mainly composed of small markets, which basically operate once a week. Interestingly, almost 40% of the markets traded animals from different governorates, which means that infectious diseases are likely to spread efficiently once introduced into the country. In relation to the species, 2% of markets traded exclusively camels, and a 17% further traded camels plus other species susceptible to RVFV. Most of the markets are located in the areas beside the Nile river and its delta, but there are also animal markets alongside the Mediterranean coast, or even in the Sahara desert (Farafra market). Of the 273 markets, only 17 were in the governorates bordering Sudan, where smuggling of small ruminants was considered more likely. The risk-ranking score for markets varied between 5 and 15, with a mean value of 7.9. The markets with the highest risk of RVFV transmission were in the proximities or within the main cities (e.g. Cairo, Minya or Aswan), while most of the camel slaughterhouses were near Cairo. That proximity to urban areas results in an increased risk of RVFV transmission to humans (zoonotic risk). In contrast to camels, all cattle imported into Egypt are slaughtered within the quarantines to which they arrive, which are located in remote areas, far from populated areas.
In order to prevent the introduction of RVFV by camels, the EVS implement a series of measures that include vaccination of imported animals, quarantines and testing of a proportion of them. However, the vaccination strategy is unlikely be effective, as camels imported from Sudan receive a single dose and camels imported from Ethiopia receive two doses, but with only 7 days difference. The inactivated vaccines against RVFV require a booster 3–6 months after initial vaccination, followed by yearly boosters [45]. Besides, the efficacy of the inactivated vaccine has not been tested in camels [45], and further research on that field would be needed. While it has been argued that the long distance that the camels from Sudan need to walk to reach the Egyptian border would be enough to prevent the importation of infected animals, there are still many uncertainties in relation to RVFV infection in camels, and in any case, that measure does not comply with OIE rules for importation of live animals [6]. On the other hand, means of transportation seem to be changing as at least some of the camels are currently transported on trucks rather than on foot (MALRE personal communication) and transmission (either direct or indirect) during the transport of camels cannot be ruled out. Seroconversions in Madagascar highlands evidenced the local circulation of RVFV in periods in which mosquitoes were rare or inactive, suggesting that other forms of transmission seem likely [46]. Even though camels are quarantined, for 3 days if coming from Sudan and 10–16 days if coming from Ethiopia, the efficiency of clinical detection of RVFV infection is limited by the fact that infections in camels are often unapparent. Besides, about 10% of the camels imported from Sudan and Ethiopia in the last 5 years were tested against RVF by RT-PCR, and were negative (MALRE personal communication). However, the sensitivity of such non-statistically based testing protocol for detecting the infection, if present, is questionable. While the measures currently implemented in camels are likely to reduce some of the risk of RVFV introduction into Egypt, their effect is likely to be limited. Similar conclusions may be drawn from the strategies implemented in cattle with the particularities that all cattle are slaughtered within the quarantine premises, where the risk of RVFV transmission by vectors was estimated to be very low. Therefore, vaccination of imported cattle is unlikely to have a significant effect on the reduction of the risk of RVFV introduction into Egypt.
On the other hand, to prevent the transmission of RVFV within Egypt, all the animals of the susceptible species (cattle, sheep, goats, buffaloes and camels) are meant to be vaccinated twice a year. However, the analysis of the vaccination data evidences a great heterogeneity in the level of vaccine coverage among species and governorates. Despite the increase in the number of vaccine doses administered in 2015 as compared to 2014, the level of vaccine coverage (30% in sheep and goats, 35% in buffaloes and 60% in cattle) would not be enough to prevent the transmission in the event of RVFV introduction into the country, in particular in those governorates where the level of vaccination is below average. Differences in the density of susceptible species throughout the country combined with the heterogenous application of vaccination programs among areas and species results in the identification of some areas of high density of susceptible hosts of RVFV, which is where efforts should be focused. Failures in the local application of vaccination programs were identified as one of the causes of previous epidemics of RVFV in Egypt [6]. There were also areas where high levels of vaccine coverage were reached resulting in a significant reduction of the risk of RVFV transmission. Conclusions in relation to vaccine coverages should be taken with caution, as the censuses used for their calculation may not be totally accurate/updated.
Even though we used the distribution of Cx. pipiens to map vector distribution, we believe it is an accurate representation of the RVFV transmission by vectors. Entomological surveys indicate that Cx. pipiens is the most common mosquito species in Egypt [19,47] and was considered the primary vector of RVFV in previous epidemics [5,16]. Besides, the primary breeding areas of other RVFV vectors (e.g. Cx. antennatus) coincide with those of Cx. pipiens [16].
The map of RVFV transmission by vectors in Egypt evidences that the majority of the country surface has a risk of zero, as it corresponds to bare soil land cover (i.e. desert), which is not suitable for the presence of neither hosts nor vectors. Therefore, RVFV transmission by vectors is basically restricted to the wet areas surrounding the Nile River and its delta. Within that area there is a great variability on the risk depending on the density of vectors and hosts. Transmission via mosquitos is considered the most important mode of transmission during the enzootic cycle [2].
The evaluation of the seasonality of Culex pipiens indicated that in general there was not much variation in Cx. pipiens favorability between the different months. Therefore the risk of RVFV transmission by vectors in the suitable areas of Egypt seem to be more or less constant throughout the year. However, results also indicate that in some specific locations there may be significant variations in Cx. pipiens density. Zayed and collaborators [17] demonstrated that in the areas where the seasonal flooding of the Nile Delta occurred, the population of vectors changed significantly, mainly in late summer-early fall.
The validation of the risk map evidenced that the risk of RVFV transmission by vectors obtained at the location of human and animal cases and RVFV-positive vectors from the previous RVFV epidemics in Egypt, varied between 0.18 and 0.85, with a mean value of 0.49. That would indicate that RVFV transmission by vectors is not restricted only to the areas with the highest risks, but that transmission may occur also in areas where the risks estimated were moderate or even low. In other words, in Egypt RVFV transmission would be possible in areas where the density of vectors or susceptible hosts is far from their maximum values. However, the limitations of the model should also be taken into account. Model results may be influenced by the accuracy of some model inputs (e.g. vaccine coverage), or the uncertainty about some model parameters (e.g. susceptibility of the different species). Besides, knowledge-driven models are always subject to some sort of subjectivity, for example in relation to the factors (layers) chosen. On the other hand, the location of outbreaks used for model validation was not exact, and given that some of them were as old as 1979, the landscape, and therefore the suitability for Cx. pipiens may have changed since then.
As expected, the linear regression analysis indicated that the distribution of Cx. pipiens was the most determinant factor in the risk of RVFV transmission by vectors. The only other statistically significant parameter was the density of susceptible goats. Relevance of goats in the risk of RVFV transmission is probably linked to their higher susceptibility, and therefore higher weight, as compared to other species (cattle, buffaloes and camels), as well as with the fact that the areas with the highest density of goats coincided with very low levels of vaccine coverage (7% and 14%, respectively).
The risk of RVFV transmission by vectors in the surroundings of the five quarantines available in Egypt was estimated to be quite low except in the case of Al Qata (where some of the camels imported from Ethiopia arrive). There is no data on the duration of viraemia in camels, but as camels imported from Sudan stay in the quarantine for only 3 days, it seems possible that they may remain infectious after being released from the quarantine, even if RVFV transmission by vectors within the quarantine was not possible. On the other hand, RVFV transmission within those quarantines may also occur by direct contact with tissues, body fluids or fomites of infected animals.
While all cattle are slaughtered within quarantine facilities, imported camels (potentially infected) may be taken to animal markets or to slaughterhouses after the quarantine period, and those would be the places where RVFV transmission would be more likely. The risk of RVFV transmission by vectors at the 273 animal markets was rather variable (between 0 and 0.84), but the mean risk was quite high, 0.60. In the case of the 18 camel slaughterhouses, the risk of RVFV transmission by vectors was quite high (mean risk of 0.59–0.60), and there was in fact less variability than in the case of animal markets. Establishment of an effective surveillance system in the areas surrounding key markets and camel slaughterhouses may allow the early detection of RVF and the timely establishment of control measures. That may include improving passive surveillance by raising farmers’ awareness for the reporting of any symptom compatible with RVF and providing the mechanisms that allowed a prompt laboratory confirmation of any suspected case. Also, the establishment of some active (sentinel) surveillance in the locations identified as having the highest risk would be essential. Besides the risk of RVFV transmission by vectors, the slaughtering of animals, which in the case of cattle is carried out within quarantine premises, poses an important risk for humans by exposure to body fluids or tissues of infected animals. Infectivity of blood during the acute phase of infection in mammals is high [2]. Moreover, given that RVFV is quite resistant to inactivation, infected tissues may remain infectious for a period of a few days [2]. In fact, Nicholas and collaborators [48] concluded that practices such as skinning or slaughtering animals were significantly associated with the risk of RVFV infection. A cross-sectional survey on 1181 abattoir workers in 15 governorates of Egypt after the 1993 RVFV epidemic evidenced a 2% prevalence of anti-RVF IgM antibodies [49]. Antibodies were found in 9 of 31 slaughterhouses and 8 of 15 governorates, including several areas where no clinical disease had been reported. Therefore, Egyptian abattoirs in which imported cattle or camels are slaughtered should implement measures (e.g. use of personal protective equipment) to prevent the infection of the personnel.
Even though we focused on the risk posed by the legal trade of animals, other mechanisms for RVFV introduction into Egypt are also possible. Illegal trade of animals through the borders is likely to occur, but its relevance in relation to the risk of RVFV introduction is difficult to assess, and would require further studies. On the other hand, the possibility of RVFV spread (potentially to distant areas) through the movement of infected (viraemic) humans and subsequent infection of competent vectors remains controversial. While for some authors humans infected by RVFV may develop a viremia sufficient to infect naïve mosquitoes [50], others consider humans as dead-end hosts [3]. Also, wind-borne transportation of infected mosquitos has been proposed [5], but not demonstrated scientifically.
|
10.1371/journal.pcbi.1004443 | Partial Unwrapping and Histone Tail Dynamics in Nucleosome Revealed by Coarse-Grained Molecular Simulations | Nucleosomes, basic units of chromatin, are known to show spontaneous DNA unwrapping dynamics that are crucial for transcriptional activation, but its structural details are yet to be elucidated. Here, employing a coarse-grained molecular model that captures residue-level structural details up to histone tails, we simulated equilibrium fluctuations and forced unwrapping of single nucleosomes at various conditions. The equilibrium simulations showed spontaneous unwrapping from outer DNA and subsequent rewrapping dynamics, which are in good agreement with experiments. We found several distinct partially unwrapped states of nucleosomes, as well as reversible transitions among these states. At a low salt concentration, histone tails tend to sit in the concave cleft between the histone octamer and DNA, tightening the nucleosome. At a higher salt concentration, the tails tend to bound to the outer side of DNA or be expanded outwards, which led to higher degree of unwrapping. Of the four types of histone tails, H3 and H2B tail dynamics are markedly correlated with partial unwrapping of DNA, and, moreover, their contributions were distinct. Acetylation in histone tails was simply mimicked by changing their charges, which enhanced the unwrapping, especially markedly for H3 and H2B tails.
| Nucleosomes, folding units of chromatin, wrap DNA about 1.75 turns and provide bottlenecks for transcription. Recent experiments showed that nucleosomes are not rigid but dynamic, showing spontaneous and partial unwrapping which is thus important for transcriptional activation. Experimentally, however, one cannot directly watch DNA unwrapping at high resolution. On the other hand, molecular dynamics simulations have high spatio-temporal resolution and thus can be powerful and complementary to experiments. Here, we put forward coarse-grained modeling of protein-DNA interactions at residue-level resolution, which is rather generic and thus can be applied to any protein-DNA complexes. By this method, we could reveal spontaneous and salt-concentration dependent partial unwrapping of DNA from nucleosomes. In addition to consistency with single molecule experiments, the simulation showed multiple and distinct intermediate states of unwrapping. Interestingly, partial unwrapping of DNA is correlated with certain parts of histone tail dynamics. Deleting positive charges in histone tails that mimics histone acetylation facilitated partial unwrapping, most significantly for H3 and H2B.
| Nucleosomes, basic units of chromatin, are made of about 147 base pair (bp) double strand (ds) DNA wrapped 1.75 turns around a histone octamer [1]. Many available X-ray crystal structures provide atomic structural information on nucleosomes [2–7]. Yet, nucleosomes are not static, but dynamic complexes changing their structures, positions along genome, and component during cell cycles. Upon DNA replication, for example, nucleosomes must globally disassemble and, after replication, re-assemble, which also involves nucleosome repositioning [8,9]. In addition to these global changes, nucleosomes regularly show partial unwrapping dynamics, which could control higher-order chromatin folding and transcriptional activity [10,11]. These dynamic aspects of nucleosomes are much less clear.
Recently, partial and global unwrapping of single nucleosomes have been intensively investigated by single-molecule FRET experiments, mechanical pulling experiments with optical traps, and so forth [12–24]. Notably, single-molecule FRET experiments were used to characterize spontaneous and intermittent partial unwrapping dynamics of nucleosome. DNA unwrapping occurs from outer stretches and the rate constants for unwrapping depends on the distance inside the nucleosome [17]. The dissociation constant for unwrapping also depends on salt concentration; as expected, higher salt concentration enhances unwrapping. Due to the spontaneous site-exposure, DNA-binding proteins can access to nucleosomal DNA. Mechanical unwrapping experiments clarified site-dependent interaction between DNA and histone cores [20]. In addition to the central dyad region where the strongest interaction has been identified, the off-dyad region which is 1/2 turn from the dyad offers another strong interaction sites [20,25].
These experiments give, albeit un-ambiguous, structurally limited information for, at most, a few FRET pair distances. For example, how unwrapping dynamics is correlated with flexible histone tail dynamics is not directly observed. As a complementary approach, higher resolution structural dynamics analysis is desired. In this sense, molecular dynamics (MD) simulations are potentially powerful because they provide full of time-dependent structural information. Yet, conventional atomistic MD simulations reach, usually, microsecond time scales as of today, which is shorter than typical time scale of intermittent DNA unwrapping from nucleosomes. To speed up MD simulations drastically, one way is to coarse grain the molecule model. These coarse-grained (CG) MD simulations are getting more and more popular for large-scale biomolecular simulations [26,27].
For single nucleosome and nucleosome-arrays, various levels of CG models have been developed and used [28–37]. For the latter, mesoscopic modeling is particularly successful where a nucleosome or a histone octamer is treated as rigid object and linker DNA is represented by continuous string or chain of beads [28,30,33]. At single-nucleosome level, unwrapping of ends of nucleosomal DNA was investigated with a higher resolution CG model [35]. Their CG model was primarily developed to approximate near-native conformations so that larger-scale unwrapping of DNA was not treated. Related to this point, this CG model does not directly treat electrostatic interactions and thus salt-dependence of unwrapping was not investigated.
To address partial unwrapping of single nucleosome from small to large scales and in comparison with single molecule experiments, we need CG models that directly capture mechanical property of histones and DNA as well as electrostatic interactions with a higher resolution than the mesoscopic level. DNA models are desired to represent major and minor grooves, while protein models need to represent residues that are inserted into these grooves. Histone tails play crucial roles and thus need to be modeled as flexible and charged polymers. Thus, to address long-time partial unwrapping dynamics of single nucleosomes, in this paper we put forward CG MD simulations in which each amino acid in proteins is represented by one CG particle and one nucleotide in DNA is represented by three CG particles. Specifically, proteins, i.e., histone octamer, are modeled by a structure-based Go-model [38,39] and dsDNA is modeled by 3SPN.1 of de Pablo group [40,41]. The structure-based Go model is known to well approximate near-native fluctuations as well as global folding [42]. Interactions between histone octamer and dsDNA are approximated by electrostatic interactions and a structure-based contact potential. The explicit treatment of electrostatics enables us to address salt-concentration dependent unwrapping dynamics.
In this paper, brief description of computational methods is followed by simulation results of spontaneous fluctuation dynamics of single nucleosome. We observed multiple and salt-dependent intermediate states of unwrapping. Then, we investigate histone tail conformations in these dynamics. Moreover, we mimic histone tail acetylation by deleting charges in tails and investigate their effects on partial unwrapping. Finally, we investigate mechanical unwrapping of single nucleosome.
We employed coarse-grained (CG) models for proteins and DNA. For the protein, we used a simple CG Go model [38,39] in which one amino acid is represented by one CG particle located at Cα position. For DNA we took a CG DNA model 3SPN.1 developed in de Pablo's group [40,41], where each nucleotide is represented by three CG particles, base, sugar, and phosphate.
The total potential energy function is divided into that for proteins, Vpro, that for DNAs, Vdna, and that for the interactions between proteins and DNAs, Vpro-dna,
Vtotal=Vpro+Vdna+Vpro−dna
The energy function for proteins is that of Clementi et al [38],
Vpro=∑ikbdpro(ri,i+1−ri,i+10)2+∑ikbapro(θi−θi0)2+∑i{kdih1pro[1−cos(φi−φi0)]+kdih3pro[1−cos3(φi−φi0)]}+∑i<j−3nat contactεgopro[5(rij0rij)12−6(rij0rij)10]+∑i<j−3non−nativeεevpro(σprorij)12
where the first, second, and third terms represent restraint potentials for virtual bond lengths, virtual bond angles, and virtual dihedral angles, respectively. The fourth term is the non-local contact potential that stabilizes amino acids pairs that are in proximity at the native (reference) structure. The last term represents a generic excluded volume effect. ri,i+1 stands for the distance of a virtual bond between i-th and i+1-th amino acids, θi is the i-th virtual angle made by two consecutive virtual bonds, ϕi is the i-th dihedral angle defined by three consecutive virtual bonds. rij is the distance between i-th and j-th amino acids. Those with superscripts 0 are parameters that take the values of the corresponding variables at the native (reference) structure. The coefficients k's and ε's are parameters that modulate relative balance among the terms. We used a default set of these parameter values in CafeMol [42].
The energy function for DNA is 3SPN.1 developed in de Pablo's group and can be written as
Vdna=∑i[ kbd1dna(ri,i+1−ri,i+10)2+kbd2dna(ri,i+1−ri,i+10)4 ]+∑ikbadna2(θi−θi0)2+∑ikφdna[ 1−cos(φi−φi0) ]+∑i<jstack4εstdna[ (rij0rij)12−(rij0rij)6 ]+∑i<jbaseεbpidna[ 5(σbpirij)12−6(σbpirij)10 ]+∑i<jnon−native{ 4εevdna[ (σdnarij)12−(σdnarij)6+14 ],ifrij<216εdna0,otherwise +∑i<jelecqiqje−rijλD4πε0εrij+∑i<jsolvεs{ [ 1−e−α(rij−rs) ]2−1 }
Here, the first, second, and third terms are restraint potentials for virtual bond lengths, virtual bond angles, and virtual dihedral angles. The fourth and fifth terms are non-local contact potentials, in which the fourth one is for the stacking energy and the fifth one represents base-pairing. The sixth term is for excluded volume effect. The seventh term is a regular Coulomb energy with the Debye-Huckel screening. In the Debye-Huckel formula, λD depends on ionic strength, and thus salt concentration of the solution. The last term represents empirical solvation energy to facilitate base-pairing. Parameters with superscript 0 take values of the corresponding variables at B-type dsDNA. The rest of parameters were tuned to approximate general properties of dsDNA. We used the default parameter values of 3SPN.1, except for the dielectric constant of water ε that was simplified as the constant value 78. See the original article for other details [41]. We note that the Debye-Huckel model is a computationally efficient, but crude approximation for such a highly charged molecule as DNA and the explicit treatment of counter ions provides higher resolution and probably more accurate estimates of electrostatics in DNA [36,43].
Modeling interactions between the histone octamer and dsDNA is not straightforward. If the interaction was very specific and many of side-chains of histones perfectly fit with DNA, using the structure-based potential, i.e., Go-potential would be reasonable. If, on the other hand, the interaction was purely non-specific, the generic electrostatic interaction alone would be reasonable, as in our earlier work of p53 [44]. We note that the single-molecule experiments often use the so-called Widom 601 DNA sequence, a selected high-affinity nucleosome positioning sequence, of which specificity is clearly non-negligible, but incomplete as well. The current DNA model does not account for detailed sequence dependent property of DNA. To this end, we decided to include a weakened Go-potential, in which the scaling parameter is tuned so that the resulting dynamics matches some of experiments.
The interactions between proteins and DNA include the electrostatic interaction in the same way as that in DNA, the general excluded volume interactions in the same form as in the protein model, and structure-based pairwise contact potential that stabilize protein-DNA complex in a reference structure, which in the current case, a crystal structure of nucleosome,
Vpro−dna=∑i<jnat contactεgopro−dna[5(rij0rij)12−6(rij0rij)10]+∑i<j−3non−nativeεevpro−dna(σpro−dnarij)12+∑i<jelecqiqje−rijλD4πε0εrij
The first term is the structure-based contact term, in which the parameter εgopro−dna controls specific attraction between histone proteins and DNA, while the last term provides sequence-non-specific attraction between positively charged histone amino acids and DNA. In the structure-based contact term, we used sugar and base sites in DNA, but not including phosphate sites because the phosphate is a charged group and is primarily represented by its charge. For charges qi, we assigned the standard ionization states; namely, all phosphates group in DNA, all the Glu, and Asp residues have -1, and Lys, Arg, and His possess +1 charges. We tested the case that all His charges equal to zero finding that the difference in DNA unwrapping between protonated (charged) and deprotonated (uncharged) His is rather minor (S3 Fig). The difference appears only in free energy depths of high energy meta-stable states.
For a quick test of the model, we calculated the root mean square fluctuations (RMSFs) and compared them with the experimental B-factors (S1 Fig). For histones and DNAs, the RMSFs as a function of residues reproduced major features of the experimental B-factors. Quantitatively, the overall correlation coefficient was 0.80.
In summary, our protein and DNA models are identical to those that have been used in literature, while the specific interaction between the histone octamer and DNA contains, in addition to standard terms, one new parameter εgopro−dna of which value needs to be calibrated.
The nucleosome we simulated is the same molecular complex as the X-ray crystal structure with the pdb code 1KX5 [2,4] (Fig 1). The complex contains dsDNA of 147-bp and a histone octamer. The DNA sequence is palindromic taken from one-half of a human a-satellite sequence repeat. The histone octamer is from Xenopus laevis. Histone tails are explicitly included except the first three residues (PEP) of H2B that are missing in the pdb data. The crystal structure is pseudo-symmetric for 180 degree rotation around an axis that goes through the dyad, the central part of dsDNA. The same pdb structure was used as the reference structure in the structure-based model.
The equation of motion that drives the system is the standard Langevin equation,
mid2ridt2=−∂Vtotal∂ri+miγidridt+miξi
in which the random noise ξi is the Gaussian white noise with the mean and variance,
〈ξi(t)〉=0, 〈ξi(t)ξj(t′)〉=2γikBTmiδ(t−t′)δi,j
respectively. Here, mi is the mass, for which we used CafeMol default value. kB is the Boltzmann constant, and T is the temperature set as 300K. γi is the friction coefficient and we used very low value (0.02 in CafeMol unit) to speed up the dynamics. The unit of time is denoted as t0. An apparent mapping leads to t0 ≈ 0.2 ps although the dynamics realized is known to be accelerated by many factors associated with coarse graining; the absence of side chain atoms, the absence of explicit water molecules, ignorance of hydrodynamic effects, the low friction coefficient, and so on. A time step of 0.1t0 was used for time integration. Each MD simulation contains 108 time steps up to 107t0 time, unless otherwise denoted. In each case, we repeated MD 20 times with different random noises to obtain the structural ensembles. To define which part of dsDNA is unwrapped from the histone octamer, we calculated the deviation dDX of every base pairs in dsDNA from those in the reference X-ray structure in each snapshot. We define the n-bp unwrapped state by that the deviations dDX of 1 to n-th bp from each end of dsDNA are larger than 10 Å and that of n+1-th bp is smaller than 10 Å. We did not impose a box boundary so that, once the histone octamer and DNA dissociates, they do not re-assemble in general. In the forced unwrapping simulation of DNA from nucleosome, in each of two ends of dsDNA, we introduced a virtual particle that is linked to terminal particles by harmonic bonds. Here, terminal particles in each end of dsDNA include 5’ end of one strand and 3’ end of the other strand. We pulled the virtual particles with a constant velocity in opposite directions. The pulling velocity was 10−4 Åt0-1. In this forced unwrapping simulations, we needed slightly different parameters in MD: time step was reduced to 0.05t0 due to a large pulling force. Each MD simulation contains 108 time steps up to 0.5·107t0 time. In each salt concentration case, we repeated MD runs 20 times with different random noises. The force was measured from the lengths of the harmonic bonds attached to the virtual particles. We note that, unavoidably, the pulling speed is orders of magnitude faster than that in referred experiments.
All the MD simulations were performed by CafeMol [42].
The degree of DNA unwrapping is expected to depend on the strength of interaction between the histone octamer and DNA, in which, as described in Methods, an appropriate value of the interaction parameter εgopro−dna is unknown beforehand. Thus, we conducted preliminary simulations of thermal fluctuations of single nucleosome with various values of εgopro−dna in the range 0≤εgopro−dna≤1.0εgopro. Here, we mostly present the results with εgopro−dna=0.8εgopro, which turned out to be a representative value comparing with experimental data [17]. In [17], Widom and his collaborators performed the FRET experiments to monitor spontaneous DNA unwrapping and subsequent rewrapping. The Cy3 donor was introduced in one of several DNA positions, while the Cy5 acceptor was connected to histones so that its distance from the donor was small enough in the wrapped state. Measuring the FRET efficiency for a range of salt concentration, they obtained both kinetic and thermodynamic parameters for partial unwrapping of different levels. In equilibrium, the end of DNA, on average, started unwrapping at the salt concentration ~250 mM, while global unwrapping was observed at and above ~750 mM. As described below, we confirmed that these overall feature can be reproduced with εgopro−dna=0.8εgopro. (Later, to see the dependence of the parameter εgopro−dna, we show some results of εgopro−dna=0.5εgopro, and 1.0 εgopro).
MD simulations of nucleosome at NaCl 300 mM showed repeated partial unwrapping in both the left and right ends of dsDNA (Fig 2). Note that the definition of "left" and "right" is arbitrary and that the bp number runs from 1 in the left end to 147 in the right end. We use them merely to distinguish two ends of dsDNA throughout this paper. In each of snapshots, a central consecutive segment of dsDNA was bound to histone octamer, while the left and/or right ends of dsDNA may be transiently unwrapped. We thus can define, in each snapshot, the first bp of the right unwrapped dsDNA segment and the last bp of the left unwrapped dsDNA segment, of which base pair numbers are illustrated in Fig 2A and 2B, respectively, for a typical trajectory. In preliminary investigation, we explored some other measures, such as distances between chromophore attaching sites in FRET experiments, and angles between two ends of DNA, to quantify the degree of partial unwrapping of DNA. These parameters showed a clear peak when DNA is fully wrapped and a broad distribution when DNA is partially unwrapped due to fluctuation of unwrapped DNA segments. Seeking a better measure that is not subject to such fluctuations of unwrapped DNA and thus is more sensitive, we reached the order parameter defined above.
Clearly, in Fig 2, partial unwrapping occurred repeatedly and transiently. Patterns of unwrapping in the left and right ends were not significantly different, probably due to its pseudo-symmetry in the crystal structure. The figure also suggests that partial unwrapping in each end takes some distinct states. One obvious state assigned is the zero-bp unwrapped state. Among snapshot structures depicted in Fig 2, those in (i) and (ii) have their left ends in the zero-bp unwrapped state. Another clear state observed has about 20-bp unwrapped, which is illustrated in the snapshot (ii) for its right end and the left end of the snapshot (iii). Rather rarely, we observed the third state where about 30-bp are unwrapped as in the snapshot (iv) for its left end. Moreover, we notice that there seem a marginal state with about 10-bp unwrapped.
Partial unwrapping in the left and right ends seems to occur independently. A statistical analysis failed to find any noticeable correlation between unwrapping of two ends (S2 Fig). The degree of correlation, if any, is expected to depend on the salt concentration of the solution because, with a low salt concentration, we expect to have relatively long-range electrostatic interactions. Yet, even at the lowest salt concentration studied here (see below), we did not find any crucial correlation of partial unwrapping in two ends (S2 Fig).
To quantify partially unwrapped states, we obtained probabilities P(bp) of the left and right ends of the unwrapped DNA segment from 20 trajectories, which are shown in Fig 3A (the left end) and Fig 3B (the right end). Equivalently, we also plotted free energy profiles, which are defined as −ln(sample_number × p(bp) + 1) in Fig 3C (the left end) and Fig 3D (the right end). At 300 mM NaCl condition (orange), we see two dominant minima in free energy profile, one at zero-bp and the other at 17-bp unwrapped states. We also find shallower minima at 5-bp and 12-bp unwrapped states, as well as a marginal state at ~ 30-bp unwrapped state. We note that, from the crystal structure, one would expect to have 5-bp periodicity. The histogram shows rather close, albeit not identical, 5-bp periodicity. We note that high free energy part of the profile that was not sampled in the current simulations can be better estimated by use of some enhanced sampling techniques.
In the same way as above, we obtained free energy profiles of the left and right ends of unwrapped bp for various different salt concentrations between 50 mM and 400 mM NaCl (Fig 3C and 3D). At salt concentrations equal to and higher than 500 mM, dsDNA progressively unwrapped and finally dissociated from the histone octamer. We note that the simulation did not account for the re-assembly process. At these conditions, we cannot well-define the free energy profiles and thus we did not plot them. Between 50 mM and 400 mM, as expected, at a higher salt concentration dsDNA tends to be unwrapped more. At 50 mM and 100 mM, we see three distinct states; zero-bp unwrapped as the dominant state, 5-bp, and 12-bp unwrapped states as minor states. At 200 mM, on top of the above three states, we also see the fourth state with ~17-bp unwrapped. At 300 mM and 400 mM, the state with 17-bp unwrapped became the major state, still retaining the above mentioned unwrapped states as metastable states. Thus, as the salt concentration changes, the positions of local minima are mostly kept fixed, while their relative stabilities vary.
DNA unwrapping involves large energy-entropy compensation, somewhat similar to that in protein folding [45]. In comparison with protein folding processes, DNA unwrapping seems less cooperative producing a series of partially unwrapped sub-states. This could be attributed to less flexibility of unwrapped dsDNA than unfolded proteins.
Next, we investigate how unwrapping is affected by the change in the interaction strength between DNA and histone octamer. All the above simulations used the coefficient εgopro−dna=0.8εgopro. Here, we tested a weaker interaction εgopro−dna=0.5εgopro and a stronger interaction εgopro−dna=1.0εgopro.
Fig 4A shows the free energy profile of the left end of wrapped DNA for the weakened case, εgopro−dna=0.5εgopro for salt concentration NaCl 50 mM, 100 mM, and 200 mM. While the overall shapes of the free energy profiles look similar to the default one, the zero-bp unwrapping state was less stable while 5-bp, 12-bp, and 17-bp unwrapped states were more stable. Furthermore, at salt concentration higher than 200 mM, after partial unwrapping, dsDNA was completely dissociated from the histone octamer. These data suggest that εgopro−dna=0.5εgopro lead to less stable nucleosome than the stability experimentally suggested for high affinity sequences, such as Widom's 601 sequence [17], where the average structure showed no partial unwrapping below 250 mM.
Next, we plotted the free energy profile for the left end of wrapped segment in the case of the stronger interaction εgopro−dna=1.0εgopro. As expected, we observed partial unwrapping less frequently than the default case (Fig 4B). Yet, at the salt concentration NaCl 400 mM, we observed partial unwrapping up to 17-bp. Even at the salt concentration of NaCl 1 M, we did not observe any complete unwrapping/ dissociation from DNA. This suggests that, with the strong interaction parameter, the structure-based Go potential alone can stabilize nucleosome no matter how the salt concentration is high within the simulation time scale. This is inconsistent with experimental results since experimentally most nucleosomes are disassembled at the salt concentration as high as 1 M [17].
These surveys of interaction parameter led us to use εgopro−dna=0.8εgopro as the default value in the current work. We note that the affinity of nucleosomal DNA with the histone octamer has a broad range that is more than a thousand, and thus it is not very important to fine tune the interaction parameter. εgopro−dna=0.5εgopro may correspond to a weak affinity sequence, while εgopro−dna=1.0εgopro is for a high affinity sequence. Hereafter, we solely use the default value.
It is interesting to look into conformations of histone tails, which can be correlated with partial unwrapping of DNA. We first plotted the distances rHTO of the histone tail terminal residues from the center of mass of histone octamer core (Fig 5A), averaged over trajectories. We clearly see that, for all histone tails, the tail termini become more distant from the center as the salt concentration increases. This is an expected result because highly basic histone tails are attracted to DNA, and the attraction is stronger at lower salt concentration. At a higher salt concentration, these attractions are weaker and disordered tails take entropically extended conformations. These results are in good agreement with recent simulation results [36]. In Fig 5A, we see that the salt concentration dependence of the distance is markedly stronger for H3 tails than the other three tails.
To understand conformations and interactions of histone tails further, we plotted the probability distribution of the distance rHTO for some salt concentrations (Fig 5B–5D). At 50 mM, rHTO for H3 tails exhibited a bimodal distribution with two peaks at ~35 Å and at ~50 Å. Even for H4 and H2B, we marginally recognize two states, a major state centered at ~60 Å and a minor one around 35 Å. On the other hand, rHTO for H2A has only one peak centered at ~60 Å. Looking at snapshot structures, we identified that 35 Å from the center of histone core corresponds to the boundary between the histone octamer core and the wrapped DNA, which makes concave surface. At 50 mM, many of histone tail terminals were located at the concave surface. We illustrated one snapshot in the left cartoon of Fig 5E where the left H3 tail and the left H2B tail terminals are located in the boundary between histone octamer and DNA. On the other hand, the distance rHTO = 50–60 Å corresponds to the case that histone tails are located near the outer surface of wrapped DNA, which can be seen in the left and right H3 tails and the right H4 tail in the left cartoon of Fig 5E. At higher salt concentration, the rHTO distributions generally moved to the larger distance (Fig 5C and 5D). In particular, at the salt concentration 200 mM all the histone tail terminals show single peak near ~60 Å. Interestingly, the H3 tail has broader distribution than the others perhaps because of its length. In these long distances, the histone tail terminal is not in contact with histone cores or DNA. Instead, at 200 mM, histone tails fluctuate apparently randomly. We note that the use of the Go model in the local potential energy function for disordered tails might have made the tails somewhat more rigid than they are. Yet, we confirmed that, with use of the flexible loop modeling [46], the results are not significantly different (S4 Fig).
Next, we investigate correlation between DNA partial unwrapping and histone protein fluctuations. First, we calculated the deviation dHX of every amino acids in histone octamer from those in the reference X-ray structure in each snapshot. The average of dHX over 20 trajectories is plotted in Fig 6A. As expected, relatively large deviation/fluctuation is localized to N-terminals. Then, we obtained the correlation coefficients between dHX and the unwrapped DNA bp in the left end, which is shown in Fig 6B. We find that the correlation coefficient is relatively large for one of H3 and H2B tails that interact with left end of DNA. These relevant parts were magnified and re-plotted in Fig 6C and 6D. It is reasonable that H3 and H2B tails have larger effects to DNA unwrapping because these two tails have their roots at the side of nucleosome and between two turns of dsDNA. In H3, high correlation appears around 25–45 residues, which is near a root of the tail (the H3 tail corresponds to residues 1–44). In H2B, high correlation is seen around 15–35 residues, which is also close to the root of the tail (the H2B tail is assigned as 4–36 residues). These clearly indicate that the root regions of H3 and H2B tails contribute to stabilize the wrapped DNA. On the other hands, both tail terminals have less correlation to the DNA unwrapping because they are already far from the wrapped DNA and highly fluctuating.
To further clarify the coupling of root regions of H3 and H2B tails with DNA unwrapping, we plotted the distribution (and the average) of deviations dHX’s for H3K36 and H2B K24 as a function of unwrapped DNA bp in Fig 6E and 6F, respectively. For H3K36, the dHX is distributed around 7 Å in the completely wrapped end, while the distribution becomes wider and the average increases to about 10Å when about 5-bp unwrapped. Yet, further unwrapping does not change the dHX distribution significantly. Thus, the H3 tail contributes to stabilize the last ~5-bp of unwrapped DNA. On the other hand, for H2B K24, the dHX distribution shifted to the large displacement when the unwrapped DNA bp reaches to ~20. Thus, H2B tail is suggested to stabilize wrapping of ~20-bp from the end, and thus is important for large-scale partial unwrapping. These results are all consistent with the X-ray crystal structure.
Here, we address effects of modification in individual histone tails on the unwrapping dynamics of nucleosome. As is well-known, N-terminal tails of histones contains quite many positively charged amino acids (K and R), which is expected to have major impact on the nucleosome stability. Acetylations of K and R reduce the positive charges in histone tails. How such reduction in tail charges affects the unwrapping dynamics are, thus, of great interest. We conducted a series of CG MD simulations where one of tails has no charge in its N-terminal tail and compare its unwrapping dynamics with that of the intact one. We note that, in each case, we deleted all the N-terminal tail charge of two molecules of each type of histone.
Fig 7 shows the free energy profiles of individual cases, each of which has zero charge in one of the four histone tails, together with the case of the intact interaction as control (denoted as "canonical"). Overall the unwrapping was more or less enhanced by deleting charge in the tails, but the extent of enhancement depended on the cases.
The most marked effect was observed for the case that we deleted charges in the H3 tails (purple dashed curves in Fig 7). The completely wrapped state became less stable rather significantly and, instead, intermediate states with < 20-bp unwrapped became more stable than the case of the intact interaction for all the salt concentration cases tested in Fig 7. This result suggests that electrostatic interactions between H3 tail and nucleosome core contribute to stabilization of the completely wrapped state. This is reasonable because in the crystal structure (Fig 1), we see that H3 tails are located near the left and right ends of wrapped DNA.
Another tail that affects the unwrapping dynamics markedly was H2B in Fig 7 (red). In this case, the deletion of the charge did not alter stabilities of states with < 20-bp unwrapping, whereas we see dramatic increase in frequency of larger-scale unwrapping with > 20-bp unwrapped. For example, at the salt concentration 300 mM, we see partial unwrapping up to 60-bp only in the case of H2B charge deletion (Fig 7C). This suggests that the charge in H2B tails contributes to stabilize the nucleosome at around 20-bp from the ends of wrapped DNA. Again, this result is perfectly in harmony with the crystal structure where H2B tails (red in Fig 1) interact with the corresponding region of DNA.
Effects of the other two tails, H2A and H4, turned out to be weaker, as in Fig 7. These tails are extended to the side of nucleosome (Fig 1) and thus seem to little contribute to the stability of nucleosomal DNA
Finally, we performed mechanical unwrapping simulations where we attached virtual particles to both ends of DNA and pulled these virtual particles to the opposing directions with a constant velocity while monitoring the pulling force.
At the salt concentration 200 mM, DNA smoothly started unwrapped, and then exhibited a major barrier in the pulling force profile once 20 trajectories were averaged (Fig 8A). This result is qualitatively in good agreement with recent single molecule experiments of mechanical unwrapping [20,22,47,48]. The peak in force profile corresponds to the unbinding at the so-called off-dyad region of DNA, where histone-DNA interactions are supposed to be strong. The peak force (at d = 280 Å) is ~20 pN. At smaller distances, there can be one or more peaks although we cannot rule out the possibility that they are simply noise.
When we decreased the salt concentration up to 100 mM, a stronger histone-DNA interaction made the force profile slightly more structured (Fig 8B). In addition to the large peak, albeit rather faint, we see at least one small peak in the force profile when DNA started unwrapping around d = 220 Å. The peak force is ~30 pN.
Experimentally, the force associated with the major peak was 27 pN at the salt concentration 50 mM in one experiment [22]. Another measurement reported 8–9 pN for a slightly different system [47]. If we directly compare the peak forces of at the same salt concentration, the force from our simulations is somewhat larger than the experimental one. The difference may be attributed to the difference in pulling speed between the simulation and the experiment, which was previously found and argued [34]. In addition, mapping between experimental salt concentration and ionic strength in our simulations may not be quantitative.
We investigated partial unwrapping dynamics of nucleosome by coarse-grained molecular dynamics simulations. Depending on the parameter for protein-DNA interaction strength, partial unwrapping dynamics is altered quite significantly. Of the three parameter values tested, we conclude that the parameter 0.8εgopro is among the most reasonable value. With this parameter, we obtained results that are in agreement with experiments such as single molecule FRET experiment. The simulations showed spontaneous unwrapping from the outer DNA and subsequent rewrapping dynamics. We found several distinct partially unwrapped states of nucleosomes. At a low salt concentration, histone tails mostly sit in the concave cleft between the histone octamer and DNA, tightening the nucleosomal DNA. At a higher salt concentration, the tails tend to be expanded outwards, which led to higher degree of unwrapping. H3 and H2B tail dynamics are markedly correlated with partial unwrapping of DNA, and, moreover, their contributions were distinct. Acetylation in histone tails was mimicked by changing their charges, which enhanced the unwrapping, especially markedly for H3 and H2B tails.
Recently, it has been suggested that histone tail acetylation alters chromatin folding, which may be biologically more relevant effect of acetylation. The method developed here can be straightforwardly extended to poly-nucleosomes, where we can address effects of histone acetylation on the chromatin folding, which will be an interesting future direction.
Although the simulation method used here is general and can be applied to many protein-DNA complexes, the method has many limitations as well and thus there is much room for improvement. First, accurate modeling dsDNA, especially its bending flexibility, is of particular importance for accurate account of DNA unwrapping in nucleosome. Recently, de Pablo group developed a new and refined version of CG DNA model that seems to approximate sequence dependent DNA flexibility rather accurately [49]. Thus, the use of this or other refined methods may be an important extension to the current work. Second, non-specific protein-DNA interactions are dominated by electrostatic interactions, where we used a simple Debye-Huckel screening model. This model is legitimated only for dilute ionic solution of monovalent ions. However, protein-DNA interactions generally involve strong electric field that leads to locally high ion concentration. Small amount of divalent ions is known to alter chromatin folding markedly, which is clearly beyond the range of Debye-Huckel model. More refined treatment of counter ions around the molecules, such as explicit treatment of them, may be a promising direction of improvement [36].
|
10.1371/journal.pmed.1002167 | Measures of Malaria Burden after Long-Lasting Insecticidal Net Distribution and Indoor Residual Spraying at Three Sites in Uganda: A Prospective Observational Study | Long-lasting insecticidal nets (LLINs) and indoor residual spraying of insecticide (IRS) are the primary vector control interventions used to prevent malaria in Africa. Although both interventions are effective in some settings, high-quality evidence is rarely available to evaluate their effectiveness following deployment by a national malaria control program. In Uganda, we measured changes in key malaria indicators following universal LLIN distribution in three sites, with the addition of IRS at one of these sites.
Comprehensive malaria surveillance was conducted from October 1, 2011, to March 31, 2016, in three sub-counties with relatively low (Walukuba), moderate (Kihihi), and high transmission (Nagongera). Between 2013 and 2014, universal LLIN distribution campaigns were conducted in all sites, and in December 2014, IRS with the carbamate bendiocarb was initiated in Nagongera. High-quality surveillance evaluated malaria metrics and mosquito exposure before and after interventions through (a) enhanced health-facility-based surveillance to estimate malaria test positivity rate (TPR), expressed as the number testing positive for malaria/number tested for malaria (number of children tested for malaria: Walukuba = 42,833, Kihihi = 28,790, and Nagongera = 38,690); (b) cohort studies to estimate the incidence of malaria, expressed as the number of episodes per person-year [PPY] at risk (number of children observed: Walukuba = 340, Kihihi = 380, and Nagongera = 361); and (c) entomology surveys to estimate household-level human biting rate (HBR), expressed as the number of female Anopheles mosquitoes collected per house-night of collection (number of households observed: Walukuba = 117, Kihihi = 107, and Nagongera = 107). The LLIN distribution campaign substantially increased LLIN coverage levels at the three sites to between 65.0% and 95.5% of households with at least one LLIN. In Walukuba, over the 28-mo post-intervention period, universal LLIN distribution was associated with no change in the incidence of malaria (0.39 episodes PPY pre-intervention versus 0.20 post-intervention; adjusted rate ratio [aRR] = 1.02, 95% CI 0.36–2.91, p = 0.97) and non-significant reductions in the TPR (26.5% pre-intervention versus 26.2% post-intervention; aRR = 0.70, 95% CI 0.46–1.06, p = 0.09) and HBR (1.07 mosquitoes per house-night pre-intervention versus 0.71 post-intervention; aRR = 0.41, 95% CI 0.14–1.18, p = 0.10). In Kihihi, over the 21-mo post-intervention period, universal LLIN distribution was associated with a reduction in the incidence of malaria (1.77 pre-intervention versus 1.89 post-intervention; aRR = 0.65, 95% CI 0.43–0.98, p = 0.04) but no significant change in the TPR (49.3% pre-intervention versus 45.9% post-intervention; aRR = 0.83, 95% 0.58–1.18, p = 0.30) or HBR (4.06 pre-intervention versus 2.44 post-intervention; aRR = 0.71, 95% CI 0.30–1.64, p = 0.40). In Nagongera, over the 12-mo post-intervention period, universal LLIN distribution was associated with a reduction in the TPR (45.3% pre-intervention versus 36.5% post-intervention; aRR = 0.82, 95% CI 0.76–0.88, p < 0.001) but no significant change in the incidence of malaria (2.82 pre-intervention versus 3.28 post-intervention; aRR = 1.10, 95% 0.76–1.59, p = 0.60) or HBR (41.04 pre-intervention versus 20.15 post-intervention; aRR = 0.87, 95% CI 0.31–2.47, p = 0.80). The addition of three rounds of IRS at ~6-mo intervals in Nagongera was followed by clear decreases in all outcomes: incidence of malaria (3.25 pre-intervention versus 0.63 post-intervention; aRR = 0.13, 95% CI 0.07–0.27, p < 0.001), TPR (37.8% pre-intervention versus 15.0% post-intervention; aRR = 0.54, 95% CI 0.49–0.60, p < 0.001), and HBR (18.71 pre-intervention versus 3.23 post-intervention; aRR = 0.29, 95% CI 0.17–0.50, p < 0.001). High levels of pyrethroid resistance were documented at all three study sites. Limitations of the study included the observational study design, the lack of contemporaneous control groups, and that the interventions were implemented under programmatic conditions.
Universal distribution of LLINs at three sites with varying transmission intensity was associated with modest declines in the burden of malaria for some indicators, but the addition of IRS at the highest transmission site was associated with a marked decline in the burden of malaria for all indicators. In highly endemic areas of Africa with widespread pyrethroid resistance, IRS using alternative insecticide formulations may be needed to achieve substantial gains in malaria control.
| Long-lasting insecticidal nets (LLINs), which prevent mosquitoes from biting people while they sleep, and indoor residual spraying of insecticides (IRS) in houses, which prevents mosquitoes from resting in houses, are the main tools used to prevent malaria in Africa.
Although LLINs and IRS have been shown to be effective, changes in the behavior of mosquitoes and people as well as the emergence of mosquitoes resistant to insecticides could compromise the benefits of these interventions.
Recently the government of Uganda distributed free LLINs throughout the country and began IRS in selected areas. In this “real world” setting, it is important to monitor for changes in the burden of malaria following the scale-up of LLIN distribution and IRS.
The researchers conducted comprehensive malaria surveillance between October 1, 2011, and March 31, 2016, at three sites in Uganda that differed in the intensity of malaria transmission.
Between 2013 and 2014, LLINs were distributed to the entire population at all three sites, and in December 2014, IRS with an insecticide different from that used in the LLINs was started in the site with the highest level of malaria transmission.
The researchers found that following LLIN distribution, there were only modest declines in some measures of malaria burden at the three sites, and no changes in other measures.
In contrast, following the addition of IRS at the highest transmission site, all measures of malaria burden declined dramatically.
The research also documented a high level of resistance to the type of insecticide used in LLINS at all three sites.
These findings suggest that in countries like Uganda, which has a heavy burden of malaria, LLINs alone may not be adequate to substantially drive down the burden of malaria, and the addition of IRS using a different insecticide may be needed to have a major impact.
However, it should be noted that IRS is more expensive and harder to implement than distributing LLINs, and generally needs to be repeated every 6–12 months to have a sustained effect.
| Over the last fifteen years, funding for malaria control activities has increased dramatically across Africa, leading to the scale-up of proven interventions including distribution of long-lasting insecticidal nets (LLINs), indoor residual spraying of insecticide (IRS), and treatment of malaria cases with artemisinin-based combination therapy (ACT) [1]. Substantial declines in measures of malaria burden have been attributed to the expansion of these interventions at various scales, from the sub-national level to the entire continent [2–4]. Despite these advances, the burden of malaria remains high, with an estimated 215 million cases and 438,000 deaths worldwide in 2015, of which 88% of cases and 90% of deaths were in Africa [1].
LLINs have been shown to reduce malaria morbidity and mortality across a range of epidemiological settings, and the World Health Organization (WHO) recommends universal coverage of populations at risk [5,6]. IRS has also been shown to be highly effective, but it is more resource-intensive and expensive to implement than distribution of LLINs [7,8]. Historically, IRS played a key role in the global malaria elimination campaign in the 1950s and 1960s, but was not widely used in sub-Saharan Africa primarily due to limited resources [9]. More recently, the use of IRS in sub-Saharan Africa has been expanded from epidemic-prone areas with seasonal transmission to areas with more intense perennial transmission [10]. However, despite the widespread use of LLINs and IRS, and the importance of quantifying the impact of malaria control interventions in operational settings, high-quality contemporary data are limited. This is largely due to a paucity of rigorous longitudinal malaria surveillance studies that can capture the impact of interventions over time. Indeed, most estimates of the impact of population-level malaria control interventions rely on health facility records, which are often incomplete and/or inaccurate, or repeated cross-sectional surveys that measure parasite prevalence, which provides only an indirect estimate of morbidity.
We report findings from a comprehensive malaria surveillance program conducted in three areas of Uganda with varied transmission intensities from October 2011 to March 2016, a period of major expansion in population-level malaria control interventions. Uganda has reported some of the highest levels of transmission intensity and ranks fourth globally in the estimated number of annual cases of malaria [1,11]. Given that malaria is endemic in over 95% of the country and the burden remains high in many areas, efforts have focused on control and not elimination [12]. From 2012 to 2014, Uganda implemented a national universal LLIN distribution campaign, with 21 million LLINs distributed to a population of approximately 35 million people [13]. In December 2014, IRS was implemented for the first time in one of our three study areas. Evaluations included enhanced health-facility-based surveillance to estimate malaria test positivity rate (TPR), cohort studies to estimate the incidence of malaria, entomology surveys to estimate transmission intensity, insecticide susceptibility testing, and repeated cross-sectional community surveys to estimate the coverage level of key malaria control interventions.
Ethical approval was obtained from the Makerere University School of Medicine Research and Ethics Committee, the Uganda National Council for Science and Technology, the London School of Hygiene & Tropical Medicine Ethics Committee, the Durham University School of Biological and Biomedical Sciences Ethics Committee, and the University of California, San Francisco, Committee on Human Research.
Comprehensive surveillance studies were conducted in three sub-counties in Uganda: Walukuba, Jinja District; Kihihi, Kanungu District, and Nagongera, Tororo District (Fig 1). These areas were chosen to represent varied malaria transmission settings. Walukuba is a relatively low transmission, peri-urban area near Lake Victoria in the south-central part of the country. Kihihi is a rural area with moderate transmission intensity bordering a national park in the southwestern part of the country. Nagongera is a rural area with high transmission intensity in the southeastern part of the country near to the border with Kenya (Fig 1). Transmission in all of these areas is perennial, with two annual peaks following the rainy seasons.
Enhanced malaria surveillance was conducted at one government-run level IV health center at each study site, as described previously [15]. Briefly, individual-level data on all patients presenting to the outpatient department of the facilities were collected electronically using a standardized data collection form. Data collected included patient age, whether symptomatic malaria was suspected (as determined by the evaluating clinician), whether laboratory testing for malaria was done, the type of laboratory testing done (either light microscopy or rapid diagnostic test), and the results of the laboratory test. Study staff visited the health facilities periodically to provide training and feedback including quality control of diagnostic testing. Enhanced surveillance was begun in 2006, but this report only includes data from October 1, 2011, to March 31, 2016, when data became available from the other malaria surveillance studies described below.
Cohort studies were performed in children from households randomly selected from enumeration surveys conducted in each of the three sub-counties, as previously described [16]. Briefly, all eligible children aged 0.5–10 y were enrolled from 100 households from each study site between August and September 2011. The cohorts were dynamic, such that all newly eligible children were enrolled during follow-up and study participants who reached 11 y of age were excluded from further follow-up. At enrolment, written informed consent was provided by parents/guardians and study participants were given an LLIN and underwent a standardized evaluation. Cohort study participants received all medical care free of charge at designated study clinics open every day. Parents/guardians were encouraged to bring their children to the clinic any time they were ill and were reimbursed for transport costs. Children who presented with a documented fever (tympanic temperature ≥ 38.0°C) or history of fever in the previous 24 h had blood obtained by finger prick for a thick blood smear. If the smear was positive for malaria parasites, the patient was diagnosed with malaria. Episodes of uncomplicated malaria were treated with artemether-lumefantrine, and episodes of complicated malaria or recurrent malaria occurring within 14 d of prior therapy were treated with quinine. Study participants were withdrawn from the study for (a) permanent movement out of the sub-county, (b) inability to be located for >4 mo, (c) withdrawal of informed consent, (d) inability to comply with the study schedule and procedures, or (e) reaching 11 y of age. Children from 31 randomly selected additional households were enrolled between August and October 2013 to replace households in which all study participants had been withdrawn and were followed using the same procedures described above. A full description of the selection of study households and participants is provided in S1 Fig.
For each household participating in the cohort studies, entomological surveys were performed, as described previously [16,17]. Briefly, mosquitoes were collected once a month from each cohort study household using miniature CDC light traps (Model 512; John W. Hock Company) with the light positioned 1 m above the floor at the foot end of the bed where a cohort study participant slept. Traps were set at 19.00 h and collected at 07.00 h the following morning by field workers. Female Anopheles mosquitoes were identified taxonomically to species level based on morphological criteria according to established taxonomic keys [18]. Members of the A. gambiae complex were identified by PCR [19] for 30 mosquitoes randomly selected at each site each month. Sporozoites were identified in individual mosquitoes stored with desiccant using an ELISA technique [20]. All female Anopheles mosquitoes captured in Walukuba and Kihihi were tested for sporozoites. Due to the large numbers of female Anopheles mosquitoes captured in Nagongera, only up to 50 randomly selected mosquitoes per household per night of collection were tested for sporozoites.
Annual cross-sectional surveys were conducted in each of the study sites in 2012, 2013, and 2015, as previously described [14]. Briefly, for each survey, households were randomly selected from our enumeration list and sequentially screened until 200 households were enrolled. The purpose of the study was discussed with the head of the household or their designate, and consent to participate in the survey was sought. Households with no adult respondent during the initial contact were revisited up to three times before excluding them from the sample selection. After obtaining written informed consent, a household questionnaire was administered to the head of the household or their designate. The questionnaire was used to capture information on the demographics of all household members and the use of malaria prevention methods. The surveys were conducted over the same 2-mo period each year in each of three sub-counties: Walukuba, Jinja District (March–April); Kihihi, Kanungu District (May–June); and Nagongera, Tororo District (January–February).
Prior to 2012, a number of sub-national LLIN distribution campaigns were carried out in Uganda. From September 2012 through August 2014, the government of Uganda carried out a countrywide mass distribution of free LLINs. An estimated 21 million LLINs were distributed, with the goal of achieving universal coverage with at least one LLIN for every two people. For our study sites, LLINs were distributed during November 2013 in Jinja and Tororo Districts, which included Walukuba and Nagongera, and during June 2014 in Kanungu District, which included Kihihi. Data on the population at risk and the number of LLINs distributed at our study sites were obtained from the Ugandan National Malaria Control Program (NMCP).
In 2006, Uganda initiated an IRS program, initially focusing on epidemic-prone areas in the southwestern part of the country. In 2009, the IRS program was moved to ten northern districts with high transmission intensity. In 2014–2015, the IRS program was moved to 14 districts in the Lango, Bukedi, and Teso sub-regions located in the central and eastern part of the country [13]. With respect to our study sites, Walukuba sub-county has not received IRS, and Kihihi sub-county received a single round of IRS using the pyrethroid lambda-cyhalothrin in February–March 2007. In Tororo District, including Nagongera sub-county, the first round of IRS using the carbamate bendiocarb was delivered in December 2014–February 2015, a second round in June–July 2015, and a third round in November–December 2015, with plans to continue IRS every 6 mo for at least 3 y. Data on the number of households targeted and the number that received IRS in Nagongera were obtained from the Ugandan NMCP.
Estimates of monthly rainfall were obtained from the NASA Tropical Rainfall Measuring Mission [21]. Insecticide susceptibility testing was conducted at the three study sites from January to June 2014, as described previously [22]. Briefly, mosquitoes were collected as larvae using the dipping method from a variety of breeding sites. Larvae were transferred to an insectary and reared to adulthood. Emerging adults were fed on a 10% sugar solution and identified as belonging to the A. gambiae species complex using morphological keys. Non-blood-fed female mosquitoes (3–5 d old) were exposed to insecticide-treated papers impregnated with WHO diagnostic concentrations (4% DDT, 0.05% deltamethrin, 0.75% permethrin, 0.1% bendiocarb, 1% fenitrothion, and 5% malathion). Batches of 20–25 mosquitoes were exposed to each insecticide for 1 h, and mortality scored 24 h post-exposure in accordance with standard WHO insecticide susceptibility testing procedures.
One of the primary objectives of establishing our comprehensive malaria surveillance study was to estimate the changes in measures of transmission intensity, infection, and disease using surveillance data at multiple sites in Uganda following the implementation of malaria control measures (as described in S1–S4 Texts). It was originally anticipated that community-level changes in the coverage of key malaria control interventions would be gradual. To make causal inferences on the effect of these interventions, a counterfactual framework was originally planned to estimate what the outcome variable of interest would be if the malaria control intervention were set to a specific level. However, following the establishment of our comprehensive malaria surveillance study, the Uganda NMCP rapidly implemented two major interventions (universal LLIN distribution and IRS) covering our study sites. Given this opportunity, we adjusted our analytical plan and performed the “before and after” comparisons described below.
All data were collected using standardized data collection forms and entered using Microsoft Access. Analyses were performed using Stata version 14 (StataCorp) and R version 3.2.1 (https://www.r-project.org/). Data for all analyses described below covered the period from October 1, 2011, through March 31, 2016. For health-facility-based surveillance, the primary metric was the TPR, defined as the proportion of patients tested for malaria who tested positive by microscopy or rapid diagnostic test. Data from health-facility-based surveillance used in this report only included children 0.5 to 10 y of age to mirror the age range included in the cohort studies. For the cohort studies, the primary metric was malaria incidence, defined as the number of new episodes of malaria per person-year [PPY] of observation. New episodes of malaria were defined as any episode of laboratory-confirmed malaria not preceded by another episode of malaria in the prior 14 d. For the entomology surveys, the primary metric was the daily human biting rate (HBR), defined as the total number of female Anopheles mosquitoes captured per house-night of collection from the same households participating in the cohort studies. Monthly time series were created for each of the three surveillance methods. An LLIN variable was created with time periods corresponding to the dates before and after the universal distribution campaigns were completed at each study site (cutoff at December 1, 2013, for Walukuba and Nagongera and July 1, 2014, for Kihihi). For Nagongera, an IRS variable was created with time periods corresponding the dates after the LLIN distribution campaign through 2 mo following the initiation of the first round of IRS (December 1, 2013–January 31, 2015) and to 2 mo after the initiation of each subsequent round of IRS (cutoffs at February 1, 2015, August 1, 2015, and January 1, 2016). Autoregressive integrated moving average (ARIMA) models were used and, specifically, the ARIMAX form of ARIMA models was used for this study, which is a multivariate transfer function model and includes current and/or past values of independent variables as predictors [23]. ARIMAX models were developed to estimate the adjusted rate ratios (aRRs) when comparing metrics before and after the interventions. Potential confounders included in the models were age, gender, and rainfall at a 1-mo lag. Seasonal terms were examined for each model, and candidate models were selected through the inspection of residual autocorrelation diagnostics via the autocorrelation function, the partial autocorrelation function, and the Ljung-Box test. The Akaike information criteria of all candidate models were compared for selection of the final model. Additional details of the ARIMA models used in the time-series analyses are provided in S1 Table. A p-value < 0.05 was considered statistically significant.
This study provided the opportunity to evaluate relationships between temporal changes in malaria incidence measured in the cohort studies and TPR measured by health-facility-based surveillance. While a direct measure of malaria incidence is considered the gold standard for estimating malaria morbidity, the TPR offers a surrogate measure [24]. Assuming that patients undergoing laboratory testing for malaria at a health facility are representative of the catchment population, the relationship between the TPR and the true incidence of malaria (Im) and non-malarial fevers (Inm) for any time interval can be defined as follows:
TPR1−TPR=ImInm
From the cohort studies, the observed relative monthly change in the incidence of malaria (observed rΔIm) was defined as Imi+1/Imi, where i represents the month of observation. For the health-facility-based surveillance, rΔIm can be predicted from the TPR using the following formula:
predicted rΔIm=Inmi+1Inmi×TPRi+1(1−TPRi)TPRi(1−TPRi+1)
For health-facility-based surveillance, the incidence of non-malarial fevers in the catchment populations was unknown. Therefore, estimates of the incidence of non-malarial fevers from the cohort studies (defined as the number of new episodes of fever with a negative blood smear PPY of observation) were used when calculating the predicted rΔIm. Relationships between relative monthly changes in the incidence of malaria observed from the cohort studies and those predicted from health-facility-based surveillance (log transformed) were investigated using the Pearson correlation coefficient.
Health-facility-based surveillance involved a total of 110,313 outpatient visits among children 0.5 to 10 y of age from all three sites combined over the 4.5-y observation period (Table 1). The proportion of visits for which malaria was suspected ranged from 54.9% to 82.4% across the three sites. Over 98% of patients with suspected malaria underwent laboratory testing at all three sites, and the TPR ranged from 26.4% in Walukuba to 48.4% in Kihihi. For the cohort studies, a total of 1,081 children were observed over 3,258 person-years. A total of 5,213 episodes of malaria were diagnosed, with an incidence ranging from 0.29 episodes PPY in Walukuba to 2.41 episodes PPY in Nagongera. Only 12 episodes of malaria (0.2% of total) met criteria for severe malaria (5 severe anemia, 4 multiple convulsions, 2 cerebral malaria, and 1 respiratory distress). There were no deaths due to malaria. Two children with negative blood smears died of diarrheal illnesses. Monthly entomology surveys were conducted in 331 households involving 15,206 nights of collection. A total of 155,613 female Anopheles mosquitoes were collected, demonstrating daily HBRs ranging from 0.88 in Walukuba to 26.12 in Nagongera and sporozoite rates ranging from 0.84% in Walukuba to 1.84% in Nagongera. Estimates of the annual entomological inoculation rate were 2.71, 20.90, and 175.54 infectious bites PPY in Walukuba, Kihihi, and Nagongera, respectively (Table 1). The primary vector species in Walukuba was A. arabiensis, followed by A. gambiae s.s. and A. funestus. In Kihihi almost all mosquitoes were A. gambiae s.s., and in Nagongera the primary vector was A. gambiae s.s., followed by A. arabiensis and A. funestus (Table 1). Insecticide susceptibility testing was performed using WHO bioassays for available vector species from the study sites in 2014. Testing of A. gambiae s.s. in Kihihi and Nagongera revealed moderate to high resistance to DDT and pyrethroids (deltamethrin and permethrin), lower resistance to bendiocarb, and full susceptibility to organophosphates (fenitrothion and malathion). Testing of A. arabiensis in Walukuba and Nagongera revealed low resistance to DDT (Walukuba only), high resistance to pyrethroids, and full susceptibility to bendiocarb and organophosphates (Fig 2).
Coverage levels of key malaria control interventions were obtained from repeated cross-sectional community surveys, routine assessments of cohort participants, and data from the Ugandan NMCP. In 2012, the proportion of households with at least one LLIN ranged from 51.0% in Kihihi to 78.5% in Nagongera, with slightly lower estimates in 2013. Considering the proportion of households with at least one LLIN per two persons, coverage levels were relatively low at all sites in 2012 and 2013 (range 17.0%–35.5%), consistent with a lack of sufficient numbers of LLINs even among households that owned at least one. According to data from the NMCP, the number of LLINs distributed in our study sites ranged from 0.63 per person in Kihihi to 0.80 per person in Walukuba. In cross-sectional surveys conducted 10–17 mo following the universal LLIN distribution campaigns, the proportion of households with at least one LLIN increased from the prior survey modestly in Walukuba (from 49.0% to 65.0%) and more substantially in Kihihi (from 37.5% to 86.5%) and Nagongera (from 71.0% to 95.5%). Considering the proportion of households with at least one LLIN per two persons, increases from the prior survey were even greater at all sites (Walukuba 31.0% to 52.0%, Kihihi 19.0% to 73.0%, and Nagongera 22.5% to 62.0%; Table 2). All cohort study participants and their family members were provided LLINs at enrolment. Over 99% of children were reported to have slept under an LLIN the prior evening at the time of routine visits. Government-supported IRS campaigns were not conducted in Walukuba and Kihihi, and cross-sectional survey data revealed only a few households that reported recent IRS use. In Nagongera, an IRS campaign was begun in December 2014, with plans to spray with the carbamate bendiocarb every 6 mo. According to data from the NMCP, over 95% of houses were sprayed during each of the three rounds. Data from our cross-sectional survey conducted in January–February 2015 estimated that 78% of houses received IRS, but this is likely an underestimate of the proportion of houses that were ultimately sprayed, as the first round of IRS was ongoing when this survey was conducted (Table 2). Among the households participating in the cohort study, over 94% received IRS during each of the three rounds conducted in Nagongera. ACT coverage among children reportedly treated for malaria from cross-sectional surveys was 75% or higher across all three surveys at all three sites, with the exception of Walukuba in 2013 (Table 2). Children in the cohort studies received treatment with artemether-lumefantrine for all episodes of laboratory-confirmed uncomplicated malaria.
Temporal trends in monthly crude estimates of malaria metrics from the different study sites are presented in Fig 3, and changes following universal LLIN distribution and IRS using time-series analyses to adjust for secular trends are presented in Table 3. In Walukuba, the lowest transmission site, malaria metrics were declining over the 26-mo period prior to universal distribution of LLINs. Over the 28-mo post-intervention period, universal LLIN distribution was associated with reductions in the TPR (aRR = 0.70, 95% CI 0.46–1.06, p = 0.09) and HBR (aRR = 0.41, 95% CI 0.14–1.18, p = 0.10) that did not reach statistical significance and no change in the incidence of malaria (aRR = 1.02, 95% CI 0.36–2.91, p = 0.97). In Kihihi, over the 21-mo post-intervention period, universal LLIN distribution was associated with a reduction in the incidence of malaria (aRR = 0.65, 95% CI 0.43–0.98, p = 0.04) but no significant change in the TPR (aRR = 0.83, 95% 0.58–1.18, p = 0.30) or HBR (aRR = 0.71, 95% CI 0.30–1.64, p = 0.40). In Nagongera, the highest transmission site, over the 12-mo post-intervention period prior to the implementation of IRS, universal LLIN distribution was associated with a reduction in the TPR (aRR = 0.82, 95% CI 0.76–0.88, p < 0.001) but no significant change in the incidence of malaria (aRR = 1.10, 95% CI 0.76–1.59, p = 0.60) or HBR (aRR = 0.87, 95% CI 0.31–2.47, p = 0.80). In contrast, compared to the 14-mo period following universal LLIN distribution, the added implementation of three rounds of IRS over a 14-mo period in Nagongera was associated with marked reductions in the TPR (aRR = 0.54, 95% CI 0.49–0.60, p < 0.001), incidence of malaria (aRR = 0.13, 95% CI 0.07–0.27, p < 0.001), and HBR (aRR = 0.29, 95% CI 0.17–0.50, p < 0.001).
This study provided the opportunity to evaluate relationships between temporal changes in two complementary measures of malaria morbidity: malaria incidence measured in the cohort studies and TPR measured by health-facility-based surveillance. Linear correlations between relative monthly changes in the incidence of malaria observed in the cohort studies and predicted from the health-facility-based surveillance TPRs are presented in S2 Fig. At all three sites, there were significant positive correlations between these two measures, with correlation coefficients ranging from 0.39 in Kihihi to 0.66 in Nagongera. The slopes for the lines of best fit were <1 for all three sites, indicating that predictions from the TPRs tended to be lower than observations from the cohort studies. The cumulative monthly relative changes in the incidence of malaria observed in the cohort studies and predicted from the health-facility-based surveillance TPRs are presented in S3 Fig. These two measures tracked well over time and in relation to the implementation of LLIN distribution and IRS, although malaria incidence relative to baseline predicted from the TPRs tended to be lower than what was observed from the cohort studies.
We utilized a comprehensive malaria surveillance system in three sites in Uganda with varied malaria epidemiology to measure changes in malaria metrics and mosquito exposure before and after malaria control interventions were delivered under operational conditions at the population level. The primary intervention was a national universal LLIN distribution campaign, with the goal of providing one LLIN for every two persons. The distribution campaign substantially increased LLIN coverage levels, but did not reach the universal coverage target. LLIN distribution was associated with modest reductions in malaria TPRs at all three sites, but the reduction reached statistical significance only in the highest transmission intensity site (Nagongera). There was a reduction in the incidence of malaria only in the medium transmission site (Kihihi) and reductions in the HBR that did not reach statistical significance at any of the sites. In contrast, in the highest transmission site (Nagongera), delivery of three rounds of IRS with the carbamate bendiocarb was associated with marked declines in all three malaria metrics. Notably, we documented high-level pyrethroid resistance among A. gambiae s.s. and A. arabiensis vectors across our study sites, which may have contributed to the limited changes seen following the distribution of LLINs. Our results suggest that IRS with non-pyrethroid insecticides is currently the most effective intervention available for reducing the burden of malaria in Uganda in areas where maximum bednet coverage is not obtained despite attempts at universal distribution.
Insecticide-treated nets (ITNs), including LLINs, are the most widely used intervention for the control of malaria in Africa, and universal coverage is recommended by WHO. The benefits of ITNs have been well established in several randomized controlled trials, with use of ITNs associated with a reduction in the incidence of malaria of 50% and a reduction in child mortality of 20% [5]. Analyses of observational data from cross-sectional surveys have suggested that the protective efficacy of ITNs under operational settings is similar to that observed in controlled trials [6]. Recently, using a large database of malaria field surveys, it was estimated that the incidence of malaria decreased by 40% across sub-Saharan Africa between 2000 and 2015, and that ITNs were responsible for 68% of cases averted [2]. Although these estimates are encouraging, they are based on measurement of parasite prevalence and not disease incidence. The relationship between parasite prevalence and the incidence of malaria remains uncertain, and contemporary, high-quality longitudinal data to clarify the relationship between these indicators are limited.
The scale-up of ITN coverage in Uganda (and many other countries in sub-Saharan Africa) has been impressive. Based on national surveys, the proportion of households in Uganda reporting ownership of at least one ITN increased from 16% in 2006 to 60% in 2011 following a series of sub-national distribution campaigns, and then to 90% in 2014 following the universal LLIN distribution campaign [25–27]. The average number of ITNs per household increased from 0.8 in 2009 to 2.5 in 2014 [27]. Although quality surveillance data to estimate the impact of the initial scale-up of ITN provision in Uganda are lacking, there is evidence from our study of a decline in the burden of malaria preceding the universal LLIN campaign in Walukuba. It is unclear what was responsible for this initial decline and why it was seen only in Walukuba, although this may have been influenced by the lower level of transmission intensity and greater urbanization in Walukuba compared to the other two sites [16,28].
Despite gradual improvements in the coverage levels of ITNs over the last 10 y in Uganda, evidence from this study suggests that the universal LLIN distribution campaign was followed by only modest declines in some malaria metrics at our three study sites. There are several potential explanations for these disappointing results. Perhaps of greatest concern is the recent emergence and spread of resistance to pyrethroids, the only class of insecticides available for LLINs. We documented high-level pyrethroid resistance among A. gambiae s.s. and A. arabiensis vectors in our study sites, a trend that is occurring across Africa [29]. Estimating the role of pyrethroid resistance in the protective efficacy of LLINs under operational conditions remains a significant challenge. However, increases in malaria cases have been attributed to the emergence of pyrethroid resistance in longitudinal studies from South Africa and Senegal [30,31]. Also of concern are putative changes in vector behavior and shifts in the relative abundance of vector species, which may increase (or leave unchanged) exposure risk during the early evening hours while people are outside of their bed nets and unprotected by LLINs [32,33]. In our study, vector behavioral characteristics were not assessed, and although there were differences in species composition between the three study sites, there was no evidence of a shift in species composition over time at any of the sites. Finally, it is possible that the LLINs were indeed effective, but that the lack of significant changes in malaria metrics was due to inadequate coverage resulting from insufficient numbers of LLINs, loss of nets, or poor compliance. Even after the LLIN distribution campaign, fully universal coverage levels were not achieved. We did not assess LLIN durability and compliance, so we are unable to characterize the contributions of these factors to the outcomes.
Historically, IRS has played a major role in the elimination of malaria in several countries outside of Africa and in greatly reducing the burden of malaria in parts of Africa with low or seasonal transmission [9,34]. However, the evidence base from randomized controlled trials for IRS efficacy when used alone is limited [35]. A number of recent observational studies and cluster randomized trials from Africa comparing the efficacy of IRS combined with ITNs versus either intervention alone have provided mixed results. In observational studies from Equatorial Guinea and Mozambique, IRS combined with ITNs was associated with a lower odds of parasitemia measured in cross-sectional surveys compared to either intervention alone [36]. An analysis of cross-sectional survey data from 17 countries in sub-Saharan Africa indicated that the combination of IRS and ITNs was associated with a lower risk of parasitemia compared to either intervention alone in medium and high transmission areas [37]. In an observational study from western Kenya using a prospective cohort study design, the combination of IRS and ITNs was associated with a lower incidence of new infection compared to ITNs alone [38]. In another observational study from western Kenya, in a setting with high transmission intensity and moderate ITN coverage, two rounds of IRS with pyrethroid insecticides was associated with a lower odds of parasitemia from cross-sectional surveys compared to ITNs alone [39]. In contrast, observational studies from Eritrea and Burundi demonstrated a protective effect of IRS and ITNs when used together, but failed to show any added benefit of the combination compared to either intervention alone [40,41].
Results from cluster randomized trials evaluating ITNs and IRS have also provided conflicting findings. In a study from northwest Tanzania, in a setting with high levels of pyrethroid resistance, 50 clusters received ITNs as part of a universal coverage campaign, and 25 clusters were randomly assigned to additionally receive two rounds of IRS with bendiocarb. The addition of IRS was associated with a significant reduction in parasite prevalence in cross-sectional surveys of children, and there was a non-significant tendency towards a lower entomological inoculation rate [42]. Although reported ITN use was suboptimal in this trial, in subgroup analyses ITNs provided some individual protection and IRS provided additional protection among both net-users and non-users [8]. In contrast, two studies from West Africa failed to show any additional benefit of IRS over ITNs alone. In a high transmission area of Benin with moderate pyrethroid resistance, the addition of IRS with bendiocarb or carbamate-treated plastic sheeting to targeted or universal LLIN coverage did not provide benefit in reducing the incidence of malaria or parasite prevalence in cohorts of children [43]. In a moderate transmission area of the Gambia with susceptible vectors, the addition of IRS with DDT to a background of high LLIN coverage did not provide any benefit in reducing the incidence of malaria, parasite prevalence, or measures of transmission intensity [44]. The results of these studies and our new findings suggest a pattern. Adding IRS to LLINs appears to be most effective in areas where LLIN coverage is low and/or pyrethroid resistance is high. In addition, in areas where LLIN coverage is high, IRS may be most effective when using non-pyrethroid-based insecticides, in line with current WHO recommendations, and as seen in our study.
There were several limitations to our study, and results should be interpreted with caution. The main limitation was the use of an observational study design, comparing outcomes before and after the interventions were implemented. Although we utilized a rigorous analytical approach to adjust for secular trends, the lack of contemporaneous control groups limited our ability to make rigorous causal inferences. In addition, the interventions were implemented under programmatic conditions and not in the setting of a rigorous clinical trial. This is of particular relevance for estimates of changes in malaria metrics following the universal LLIN distribution campaign, which was preceded by other sub-national distribution campaigns. Also these estimates do not include consideration of adherence or net quality. Because cohort participants received LLINs at the start of the study, the cohort study and entomology survey data utilized in this study provided estimates of changes in malaria incidence and HBR following universal LLIN distribution in the setting of existing LLIN ownership. In contrast, data from the health-facility-based surveillance measured direct changes in malaria TPR following universal LLIN distribution. Thus, these measures are complementary. Despite these limitations, our study benefitted from a comprehensive malaria surveillance system focusing on multiple indicators measured longitudinally, including disease incidence, a major improvement compared to cross-sectional surveys measuring only parasite prevalence.
Uganda is representative of several countries in sub-Saharan Africa that have made great strides in reducing the burden of malaria but face substantial challenges on the road to elimination. LLINs and IRS remain the primary interventions for the prevention of malaria in Africa; however, their relative roles in the setting of limited resources are controversial. Like most African countries, Uganda has focused on maximizing coverage with LLINs. Over the last decade, this has led to dramatic increases in coverage levels, culminating in a national universal LLIN distribution campaign resulting in 90% of households owning at least one LLIN and 72% of persons reportedly sleeping under an LLIN [27]. IRS has also become a key component of Uganda’s malaria control strategy, but resource constraints have limited its use to selected areas of the country, with less than 10% of the population protected by IRS, a level similar to that in several other African countries [10]. In this study, following the recent universal LLIN distribution campaign, we found evidence of a modest decline in the burden of malaria at one relatively low transmission site (Walukuba), but no reduction in two higher transmission sites (Kihihi and Nagongera). In contrast to the limited changes following LLIN distribution, three rounds of IRS with a carbamate insecticide at one high transmission site (Nagongera) was followed by a marked decline in the burden of malaria. These data strongly suggest that IRS with non-pyrethroid-based insecticides in combination with LLINs is the most effective intervention currently available for reducing the burden of malaria in Uganda. Resources are needed to increase LLIN coverage and expand the IRS program to cover a greater proportion of the Ugandan population. Given the complex and dynamic nature of vector populations, insecticide resistance patterns, local epidemiology, and the operational effectiveness of malaria control interventions, it is not possible to develop a “one size fits all” approach to malaria control in Africa. More resources are needed to support high-quality malaria surveillance to assess the effectiveness of malaria control interventions over time and space to support evidence-based policy decision making.
|
10.1371/journal.ppat.1004244 | Genetic Analysis of Leishmania donovani Tropism Using a Naturally Attenuated Cutaneous Strain | A central question in Leishmania research is why most species cause cutaneous infections but others cause fatal visceral disease. Interestingly, L. donovani causes both visceral and cutaneous leishmaniasis in Sri Lanka. L. donovani clinical isolates were therefore obtained from cutaneous leishmaniasis (CL-SL) and visceral leishmaniasis (VL-SL) patients from Sri Lanka. The CL-SL isolate was severely attenuated compared to the VL-SL isolate for survival in visceral organs in BALB/c mice. Genomic and transcriptomic analysis argue that gene deletions or pseudogenes specific to CL-SL are not responsible for the difference in disease tropism and that single nucleotide polymorphisms (SNPs) and/or gene copy number variations play a major role in altered pathology. This is illustrated through the observations within showing that a decreased copy number of the A2 gene family and a mutation in the ras-like RagC GTPase enzyme in the mTOR pathway contribute to the attenuation of the CL-SL strain in visceral infection. Overall, this research provides a unique perspective on genetic differences associated with diverse pathologies caused by Leishmania infection.
| Visceral leishmaniasis is one of the most lethal parasitic diseases, and the mechanisms that govern its survival in visceral organs are not understood. Here, we obtained an atypical cutaneous Leishmania donovani clinical isolate from Sri Lanka and compared it to a typical visceral disease causing clinical isolate. Through whole genome sequencing, bioinformatics analysis, experimental infection in mice and functional genomic analysis, this study provides novel information on what differentiates a deadly visceral strain from a benign cutaneous strain. Results indicate that the ability of Leishmania parasites to cause visceral or cutaneous leishmaniasis may be determined by mutations or amplification of a few genes, or combinations of these factors. Overall, this work contributes to the understanding of parasite virulence and may help guide disease control efforts.
| Leishmaniasis is a neglected tropical disease present in 98 countries, with over 350 million people at risk of infection and is caused by Leishmania protozoan parasites transmitted by infected sand flies [1], [2]. Visceral leishmaniasis is the most serious form of this disease and it is among the most lethal parasitic infections after malaria. Cutaneous leishmaniasis in comparison causes skin lesions which usually self-heal. Over 20 Leishmania species can infect humans; however only the Leishmania donovani complex including L. infantum cause the vast majority of visceral leishmaniasis cases worldwide [2], [3].
In Sri Lanka, an atypical L. donovani (strain MON-37) has been responsible for thousands of cutaneous leishmaniasis cases in the past decade [4], [5], [6]. This is of considerable interest because L. donovani typically causes visceral leishmaniasis in Asia and Africa. Visceral leishmaniasis is rare in Sri Lanka, with the first recorded case only in 2007 [7] and only 4 cases of autochthonous visceral leishmaniasis reported so far. We have recently demonstrated that the visceral leishmaniasis-causing strain is also L. donovani MON-37 [8]. It is however unknown whether the same or different sub-strains of L. donovani MON-37 are responsible for visceral and cutaneous leishmaniasis in Sri Lanka.
Among the most important questions in Leishmania research is why some species remain at the site of the sandfly bite and cause cutaneous infections and others metastasize to the internal organs where they cause visceral disease. By comparing genomes of Leishmania species which cause different pathology, we previously identified several L. donovani genes including A2 and Ldbpk_280340 which are required for visceral organ tropism [9], [10]. However, comparing genomes of different Leishmania species is insufficient to fully define determinants of disease tropism and pathology because of their evolutionary distance, introducing genetic changes unrelated to human pathology [3]. A more effective strategy is to compare genomes of closely related Leishmania isolates of the same species that cause different human pathologies. We therefore undertook a phenotypic and genotypic analysis of L. donovani clinical isolates derived from cutaneous and visceral leishmaniasis patients in Sri Lanka. Characterization of these L. donovani clinical isolates provides invaluable insight into the etiology of visceral and cutaneous leishmaniasis in Sri Lanka and provides unique insight into the genetic basis of visceral leishmaniasis.
The cutaneous leishmaniasis L. donovani isolate (CL-SL) from Sri Lanka was derived from a skin lesion as detailed in methods and the visceral leishmaniasis L. donovani strain (VL-SL) has recently been reported [8]. We initially re-sequenced the 6-phosphogluconate dehydrogenase (6PGDH) isoenzyme gene from the CL-SL and VL-SL isolates to confirm they were both L. donovani MON-37 (Fig. S1).
These isolates were then compared with respect to their ability to cause disease when experimentally introduced into mice. BALB/c mice were injected in the tail vein with CL-SL or VL-SL to compare their ability to cause visceral infections and subcutaneously with CL-SL or VL-SL in the rear footpad to assess cutaneous infection. Liver and spleen parasite burdens were determined 4 weeks after visceral infection, and footpad swelling was monitored for 11 weeks following cutaneous infections. As shown in Fig. 1 A–C, mice injected with the VL-SL isolate had high levels of visceral infection in the liver and spleen. Mice injected with the CL-SL isolate had very little detectable visceral organ infection, demonstrating that the CL-SL parasite has lost the ability to survive in visceral organs (Fig. 1B, C). With respect to cutaneous infections, both the CL-SL and VL-SL isolates demonstrated low virulence; however, the CL-SL isolate was able to induce transient footpad swelling while the VL-SL isolate was unable to do so (Fig. 1D). Although the CL-SL isolate induced more footpad swelling than the VL-SL isolate, this was not associated with a significant increase in parasite number (Fig. S2) and therefore was likely due to a stronger inflammatory response.
Overall, these results conform to Koch's postulates demonstrating that cutaneous and visceral leishmaniasis are caused by distinct but closely related strains of L. donovani. These include 1) isolates were obtained directly from cutaneous or visceral sites of human infection; 2) isolates could be cultured and identified by isoenzyme gene sequencing; 3) isolates induced similar pathologies in experimental animals as in humans; and 4) the same strains could be re-isolated from the experimental infection site. Since the mouse infection phenotypes were similar to the human infections, this would argue that these isolates are representative of the cutaneous and visceral disease causing strains circulating in Sri Lanka.
The above experimental infections demonstrated that the CL-SL and VL-SL L. donovani isolates were phenotypically distinct and therefore provided strong justification for comparing their genomes. Whole genome sequencing was performed using the Illumina GAIIx next generation sequencer >200× coverage as detailed in supplementary methods (Text S1). The sequences of the CL-SL and VL-SL genomes were aligned and compared against the L. donovani reference strain BPK282A1 originating from Nepal [11], [12].
There were no gene deletions detected in either genome. However, coverage analysis of gDNA-seq reads in 10 kb segments from the CL-SL and VL-SL isolates indicated that there were nine regions with copy number variations some of which contained several genes (Table 1). Those higher in the VL-SL isolate included Ldbpk_111220.1 (ABC transporter) repeat cluster, LdbpK_161030.1 thru LdbpK_161110.1 hypothetical gene cluster, and LdbpK_200120.1 (phosphoglycerate kinase B, cytosolic and an rRNA locus in chromosome 27. The species-specific A2 multi-gene family coding regions and the flanking 5′ and 3′ A2rel genes [13] were assembled and examined manually because they are incorrectly assembled in the reference genome BPK282A1. Aligning genomic DNA reads against the reconstructed A2 region revealed the presence of more copies of the A2 genes in the VL-SL strain than the CL-SL strain (Table 1). The CL-SL isolate has higher copy number in a region of chromosome 23 which contains ABC thiol transporters (MRPA) H-region (LdbpK_230290.1), terbinafine resistance locus protein (yip1; LdbpK_230280.1) and regions of chromosome 1 (eukaryotic initiation factor 4a, putative), chromosome 19 (glycerol uptake proteins) and chromosome 29 (hypotheticals).
Differences identified between the CL-SL, VL-SL isolates and the reference L. donovani (BPK282A1) genome are summarized in Table 2. More than 80% of the differences are common to VL-SL and CL-SL revealing that the Sri Lanka isolates are more closely related to each other than to the reference L. donovani strain from Nepal. Furthermore, although the majority of variants are homozygous, about 20% of variants are heterozygous. As expected, SNPs accounts for the vast majority of variants which occur mostly in the intergenic regions, while about 20% of the SNPs are located in coding regions. Interestingly, there are over 70 pseudogenes resulting from frame shifts or stop codon variations which are in common to the Sri Lanka CL-SL and VL-SL isolates yet are functional in the reference L. donovani strain (Table 2). These genes therefore appear not to be essential or are not involved in virulence.
Variations between the VL-SL or CL-SL isolates are among the most interesting, and those specific to the CL-SL isolate would be expected to cause loss of the ability to survive in visceral organs. Surprisingly, as summarized in Table 3, there are only 5 genes unequally affected in the CL-SL and VL-SL isolates by frame shift or altered stop site that could be considered to be pseudogenes. Two genes with frame shifts (LdBPK_312060.1 and LdBPK_320030.1) are only present as homozygous or heterozygous in the VL-SL isolate. There are no homozygous pseudogenes specific to the CL-SL isolate except LdBPK_311390.1 which only differs by 3 amino acids at the C-terminus when compared to its homolog from the VL-SL isolate.
SNPs resulted in 117 (70 homo+47 hetero) non-synonymous and 92 (60 homo+32 hetero) synonymous changes specific to the CL-SL isolate (Tables 2, S1). Although the majority of these non-synonymous SNPs involve hypothetical proteins, 15 of the genes with homozygous mutations in the CL-SL isolate have putative functions (Table S1). Some of the non-synonymous SNPs in these 15 genes are located in functionally important regions of the genes described below. For instance, the non-synonymous change (G226D) in LdBPK_220120.1 phosphoinositide phosphatase is in a highly conserved region. Likewise, the non-synonymous change P147Q in LdBPK_171200.1 is in a highly conserved region of a putative 3-oxo-5-alphasteroid 4-dehydrogenase, responsible for the formation of dihydrotestosterone in higher eukaryotes [14]. LdBPK_252290.1 encodes a putative highly conserved DnaJ family protein that plays a role in protein translation, folding, translocation, and degradation [15]. LdBPK_322160.1 is a putative Rab GTPase family 2 (Rab2) protein involved in membrane trafficking [16]. LdBPK_341910.1 encodes a putative NAD dependent deacetylase which catalyzes NAD+-dependent protein/histone deacetylation and has been shown to regulate gene silencing, DNA repair, metabolic enzymes, and cellular life span [17]. LdBPK_366140.1 encodes a ras-like small GTPase-RagC which may be involved with mTOR regulated translation and cell growth [18]. SNPs in the non-coding region could also have an impact on protein levels since protein expression in Leishmania is regulated at the post-transcriptional level, including via sequences in the mRNA 3′UTR [19].
We considered non-synonymous SNPs in the CL-SL isolate of potential interest if the sequence differed from both the VL-SL isolate and the reference L. donovani strain and if the SNP appeared to be in a functionally important part of the gene as described above. To determine the potential impact of CL-SL specific SNPs on visceral infection and to complement potential functional defects, we cloned 7 corresponding genes from the VL-SL isolate described above, transfected them into the CL-SL isolate and measured the survival of the transgenic parasites in the spleen 4 weeks after infection. All of the CL-SL specific SNPs were verified by Sanger sequencing prior to transfection and their expression in the transfected CL-SL isolate verified by RT-PCR (Fig. 2A). As shown in Fig. 2B, only transfection of the VL-SL Rag C gene into the CL-SL isolate significantly increased infection levels in the spleen. It is noteworthy that the increased infection levels in separate experiments ranged from a 5 to 40 fold and were statistically significant revealing that expression of the VL-SL Rag C gene in CL-SL increased survival in the visceral organs.
The Rag C gene (LdBPK_366140.1) in the CL-SL isolate has a homozygous non-conservative amino acid substitution (R231C). As shown in Fig. 2C, the R231C polymorphism is in a highly conserved amino acid across trypanosomatids. RagC is a small ras-like GTPase involved in the activation of the target of the mammalian rapamycin (mTOR) protein in higher eukaryotes. The mTOR protein is a conserved serine-threonine protein kinase involved in multiple cellular processes including cell stress and proliferation [18]. There are 3 TOR genes (TOR 1–3) identified in the Leishmania genome and TOR 1–2 are essential whereas TOR3 is not essential for survival in the promastigote stage but is required for survival in macrophages and infected mice [20].
Chromosome somy can vary in different Leishmania species [11], [12], [21] and therefore it was necessary to determine if there were differences in chromosome somy that could account for the differences in pathology. As shown in Figure 3A, most of the chromosomes were disomic in both isolates except chromosome 31 which was tetrasomic as previously shown for L. donovani [11] and chromosome 23 which appears to be trisomic. With respect to differences between the isolates, chromosomes 13 and 20 appeared to be trisomic in VL-SL and diploid in CL-SL.
We next examined whether the differences in chromosome somy was reflected in differences at the transcript level. cDNA libraries were prepared using mRNA derived from axenic promastigotes (Pro), axenic amastigotes (Ax) and macrophage-derived amastigotes (Am) and entire chromosome transcript levels were compared between these two isolates (Fig. 3B). Several variations between chromosomes were apparent when the median mRNA levels (normalized to the entire genome) were compared; most appear to be unrelated to somy. For example, the mRNA levels are lower than average for many of the small chromosomes (except chromosome 5); chromosome 31 has average mRNA levels despite being tetrasomic; and the mRNA levels for chromosome 13 are high for both isolates even though chromosome 13 is trisomic in the VL-SL isolate. Despite the fact that chromosomes 9 and 26 are disomic in both isolates, the mRNA level for chromosome 9 is about 40% higher in the VL-SL isolate than in the CL-SL isolate; conversely, the mRNA level for chromosome 26 has at least a 30% increase in the CL-SL isolate. However, consistent with the increased somy, the median mRNA level for chromosome 20 has been increased nearly 1.5 times in the VL-SL isolate. Overall these data suggest that chromosome somy may not be a major factor in the different pathologies caused by these isolates since ploidy is not closely mirrored by transcript level.
When comparing the individual gene transcript levels in different life cycle stages, the vast majority of genes were expressed at similar levels. A list of the most differentially expressed genes with at least a 4 fold difference in mRNA levels in the amastigote infected macrophage stage is shown in Table 4. Genes with at least a 2 fold difference are shown in Table S2 for genes more highly expressed in the VL-SL isolate and Table S3 for genes more highly expressed in the CL-SL isolate. Two gene families identified in Table 4, the ABCA3 and A2 gene families, also have higher gene copy numbers as shown in Table 1. One of the rate limiting enzymes in the glycolysis pathway, glycosomal phosphoglycerate kinase (Ldbpk_200110.1) and two hypothetical proteins were also highly up-regulated in the VL-SL strain. The CL-SL isolate had higher mRNA levels of a ribosomal subunit protein and a putative adaptin protein. Overall, these mRNA level differences could impact energy production, protein synthesis and, with respect to A2, visceral organ survival.
One of the genes with an increased copy number (Table 1) and higher transcript levels (Table 4) in the VL-SL isolate was the A2 gene family. A2 proteins are stress-inducible factors necessary for visceral infection [9], [22]–[24]. We therefore compared sequence reads of the region of chromosome 22 containing several A2 gene family members (see Text S1), and this revealed an amplification of the A2 coding regions in the VL-SL isolate compared to the CL-SL isolate (Fig. 4A). We also performed Western blot analysis to determine whether the increased number of A2 gene sequences resulted in an increase in A2 protein levels. As shown in Figure 4B, there were higher levels of A2 and more A2 protein species in the VL-SL isolate compared to the CL-SL isolate. This observation would be consistent with previous studies showing that A2 genes with greater number of repeat sequences results in larger A2 proteins and are all inducible with heat stress [22]–[24].
Because of the difference in A2 proteins between the VL-SL and CL-SL isolates, we investigated whether this was functionally relevant for the different pathologies displayed by these strains. A2 expression was therefore experimentally increased in the CL-SL strain by transfection with the A2 gene-containing pKSneo plasmid expression vector. This construct was generated from a L. donovani genomic library and previously used to characterize the role of the A2 locus in visceralization [13]. Expression of the A2 transgene product (shown with arrow) was effectively up-regulated by temperature stress (40°C) similar to the endogenous chromosomal A2 genes (Fig. 4C). As shown in Fig. 4D, the CL+A2 transfected parasite expressing the additional A2 transgene displayed a higher level of survival in the spleen than the control-transfected CL-SL 4 weeks after infection. It is noteworthy that the increased infection levels in separate experiments ranged from a 5 to 50 fold and were statistically significant revealing that expression of the A2 transgene increased survival in the visceral organs.
We further confirmed the importance of A2 in these clinical isolates by downmodulation of A2 expression in the VL-SL strain and this was associated with impaired survival in the visceral organs (Fig. S3). Overall, these results represent a validation of the importance of A2 in a natural setting and highlight the impact that alterations of A2 levels can have on the parasite's ability to survive in visceral organs. Nevertheless, virulence remained severely attenuated in CL-VL parasites in which A2 expression was partially restored (Fig. 4 C, D) relative to the VL-SL isolate (See also Fig. 1B, C), demonstrating that factors in addition to A2 are required for full virulence in the visceral organs.
A major question concerning leishmaniasis in Sri Lanka is whether one or different L. donovani strains are responsible for cutaneous and visceral leishmaniasis. We provide compelling evidence here including fulfilment of Koch's postulates that different strains of L. donovani MON-37 are responsible for visceral and cutaneous disease in Sri Lanka. Most strikingly, the CL-SL clinical isolate was severely attenuated for survival in visceral organs in experimentally infected BALB/c mice, yet acquired the ability to cause cutaneous leishmaniasis in humans. Atypical cutaneous leishmaniasis in Sri Lanka caused by L. donovani is therefore most likely due to parasite-specific rather than host-specific determinants.
The L. donovani VL-SL and CL-SL clinical isolates examined here represent unique strains to investigate genetic determinants affecting disease tropism. There were no homozygous gene deletions and only one pseudogene (containing an additional 3 aa at the C-terminal) specific to the CL-SL strain that were wildtype in the VL-SL strain. SNPs and protein level variations are most likely responsible for disease tropism and pathology differences, and the Rag C and A2 genes identified within are examples of genes which may contribute to the different pathologies caused by the CL-SL and VL-SL strains. It will be of considerable interest to characterize additional CL-SL isolates and determine how widespread changes in these and other genes are on a larger population of parasites from cutaneous leishmaniasis patients in Sri Lanka.
Genome sequencing (Table 1, Fig. 4A) and transcription analysis (Table 4) provided evidence that A2 expression levels were higher in the VL-SL isolate relative to the CL-SL isolate and this was confirmed by Western blot (Fig. 4B). A2 is a multigene family present in L. donovani and L. infantum but found as pseudogenes in Old World cutaneous leishmaniasis species such as L. major and L. tropica [22]. Although A2 is present in New World cutaneous L. mexicana [3], [25], its size and sequence differ from Old World A2, and therefore may serve a different function. In L. donovani, A2 is required for survival in visceral organs and protects L. donovani against high temperature (fever) and oxidative stress [23], [24]. As demonstrated within, the expression of a single additional copy of the A2 gene in the CL isolate resulted in increased survival in the spleen. These observations argue that the L. donovani CL-SL strain has lost the threshold level of A2 expression necessary for survival in visceral organs and that this represents a major determinant of its attenuation.
It was of interest to identify a non-synonymous SNP in the CL-SL isolate in the Rag C gene which in higher eukaryotic cells is involved in the regulation of the mTOR pathway. There are 3 TOR genes in Leishmania which are all essential for Leishmania survival in macrophages or infected mice [20]. Although we did not identify SNPs in the TOR genes, this pathway could be compromised by polymorphism in the Rag C gene. Expression of the VL-SL Rag C gene in the CL-SL isolate significantly increased the ability of the CL-SL isolate to survive in visceral organs (Fig. 2) suggesting that this pathway plays an important role for visceral disease. It has also been demonstrated that targeting the TOR pathway with inhibitors represents a novel opportunity to treat trypanosomatid infections [26]. The observations made here further support this pathway as a drug target for visceral leishmaniasis.
Not only is the L. donovani CL-SL strain attenuated for survival in visceral organs, but it has also gained the ability to cause thousands of cutaneous leishmaniasis cases in Sri Lanka. It will be important to identify any gain of function enabling the CL-SL strain to cause cutaneous disease when other naturally-occurring strains of L. donovani are unable to do. This may require a better biological readout for the CL-SL L. donovani isolate and a better understanding of proteins currently classified as hypothetical.
One of the major observations from recent studies sequencing several different species of Leishmania is that although there are relatively few species-specific genes, there is considerably variation in individual gene and chromosome copy numbers [11], [12], [21]. We observed strain-specific differences in the somy of chromosomes 13 and 20 which were trisomic in the VL-SL strain and disomic in the CL-SL strain (Fig. 3A). This lead to an increased level of transcripts from chromosome 20 in the VL-SL isolate relative to the CL-SL isolate though the transcript levels from chromosome 13 are similar in the CL-SL and VL-SL strains (Fig. 3B). This suggests that certain genes encoded on chromosome 20 may play a role in VL-SL pathogenesis. Increased copy number for chromosome 20 would allow for simultaneous amplification and overexpression of these genes. Chromosome 20 encodes 125 hypothetical proteins. Non-hypothetical proteins of interest include glycosomal phosphoglycerate kinase C, one of the five genes with greater than four-fold differences in transcript levels compared to the CL-SL isolate, cytosolic phosphoglycerate kinase B, glycerol-3-phosphate dehydrogenase-like protein, eight calpain-like cysteine peptidases, two DNAj-like protein chaperones, a member of the glutaredoxin antioxidant family, five kinases and three phosphatase subunits. Calpains may play a role in resistance to miltefosine and calpain inhibitors arrest Leishmania growth [27].
Overall, this work provides valuable insight into the pathogenesis of visceral leishmaniasis. In particular, gene deletions or pseudogene formation do not appear to be required for L, donovani to lose the ability to survive in visceral organs and cause cutaneous disease. The ability of Leishmania parasites to cause visceral or cutaneous leishmaniasis may be determined by sequence polymorphisms or amplification of a few genes. This contributes to understanding L. donovani virulence and may help to identify intervention targets required for visceral organ infection by this important parasite.
Isolation of L. donovani from the patients for this study was approved by the ethics review committee of the faculty of Medical Sciences, University of Sri Jayewardenepura, Sri Lanka (approval no 482/09). Written informed consent was obtained prior to parasite isolation from the two adult patients.
All BALB/c mouse infections (permit # 7395) were approved by the McGill University Animal Care Committee following guidelines from the Canadian Council on Animal Care (CCAC).
L. donovani parasites were isolated from two Sri Lankan patients from the Vavuniya district as previously shown geographically [8]. The VL-SL clinical isolate was from a bone marrow aspirate from a 53-year-old male with chronic fever, hepatosplenomegaly, low haemoglobin and seropositivity for rK39 anti-Leishmania antibodies as previously described [8]. The CL-SL clinical isolate was derived from a skin lesion on the nose from a 28 year old male. Biopsy samples were directly inoculated into Leishmania promastigote culture medium [10], and genomic DNA for genome sequencing was prepared within one month from these promastigote cultures. For axenic amastigote culture, Leishmania promastigotes were shifted to 37°C, pH 5.5 culture media to mimic the macrophage phagolysosome environment. Transfections were performed as previously described with the pKSneo-control plasmid, the pKSneo-A2 plasmid or the pKSneo-A2 (R) antisense plasmid [9]. Transfected parasites were maintained in media supplemented with 200 µg/mL G418 (Wisent).
Female BALB/c mice weighing 17–20 g were purchased from Charles River Breeding Laboratories and maintained in the animal care facility under pathogen-free conditions. BALB/c mice were infected by tail vein injection of stationary-phase promastigotes. Four weeks post infection, amastigotes were isolated from the liver and spleen, and the Leishmania parasite number determined by limiting dilution. The number of stationary promastigotes injected is indicated for each experiment in the Figure legends. Liver parasite burdens were also measured by counting the number of amastigotes in Giemsa-stained liver imprints, expressed as Leishman-Donovan Units (LDU): number of amastigotes per 1000 cell nuclei×liver weight [9]. For cutaneous infections, mice were infected subcutaneously with stationary-phase promastigotes in the hind footpads. Disease progression was monitored by weekly caliper measurement of footpad swelling.
Genomic DNA was sheared into fragments of 100–1,200 bp (Nebulizer, Illumina) and was made into a paired-end DNA sequencing (DNA-seq) library using the Genomic DNA Sample Preparation Kit (Illumina, San Diego, USA). Libraries were sequenced using the Genome Analyzer IIx (Illumina) at the High Throughput Genomics Unit at the University of Washington. Reads were aligned against the reference genome (L. donovani BPK282/Ocl4, cloned line from Nepal). Somy and copy number information for each chromosome were calculated independently using custom written perl script entitled “find_copy_number.pl” (see supplementary methods, Text S1). Single nucleotide polymorphism and small indels were called by inputting alignment files (bam) from all the four libraries into GATK software [28]. A thorough manual inspection revealed around 30% of variant calls were false positives or incorrectly genotyped. Calls were then validated manually and reassigned genotypes if necessary. All the validated variants that were consistent within each group (VL-SL and CL-SL) but different between them were then analyzed in detail to study the effects on protein coding genes using SNP EFF(v3.2) tool [29]. Further details concerning genomic library preparation and sequencing are presented in Supplementary Methods (Text S1).
Promastigote total RNA was extracted from late log phase promastigotes culture. Axenic amastigote total RNA was extracted from axenic amastigote cultures 22 hours after shifting the cells to amastigote culture conditions. B10R macrophages were infected with CL-SL or VL-SL amastigotes at an amastigote/macrophage ratio of 10/1 in suspension in DMEM medium for 12 hours. The percentage of infected macrophages was 72% for CL-SL and 95% for VL-SL; the average amastigotes per infected B10R cell was 3.6 for CL-SL and 3.3 for VL-SL. Infected macrophages were washed and centrifuged to remove extracellular amastigotes and directly suspended in Trizol reagent for total RNA extraction.
SL RNA-seq libraries were prepared as previously described [30]. The SL RNA libraries were sequenced using the Genome Analyzer IIx (Illumina) at the High Throughput Genomics Unit at the University of Washington to generate 36-nt long single-end reads. Details of the mapping of high-throughput sequencing reads can be found in the supplementary methods (Text S1). The SL RNA-seq data have been submitted to the GEO database under accession no. GSE48475.
PCR cloning of VL-SL genes, expression vector constructions and transfection were performed as previously described [10]. To confirm that the corresponding VL-SL genes were expressed in transfected L. donovani CL-SL cells, the reverse transcription PCR (RT-PCR) reactions were performed with QIAGEN OneStep RT-PCR Kit. One primer was specific for the transfected VL-SL gene and other primer was specific to the expression vector sequence contained in the 3′ untranslated region (See supplementary Table S4 for primer sequences). 1 µg of total RNA extracted from these transfected CL-SL cells and pretreated with DNase I was used in each RT-PCR reaction. SDS-PAGE and Western blotting to detect A2 was performed as previously described [23]. A2 was detected with a 1∶8,000 dilution of mouse monoclonal anti-A2 antibody and tubulin with a 1∶2,000 dilution of mouse monoclonal anti-tubulin antibody (Oncogene).
All information was deposited in GenBank under bioproject ID PRJNA210295. DNA sequencing data can be accessed from the SRA database using accession no. SRS484822 and SRS484824. RNA-seq data has been deposited in the GEO database using accession no SRP026479. Open source software tools we used are referenced from methods section. The locally developed scripts are publicly available to download from github project ‘bifxscripts’ under GPLicence (https://github.com/bifxcore/bifxscripts/tree/master/Ldonovani_tropism)
|
10.1371/journal.pgen.1004839 | Topoisomerase I Plays a Critical Role in Suppressing Genome Instability at a Highly Transcribed G-Quadruplex-Forming Sequence | G-quadruplex or G4 DNA is a non-B secondary DNA structure that comprises a stacked array of guanine-quartets. Cellular processes such as transcription and replication can be hindered by unresolved DNA secondary structures potentially endangering genome maintenance. As G4-forming sequences are highly frequent throughout eukaryotic genomes, it is important to define what factors contribute to a G4 motif becoming a hotspot of genome instability. Using a genetic assay in Saccharomyces cerevisiae, we previously demonstrated that a potential G4-forming sequence derived from a guanine-run containing immunoglobulin switch Mu (Sμ) region becomes highly unstable when actively transcribed. Here we describe assays designed to survey spontaneous genome rearrangements initiated at the Sμ sequence in the context of large genomic areas. We demonstrate that, in the absence of Top1, a G4 DNA-forming sequence becomes a strong hotspot of gross chromosomal rearrangements and loss of heterozygosity associated with mitotic recombination within the ∼20 kb or ∼100 kb regions of yeast chromosome V or III, respectively. Transcription confers a critical strand bias since genome rearrangements at the G4-forming Sμ are elevated only when the guanine-runs are located on the non-transcribed strand. The direction of replication and transcription, when in a head-on orientation, further contribute to the elevated genome instability at a potential G4 DNA-forming sequence. The implications of our identification of Top1 as a critical factor in suppression of instability associated with potential G4 DNA-forming sequences are discussed.
| Genome instability is not evenly distributed, but rather is highly elevated at certain genomic loci containing DNA sequences that can fold into non-canonical secondary structures. The four-stranded G-quadruplex or G4 DNA is one such DNA structure capable of instigating transcription and/or replication obstruction and subsequent genome instability. In this study, we used a reporter system to quantitatively measure the level of genome instability occurring at a G4 DNA motif integrated into the yeast genome. We showed that the disruption of Topoisomerase I function significantly elevated various types of genome instability at the highly transcribed G4 motif generating loss of heterozygosity and copy number alterations (deletions and duplications), both of which are frequently observed in cancer genomes.
| In addition to the canonical Watson-Crick helical duplex (B-DNA), genomic DNA, especially repetitive sequences, can assume other types of structures such hairpins, Z-DNA, triplex DNA (H-DNA) or tetrahelical DNA structures [1]–[6]. Impediments to normal DNA metabolic processes including transcription and replication imposed by such secondary DNA structures explain the correlation between repetitive sequence elements and elevated genome instability. Genomic instability at purine-rich GAA•TTC repeats and CAG•CTG repeats, which can fold into three-stranded H-DNA [7] and a slipped hairpin structure [8], forms the molecular basis of multiple neurodegenerative diseases, such as Freidreich's Ataxia and Huntington's disease, respectively.
G-quadruplex or G4 DNA is another non-B secondary DNA structure that potentially interferes with normal DNA transactions [5], [9]–[11]. G4 DNA contains a stacked array of multiple G-quartets, which are comprised of four guanines interacting in a planar configuration [5], [10]. G4 DNA can be readily formed in solution by oligonucleotides containing multiple runs of guanines and by actively transcribed plasmid DNA [12], [13]. G4 DNA-forming sequences or G4 motifs are present in the genomes of diverse organisms and conserved throughout evolution; they number >375,000 in the human genome and >1,400 in the Saccharomyces cerevisiae nuclear genome [14]–[17]. The distribution of G4 motifs is highly concentrated at telomeres, rDNA loci, immunoglobulin heavy-chain switch regions, and G-rich minisatellites and significantly correlates with nucleosome-free regions and transcription start sites (TSSs) [10], [11], [18]. In oncogenes, G4 motifs are mostly enriched in the regions flanking TSSs, which suggests that G4 DNA may be involved in transcriptional regulation [15], [17].
G4 DNA becomes a structural barrier to transcription and replication in vitro indicating that it might play a significant role in genome instability [19]–[21]. In the absence of Pif1, a potent G4 DNA unwinding helicase, replication forks slow down near G4 motifs present in the yeast genome strengthening the argument that an unresolved G4 structure can lead to increased genome instability [22]. G4 motifs are frequently found at unstable genomic loci including proto-oncogenes and sites of frequent translocation breakpoints [23] and at preferred mitotic and meiotic DNA break sites [17]. In some human cancers, G4 motifs have been identified at frequent breakpoints involved in chromosomal translocations including the major breakpoint region in the proto-oncogene BCL2 [24]. Chromosomal translocations involving G-rich immunoglobulin switch regions have long been observed in various cancer cell lines [25]. The identification in silico of potential G4 DNA-forming sequences at sites of genome instability, however, has not yet been fully verified by in vivo demonstration of biological relevance of G4 DNA structure.
Seminal advances in understanding the genome instability induced by the repetitive, DNA secondary structure-forming sequences have been made using yeast and bacterial model systems [26]–[28]. A large tract of GAA•TTC repeats in yeast, for example, acted as a hotspot of gross chromosomal rearrangements (GCRs) and interstitial deletions [26], [27]. In bacteria, CTG•CAG repeats from the Myotonic Dystrophy gene induced large deletions when the repeats were highly transcribed [29]. When a guanine-run containing a human subtelomeric minisatellite was integrated into the yeast genome, it significantly elevated GCRs [30] and resulted in frequent repeat expansion and contraction [31]. The guanine-rich yeast telomere repeats, when placed within an intron of an interstitially located gene, cause various types of chromosomal rearrangements including deletions and inversions [32]. Accumulating evidence pointing to G4 motifs as genome instability hotspots underscores the importance of defining endogenous and exogenous factors that influence the integrity of genomic loci containing these motifs.
Active transcription, when oriented in the direction to place the guanine-runs in the transiently single stranded non-transcribed strand (NTS), was shown to stimulate formation of G4 DNA structure both in vitro and in bacterial cells [13]. To determine the effect of G4 DNA on yeast genome stability, we previously constructed a genomic reporter assay where a potential G4 DNA-forming sequence was highly transcribed to promote secondary DNA structure formation. By normalizing the extent of genome instability occurring at this reporter construct to that occurring at the exact same sequence transcribed in inverse orientation (that is, with the G4 motifs on the transcribed strand), we were able to apply a stringent control for the correlation between elevated genome instability and G4 DNA. Using this approach, we found that gene conversion recombination was significantly elevated by highly transcribing guanine-run containing sequence in a strictly strand-specific manner [33].
In the current report, we demonstrate that active transcription transforms a guanine-run containing sequence into a strong hotspot for gross chromosomal rearrangements and loss-of-heterozygosity (LOH). Our data also show that the direction of replication can significantly alter the level of instability at a potential G4 DNA-forming sequence suggesting that genomic context in terms of both transcription and replication is important when considering G4 motifs as potential genome instability hotspots. Finally, we identify a critical role of Topoisomerase I (Top1) in suppressing various types of genome rearrangements associated with co-transcriptionally formed G4 DNA.
We previously showed that gene conversion resulting from ectopic recombination is increased due to co-transcriptionally formed G4 DNA. In order to determine whether co-transcriptionally formed G4 DNA can also elevate gross chromosomal rearrangements (GCRs), we modified the GCR reporter system previously described by Chen and Kolodner [34], [35]. In this reporter system, the URA3 gene was integrated into the left arm of chromosome V (CHR5) replacing the HXT13 gene located ∼8.5 kb centromere-distal to the CAN1 gene (Fig. 1A). The loss of functional CAN1 or URA3 results in resistance to the drug canavanine (Can) or 5-Fluoroorotic acid (5-FOA), respectively. Because the first essential gene on the left arm of CHR5, PCM1, is located ∼60 kb from the telomere, the region containing CAN1 and URA3 genes can be lost without affecting viability of haploids. The hypothetical rate of double drug resistance (CanR/5-FOAR) occurring via independent mutations in CAN1 and URA3 is approximately 10−12, which is significantly lower than the observed rate of deletion of the left arm of CHR5 (10−11–10−10 in a wild-type background) [35]. Therefore, by selecting for colonies resistant to both Can and 5-FOA, GCR events resulting in simultaneous loss of CAN1 and URA3 genes are detected.
We modified the GCR assay by integrating the pTET-lys2-GTOP or -GBTM cassette immediately centromere-proximal to CAN1 (Fig. 1A). As previously described [36], the pTET-LYS2 cassette contains the LYS2 gene transcribed from the heterologous tetracycline-repressible promoter (pTET) linked to a marker gene conferring G418 resistance (G418R). Within the LYS2 ORF, at ∼390 bp from the start codon, a 760 bp fragment from the mouse immunoglobulin (Ig) switch Mu (Sμ) region was inserted to generate pTET-lys2-GTOP or -GBTM cassettes [33]. Switch regions, which are required for Ig heavy chain class-switch recombination (CSR), are comprised of multiple, degenerate guanine-rich repeats of several kb in length [37]. The Sμ sequence is a model G4 motif and was previously demonstrated to form G4 DNA structures both in vivo and in vitro when highly transcribed [13]. The sequence of the Sμ fragment incorporated into the LYS2 ORF (containing ∼17 of (GAGCT)nGGGGT repeats) is shown in Figure S1. This fragment was inserted either in the physiological (-GTOP) or in the inverted orientation (-GBTM), placing the guanine runs on the non-transcribed (NTS) or the transcribed strand (TS), respectively (Fig. 1B) [33].
Topoisomerase 1 (Top1) is a highly conserved enzyme that relieves positive or negative torsional stress associated with transcription and replication [38], [39]. Top1 functions by covalently attaching to the 3′ end of nicked DNA, which is quickly re-ligated after swiveling of the DNA strands to remove supercoiling. Although not essential for viability in yeast, replication slows down and sometimes stalls in the absence of Top1, especially at highly transcribed regions [40]. We previously reported that disruption of Top1 leads to an increase in Sμ-induced gene conversion events [33].
In order to determine whether Top1 also plays a role in preventing GCRs initiating at G4 DNA, we deleted the TOP1 gene from strains containing pTET-lys2-GTOP and –GBTM constructs. In wild-type (WT) backgrounds under high-transcription conditions, the rates of CanR/5-FOAR events were 0.88×10−10 and 0.20×10−10 for pTET-lys2-GTOP and –GBTM constructs, respectively (Fig. 1C). These rates are not significantly different from each other and are comparable to GCR rates previously reported in the absence of an inserted pTET-lys2 cassette [34], [35]. Upon TOP1 deletion, the rate of GCR (CanR/5-FOAR) was significantly elevated for both the pTET-lys2-GTOP and -GBTM construct. Importantly, GCR occurred at a significantly higher (∼30-fold) rate for pTET-lys2–GTOP, where guanine-runs are present on the NTS, compared to the pTET-lys2–GBTM construct where guanine-runs are on the TS.
We tested whether active transcription of the guanine-run-containing sequence is required for the elevated GCRs by growing the top1Δ strains in medium containing doxycycline, an analog of tetracycline, which resulted in ∼60- to 200-fold reductions in the transcription rates (Table S1). Repression of transcription from the pTET promoter by doxycycline resulted in a >200-fold decrease in the rate of CanR/5-FOAR for the strain associated with the pTET-lys2-GTOP construct (Fig. 1C). For the strain containing the pTET-lys2-GBTM construct, transcriptional repression also led to a significant decrease in the rate of CanR/5-FOAR.
In order to determine whether the GCRs in top1Δ backgrounds initiate at the G4 DNA containing reporter construct, we carried out PCR analysis to map the GCR initiating breakpoints. For the strain containing the pTET-lys2-GTOP construct in the top1Δ background, 27 of 30 CanR/5-FOAR isolates tested had lost the portion of LYS2 gene with the Sμ fragment insertion but still retained the G418R cassette located just centromere-proximal to this region (Fig. 1A and Table 1). Thus, 90% of the GCR events occurring in this strain initiated within the ∼4 kb region comprised of the pTET-lys2-GTOP cassette between G418R cassette and CAN1. This is proportionally greater than the GCR initiating in the same region for the pTET-lys2-GBTM construct in top1Δ strain (9 out 23; P<0.0005 by chi square analysis).
To further characterize the chromosome breakpoints in GCR events associated with the pTET-lys2-GTOP cassette, we carried out PCR with a degenerate primer annealing to generic yeast telomere sequence (CA16; 5′ CACCACACCCACACAC 3′) and a primer annealing to the 5′ untranslated region of the LYS2 gene. Out of 27 samples where the disruption of the pTET-lys2-GTOP cassette was confirmed by PCR mapping, telomere-anchored PCR products of 700 to 1900 bp were obtained for 16 samples. Subsequent sequencing of these fragments showed that, in 15 CanR/5-FOAR clones, de novo telomere additions occurred at various locations within the G4-forming Sμ fragment (Class I events in Fig. 2 and Fig. S1). Due to the high G/C content and repetitiveness of the Sμ sequence, sequencing analysis failed to identify the site of telomere addition in one of the PCR fragments. In concurrence with preferential telomere addition sites previously identified [41], the junctions of de novo telomere addition were located at GT dinucleotides and frequently at 5 to 6 nt clusters of GT-rich sequence. Additionally, mutations, deletions/insertions, and duplications within the Sμ fragment were detected in seven of the 15 CanR/5-FOAR clones with telomere additions.
We further characterized the genome rearrangements associated with G4 DNA using pulse field gel electrophoresis (PFGE) and microarray-based comparative genome hybridization (array-CGH). PFGE showed that, in the 16 samples where de novo telomere addition at Sμ was confirmed by sequencing (Class I), CHR5 was reduced in size by ∼35 kb and co-migrated with CHR8 (Fig. 2A and Fig. S2A). The loss of CHR5 sequences from the left telomere to the integration site of pTET-lys2 cassette (at 34,000 NT) was confirmed by array-CGH analysis (Fig. 2B and Fig. S2A).
In another 8 samples (Class II) where CHR5 appeared smaller by only ∼15 kb (Fig. 2A and Fig. S2B), array-CGH identified segmental deletions between the original location of HXT13 (at 23,000 NT) and the pTET-lys2 cassette. PCR and sequencing analysis showed that this recurrent deletion was mediated by a pair of 21 bp direct repeats introduced into the two respective sites of CHR5 as parts of plasmid constructs used for integration of the URA3Kl marker at HXT13 locus (pUG72; Euroscarf) and pTET promoter (pCM225; Euroscarf) (Fig. S3).
Finally, in class III events, deletion of CHR5 sequences from the left telomere to the pTET-lys2 integration site occurred in combination with duplications of the immediate proximal sequences on the left arm, and duplication of a terminal segment on the right arm (Fig. 2B and Fig. S2C–D). The two clones from class IIIa showed duplications from pTET-lys2 to the Watson-oriented YELWdelta1 dispersed long terminal repeat element (LTR), and the one example of class IIIb showed a duplication extending further to the full length Crick-oriented Ty1 element insertion at the URA3 locus (ura3-52 allele). All three class III clones had duplications of a segment of the right arm extending from position ∼446,000 NT (containing the full length Crick-oriented YERCTy1-1 element and the Watson-oriented YERWdelta20b LTR) all the way to the right telomere. These clones displayed longer versions of CHR5 of ∼700 kb (migrating just below CHR10) and ∼770 kb (co-migrating with CHR2 and CHR14) for class IIIa and class IIIb, respectively (Fig. 2A and Fig. S2C–D).
The class III chromosome sizes were consistent with the deletions and duplications detected by array-CGH and suggested a complex mechanism of formation. Similar array-CGH patterns were observed recently in the analyses of GCR events in yeast CHR5 as well as in humans [42], [43]. These studies described breakpoint structures consistent with an intra-strand fold-back mechanism in which a resected free 3′ end folds back on itself, re-anneals to a microhomology region and primes break-induced DNA replication (BIR). By plasmid-rescuing this regions subcloning and sequence analysis (See Materials and Methods), we confirmed that the duplicated regions proximal to the Sμ breakpoints in class IIIb clone A7 and class IIIa clone A8 were comprised of inverted duplications separated by single copy regions corresponding to the original ssDNA loops, as predicted by the fold-back mechanism (Figs. S4A and S4B). We were not successful in rescuing the duplicated regions from the Class IIIa clone A16. The recovered rearrangement structures observed in the A7 and A8 clones can be explained by two different models (Fig. S4C). In the first scenario, as the BIR event initiated by the intra-strand fold-back reached the Ty/LTR sequences on the left arm, it collapsed, re-annealed at the Ty/LTR sequences on the right arm (template switching), and continued on to reach the right telomere to produce a stable monocentric chromosome. The second possibility is that BIR continued all the way to the right telomere forming an unstable dicentric chromosome, which was then stabilized by a secondary homologous recombination event between Ty/LTR repeats leading to loss of one of the centromeres. Although our data does not allow us to distinguish between these two possibilities, we favor the template switching model (Model 1 in Fig. S4C) since BIR is generally thought to be impeded by centromeric structures, and BIR template switching has been shown to be frequent in yeast [44], [45].
Spontaneous DNA breaks in diploid cells are frequently repaired by allelic mitotic recombination using as template either a sister chromatid or a homologous chromosome. We designed an assay that can measure G4 DNA-induced mitotic recombination between chromosome III (CHR3) homologs in diploids (Fig. 3A). First, we integrated the URA3 gene near the telomere of the left arm of CHR3 in a haploid strain derived from YPH45 (a S288c derivative). On the same arm of CHR3, about 44 kb centromere-proximal to the URA3 integration site, pTET-lys2-GTOP or -GBTM was integrated replacing HIS4. As described above for the CHR5 GCR assay, the pTET-lys2-GTOP and -GBTM cassettes contained the 760-bp fragment of Sμ sequence and were adjacent to an aminoglycoside phosphotransferase gene conferring resistance to the drug G418 (G418R). Because the direction of replication fork movement relative to the direction of transcription can affect recombination at highly transcribed regions [46], [47], each cassette was integrated in two orientations relative to the nearby replication origin ARS306. This yielded constructs in which the transcription and replication forks are co-directional (SAME) or in head-on orientation (OPPO) (Fig. 3B). The direction of replication fork movement through this region of CHR3 was previously confirmed by 2D-gel analysis [36].
Heterozygous diploids were generated by mating the YPH45-derived haploid strains described above to a haploid strain derived from YJM789, a clinically isolated strain with ∼0.5% sequence divergence relative to the S288c reference strain [48]. Because the YJM789-derived strain is Ura−, loss or mutation of the URA3 gene on the CHR3 from the YPH45 parent will result in resistance to 5-FOA in the heterozygous diploid cells. The types of genome rearrangements that can lead to loss of URA3 include (a) complete loss of YPH45 CHR3, (b) partial loss of the left arm of YPH45 CHR3, (c) Break Induced Replication (BIR) or reciprocal crossover (RCO) initiating between URA3 and CEN3 or (d) translocation/BIR events involving a heterologous chromosome. In RCO, the distal ends of the two CHR3s are exchanged without loss of genetic material. In BIR involving the homolog, the break in YPH45-CHR3 will be repaired through replication using YJM789-CHR3 from the break to telomere as template. In our assay, we cannot distinguish between these two mechanisms (Fig. S5). Because all of these events result in loss of heterozygosity (LOH) for the segment of CHR3 containing the URA3 marker, we hereafter refer to this assay as the LOH assay.
In order to map the position of LOH in 5-FOAR isolates, we devised a PCR-based restriction fragment length polymorphism (RFLP-SNP) assay described in Fig. S6. We defined the initiation of recombination point as between the last telomeric SNP site displaying LOH and the first centromeric SNP site displaying heterozygosity (Table 2). A 5-FOAR isolate was defined as resulting from recombination initiating at or near the G4 repeats (G418-pTET-lys2-GTOP or GBTM), when it was homozygous for the centromere-distal SpeI site and heterozygous for the centromere-proximal NarI site and the G418S cassette was not present (LOH class E in Table 2 and Figure 3A). When mitotic recombination is initiated by DNA breaks near the pTET-lys2-GTOP or –GBTM cassette, resection must extend into the region of the YPH45 CHR3 with homology to YJM789 CHR3 and, therefore, remove the pTET-lys2-GTOP or –GBTM cassette along with the G418R marker.
To determine whether highly transcribed G4 motifs elevate LOH on CHR3, we measured the rate of 5-FOAR in WT strains containing pTET-LYS2 (no Sμ sequence) or pTET-lys2-GTOP cassette. In either the SAME or OPPO orientations, the overall rate of LOH (5-FOAR) associated with pTET-LYS2 or pTET-lys2-GTOP was not significantly different (Fig. 4A). Using the RFLP-SNP assay, we identified 3/46 or 9/46 LOH events initiated at the highly transcribed pTET-LYS2 cassette when in the SAME or the OPPO orientation, respectively (Table 3). These proportions were not statistically different from those for LOH initiating at pTET-lys2-GTOP in the SAME or OPPO orientation (by Fisher's exact test; P = 0.31 and 0.39, respectively).
We observed a dramatic and specific increase in the rates of gene conversion [33] and GCR (see above) associated with highly transcribed Sμ sequences when Top1 was disrupted in a haploid YPH45 background. Importantly, both occurred at significantly higher rates when the guanine-run containing strand was on the NTS where its single stranded nature fosters G4 DNA formation. In order to determine whether LOH on CHR3 is similarly affected by the location of G4-forming sequence on the TS vs. NTS, we compared LOH rates associated with pTET-lys2-GTOP and -BTM constructs in top1Δ/top1Δ backgrounds. There was no significant difference between overall rates of LOH events associated with pTET-lys2-GTOP and -GBTM constructs when replication was in the SAME direction (Fig. 5A). However, when replication was in the OPPO orientation, the overall rate of LOH events was ∼3 fold higher for the pTET-lys2-GTOP than for the pTET-lys2-GBTM construct. The rates of LOH initiating at the pTET-lys2-GTOP/GBTM cassette in the SAME or OPPO orientation in top1Δ/top1Δ background were determined by analyzing 47–93 5-FOAR isolates by the RFLP-SNP assay (Table 4). When transcription from the pTET promoter was in the SAME orientation relative to replication originating at ARS306, the rate of LOH initiating at the G4-containing sequence was similar whether the guanine-runs were on the NTS (pTET-lys2-GTOP) or on the TS (pTET-lys2-GBTM) (Fig. 5B). However, the rate of LOH initiated near pTET-lys2-GTOP in the OPPO orientation was >20 fold higher than at pTET-lys2-GBTM in OPPO orientation and ∼4 fold higher than at pTET-lys2-GTOP in the SAME orientation.
Upon Top1-disruption, when the replication and transcription is in “head-on” or OPPO orientation, the rate of loss of heterozygous SNP at NarI site (at 78380 NT) is also significantly elevated (Table 4). The NarI-SNP is not lost at a high rate in pTET-lys2-GBTM-containing strain, which suggests that recombination initiating at the pTET-lys2-GTOP cassette (at 68300 NT) are often associated with long conversion tracks resulting in the loss of NarI-SNP. This is consistent with the average length of conversion tracks associated with reciprocal crossover events, which is reported to be about 12 kb [49].
Using the RFLP-SNP assay described above, we characterized 93 5-FOAR isolates from the top1Δ/top1Δ strain containing the pTET-lys2-GTOP cassette in the OPPO orientation and identified 33 isolates with LOH events initiating near the G4 motif-containing reporter construct (Table 4 – LOH Class E). Such LOH events can occur by (1) mitotic recombination with the other CHR3 in the heterozygous diploid cell via reciprocal crossover (RCO) or break-induced replication (BIR) with the homolog, (2) the partial loss of the chromosome arm, or (3) translocation to another chromosome with a short stretch of homology (Figure 3A). In order to determine the types of rearrangements occurring at this locus, separation of the CHR3 homologs by PFGE was carried out for the 33 LOH Class E isolates (Figure S7). The CHR3 homologs of YPH45 and YJM789 differ in size by about 56 kb, likely reflecting different polymorphic subtelomeric gene content and retrotransposon insertions. In 32 out of 33 5-FOAR isolates analyzed, the YJM789-derived CHR3 was unchanged in size and the YPH45-derived CHR3 appeared slightly smaller than that of the parental haploid. A smaller CHR3 can be generated by RCO or BIR initiating on YPH45-derived CHR3 using YJM-789 derived CHR3 as the repair donor. The reduction in chromosome size was approximately 13 kb in 31/33 5-FOAR isolates (Fig. S7). In one of 33 isolates analyzed, YPH45-CHR3 was reduced in size by ∼24 kb. An ∼8.5 kb reduction was expected from the loss of hemizygous URA3 maker and the pTET-lys2-GTOP cassette with additional reduction in size resulting from the loss of other hemizygous sequences. Translocation to a heterologous chromosome was observed in 1/33 5-FOAR isolates. None of the analyzed isolates contained CHR3 shortened by ∼80 kb, which is predicted in case of the loss of left arm from the pTET-lys2-GTOP cassette to the telomere followed by a telomere addition at the break site (Figure 3A, bottom panel).
In our previous report regarding the rate of gene conversion events induced by the highly transcribed Sμ sequence, the pTET-lys2-GTOP cassette was integrated in the orientation and location identical to the “OPPO” construct described above for the LOH assay (Fig. 6A). In this orientation, transcription from pTET promoter and replication originating at ARS306 are in the convergent or “head-on” orientation. In order to determine whether the rate of gene conversion is dependent on the relative orientation of transcription and replication, we deleted ARS306 by replacing the ARS consensus sequence (5′-WTTTAYRTTTW-3′) [50] with the gene encoding hygromycin B phosphotransferase (Hph). In the resulting ars306Δ strain, replication through the pTET-lys2-GTOP cassette originates from the ARS305 located about 27 kb away and is in co-directional orientation relative to transcription [51]. As previously reported, in the gene conversion assay, recombination initiating at the pTET-lys2 cassette can be completed using a truncated lys2 gene fragment integrated on CHR15 resulting in lysine prototrophy (Lys+) (Fig. 6A) [33]. In a top1Δ strain containing the pTET-lys2-GTOP cassette, reversing the replication orientation by deletion of ARS306 resulted in a three-fold decrease in the Lys+ rate indicating that replication-transcription conflict is a factor in elevated gene conversion initiating at co-transcriptionally formed G4 DNA (Fig. 6B). The deletion of ARS306 did not significantly affect the gene conversion rates at pTET-lys2-GBTM in top1Δ background or at pTET-lys2-GTOP or –GBTM cassette in WT backgrounds.
G4 motifs have been implicated in various types of genome instability events. However, the large number of sequences predicted form G4 DNA structures have not all been validated as potential hotspots of genome rearrangements. We here focused on the level of transcription as a singularly important genomic context that distinguishes genetically unstable G4-forming sequence. At the Ig heavy chain locus, class-switch recombination (CSR) in activated B cells requires switch regions consisting of long repetitive sequences dense with guanine-runs [52]. In the pathogenic bacteria Neisseria gonorrhoeae, a G4 DNA-forming sequence was identified to be essential for the gene conversion occurring at the PilE locus, which facilitates evasion of host adaptive immune system by the production of variant pilin subunits [53]. Transcription of the guanine-run containing sequences is required to initiate recombination in both of these processes [54], [55]. One possible role played by transcription is to provide the strand separation necessary for guanine-runs to fold into G4 structures. In order to determine the effect of G4 DNA on yeast genome stability, we designed our genetic assays to assess the role of transcription in biological processes. For effective formation of G4 DNA during transcription, we placed a model G4 motif from the mouse Ig switch Mu region into the highly transcribed pTET-LYS2 cassette. The guanine-runs were placed on non-transcribed strand (-GTOP) to promote co-interaction in single-strand context; as a negative control, genomic instability associated with the same G4 sequence was measured when transcribed in inverse orientation (-GBTM) where guanines on the TS interact with the nascent RNA and are not available for G4 formation (Fig. 1B).
In the gene conversion assay we used previously, only those recombination events initiating specifically at the reporter construct could be phenotypically selected [33]. In the current study, we designed two other genomic assays that allowed us to survey instability initiated over large genomic areas that includes our G4 motif-containing reporter construct. Loss of part of a chromosome arm initiating over a ∼20 kb region of CHR5 in a haploid background (GCR assay) or recombination initiating over a ∼100 kb region of CHR3 in a diploid background (LOH assay) yielded selectable, CanR 5-FOAR or 5-FOAR colonies, respectively (Fig. 1A and 3A). By further analysis of the individual isolates, we were able to estimate the locations of GCR or mitotic recombination initiation. Importantly, this unbiased measurement of spontaneous genome rearrangements enabled us to define the conditions under which GCR and mitotic recombination resulting in LOH are specifically elevated at the highly transcribed G4 motif.
Eukaryotic Top1 has multiple functions during DNA transactions [56]. Both negative and positive supercoils accumulated during transcription are removed by Top1 activity. Together with the type II topoisomerase Top2, Top1 is recruited to the genomic regions undergoing replication [57] and has a role in relieving transcription-replication conflicts [40]. Top1 function can also adversely affect genome stability; its endo-ribonuclease activity generates unligatable single-strand breaks at ribonucleotides embedded in DNA, which leads to replication stress and accumulation of deletion mutations [58], [59]. We have shown here that, at the co-transcriptionally generated G4 DNA, Top1 activity is required to suppress various types of genome instability. Top1 disruption greatly elevated overall GCR rates when the pTET-lys2-GTOP was present on CHR5, with 90% of GCR breakpoints mapping to the G4 DNA-forming sequence (Fig. 1C and Table 1). This elevation was completely dependent on the level of transcription and the location of the guanine-runs on the NTS, reinforcing the conclusion that a potential G4 sequence motif is transformed by transcription into a genome instability hotspot. In the LOH assay, even though the overall mitotic recombination rate was not considerably elevated in the top1Δ/top1Δ background, a significantly higher proportion of the LOH tract breakpoints mapped to the pTET-lys2-GTOP but not to the pTET-lys2-GBTM cassette within the left arm of CHR3 (Fig. 5 and Table 4).
In absence of Top1, the rates of both gene conversion and LOH occurring at the G4-forming sequence were significantly higher when the transcription was in “head-on” or collisional orientation with replication fork movement than when it was in co-directional orientation (Figure 5B and 6B). This suggests that Top1-dependent suppression of G4-associated genome instability involves its activity of resolving transcription-replication conflict in addition to its activity of resolving transcription-associated torsional stress. This also suggests that, in addition to the level of transcription, the relative orientation of replication is an important genomic context that can render certain G4 motifs genetically unstable. Accumulation of stalled replication forks at gene-rich regions have been observed in Top1-deficient cells indicating that one of the ways Top1 prevents genome instability is to prevent replication fork collapse due to collision with transcription [40]. Alternatively, the significantly higher LOH and gene conversion rates observed when the transcription and replication are in “head-on” orientation can be due to the intrinsic asymmetry in the replication process. In “head-on” or co-directional orientation, G-runs are present in the leading strand or lagging strand, respectively. It is possible that, upon encountering G4 DNA, the lagging strand synthesis is less prone to replication arrest since re-priming downstream will allow continued replication fork movement. It was previously reported that replication orientation did not have an effect when the G4-forming human subtelomeric minisatelite CEB1 was placed into the yeast genome [30]. The rates of GCR at this G4 forming sequence were elevated to similar degrees whether the G-runs were on the leading strand or lagging strand. In this experiment, CEB1 was not transcribed, supporting the argument that the orientation bias we observed with the highly transcribed Sμ is due to conflict between replication and transcription.
In cultured mouse B cells, it was reported that class switch recombination (CSR) at Ig heavy chain locus was inhibited by camptothecin (CPT) treatment and significantly elevated by siRNA-mediated knock-down of Top1 [60]. It was suggested that G4 DNA formation is facilitated by reduced Top1 activity and that DNA ligation by Top1, which requires the proper alignment of 3′ and 5′ ends of the breaks, is inhibited by interaction with DNA secondary structures resulting in Top1 cleavage complex and unresolved DNA breaks that initiate CSR. However, we demonstrated here that genome instability associated a G4 motif is stimulated by the complete absence of the Top1 protein (top1Δ) (Fig. 1C, 5B and ref. [33]). DNA breaks initiating gene conversion, LOH or gross chromosomal rearrangements may originate from other sources such as G4-specific nucleases or from collapsed DNA replication forks.
Multiple helicases including human FANCJ, PIF1, BLM and yeast Sgs1 can unwind G4 structures in vitro [61], [62]. In BLM-deficient cells, G4 motifs are frequently found near the transcription start sites of those genes with perturbed expression profile suggesting a role of BLM helicase in G4-mediated gene regulation [63]. Using GCR assays similar to that described above, two independent investigations into the effect of mutation of the 5′-to-3′ DNA helicase Pif1 on the rate of GCR initiated by G4-forming sequences reported dramatically disparate results [30], [61]. Although greatly increased rates of FOAR/CanavanineR colonies were observed in both investigations, further analyses revealed that the 5-FOAR/CanavanineR colonies in yeast cells expressing mutant Pif1 arose mostly via rearrangements or partial loss of the chromosome in one case [30] but mainly via epigenetic silencing of the CAN1 and URA3 genes in another [61]. In case of the G4-associated GCR events occurring due to Top1-deficiency reported here, we demonstrated that the simultaneous resistance to 5-FOA and canavanine resulted from the loss of the region of the CHR5 containing CAN1 and URA3 genes (Table 1). In 60% of the GCR events involving the highly transcribed G4 motif, de novo telomere addition occurred within the guanine-run containing Sμ region (Fig. 2 and Fig. S1). Other types of events included 15 kb segmental deletions and complex genome rearrangements involving terminal deletions and segmental duplications (Fig. 2B. Fig. S2 and Fig. S3). When combined with co-transcriptionally formed G4 DNA, Top1 disruption significantly reshapes the genome not just through elevated non-crossover and allelic interhomolog recombination but also through gross deletions and duplications resulting in copy number variations. This result suggests that, whereas the function of Pif1 and BLM is possibly linked to the role of G4 DNA as an epigenetic and transcriptional regulator [17], [61], [63], Top1 functions directly to prevent chromosomal rearrangements and gross loss of genetic information associated with the G4 DNA, particularly at highly transcribed areas.
Activated transcription through G/C rich sequence can lead to formation of R-loops, which comprise of a long and stable hybrid between nascent RNA and template DNA strand [64]. R-loop accumulation and associated hyper-recombination can ensue when the mRNA packaging and export is disturbed in THO-TREX defective strains or when the degradation of RNA in RNA∶DNA hybrid is deficient due to absence of RNase H activity [65]. Accumulation of negative supercoils in Topoisomerase-deficient cells can also lead to R-loop accumulation [66]. Duquette et al reported that the combination of R-loop and G4 DNA, referred to as G-loop, is identifiable by electron microscopy when the Ig switch sequence is highly transcribed either in vitro or in bacteria [13]. At the pTET-lys2-GTOP cassette containing the Sμ sequence, therefore, the elevated genome instability could be the result of RNA∶DNA hybrid and/or G4 DNA. G-loop formation can be instigated by G4 DNA nucleation in the NTS, which lead to the stable annealing of the nascent RNA with the unpaired TS of DNA (Fig. 7). Alternatively, G-loop formation can initiate via the formation of RNA∶DNA hybrid, which leaves the NTS unpaired and free to fold into G4 structure. In this case, the higher stability of rG∶dC base pairing compared to rC∶dG could account for the greater instability we observed when the G-runs are on the NTS (-GTOP) [67]. During in vitro transcription, rG∶dC containing RNA∶DNA hybrid is critical for the formation of G-loop structure by Ig switch sequence [13], and required for the transcription blockage by a guanine-run [19]. We tested whether G4-induced hyper-recombination is dependent on RNA∶DNA hybrid formation by overexpressing RNase H1 in top1Δ background. RNase H1 is an enzyme that degrades RNA hybridized to DNA and was shown to counteract the hyper-recombination phenotype associated with R-loops in THO/TREX mutant background [65]. As shown in Fig. 7B, RNase H1 overexpression did not reduce the elevated recombination at the highly expressed pTET-lys2-GTOP cassette, which suggests that RNA∶DNA hybrid is not required for the elevated recombination occurring at the G4 motif in absence of Top1. We previously reported that, upon disruption of both RNase H1 and RNase H2 (rnh1Δ rnh2Δ), the rates of gene conversion for the pTET-lys2-GTOP and –GBTM constructs were elevated by 28- and 8- fold, respectively [33]. RNase H1 overexpression led to significant decreases in the rates of gene conversion in rnh1Δ rnh2Δ backgrounds indicating that RNA∶DNA hybrid is responsible for the elevated recombination at both the pTET-lys2-GTOP and –GBTM constructs in rnh1Δ rnh2Δ mutant strains (Fig. 7B). We postulate that RNA∶DNA hybrid and G4 DNA can each result in genome instability but that R-loop is not the primary cause of elevated recombination we observed for the pTET-lys2-GTOP upon disruption of Top1. A function of Top1 other than the prevention of R-loop formation is relevant in suppressing genome instability at G4 DNA, which will require further investigation to identify.
In summary, we report here the identification of Top1 as an important factor in suppressing genome rearrangements instigated by co-transcriptionally formed G4 DNA. In the absence of Top1, LOH-inducing mitotic recombination as well as GCR is highly elevated but only when the guanine-run containing sequence is located on the NTS and is highly transcribed. The exquisitely specific effect of Top1 at G4 DNA is underscored by recent reports that, without co-transcriptionally formed G4 DNA, Top1-disruption had no significant effect on GCR or gene conversion rate and even suppressed the GCR or gene conversion occurring in cells with defects in the ribonucleotide excision repair (RER) pathway [68], [69]. One possible explanation for its functional specificity is the high affinity binding of G4 DNA by Top1 demonstrated in in vitro experiments [70]–[72]. Top1 activity in suppressing G4-assoicated genome instability becomes even more important when transcription is in the collisional orientation with replication. This result suggests that, besides the transcription-conferred strand bias, the genomic location relative to a replication origin might determine which of the numerous G4 motifs so far identified in the eukaryotic genomes might be a hotspot of genome instability. The data presented in this report opens up the possibility that other factors suppressing G4-associated genome instability will be found among proteins with known physical and/or genetic interactions with Top1.
Yeast strains used for the GCR assay and the gene conversion assay were derived from YPH45 (MATa, ura3-52 ade2-101 trp101; [73]). For construction of the GCR assay, procedures for deletion of endogenous LYS2 on chromosome II and insertion of tetR′-SSN6 repressor-expressing cassette at LEU2 locus (pCM244 from Euroscarf [74]) were as described previously for the gene conversion assay [33]. A PCR-amplified LYS2 gene fragment was then integrated upstream (centromere proximal) of the CAN1 ORF on the left arm of chromosome V. The LYS2 promoter was replaced with a PCR-generated cassette from pCM225 (Euroscarf) containing the pTET promoter with 7 repeats of tetO and the tetR-VP16 activator coding sequence. The replacement of LYS2 with either the lys2-GTOP or -GBTM allele was carried out using the two-step allele replacement method. Finally, the loxP-flanked URA3Kl cassette was amplified from pUG72 (Euroscarf; [75]) to replace the HXT13 gene through one-step allele replacement. The construction of pTET-LYS2 and pTET-lys2-GTOP or -GBTM (previously referred to as pTET-lys2-SμF and –SμR) cassettes and the genomic integration on chromosome III in the YPH45 strain background were previously described [33], [36]. For the loss of heterozygosity (LOH) assay, the YPH45-derived haploids were mated to an YJM789-derived haploid (MATα, ura3 lys2; [48]).
The plasmid pGAL1-RNH1 with the yeast RNH1 gene under the galactose-inducible GAL1 promoter was a gift from R. Crouch (NCI; Bethesda, MD).
For the GCR assay, 5 ml cultures in YEPD medium (1% yeast extract, 2% Bacto-peptone, 2% dextrose, and 250 µg/mL adenine hemisulfate, 2% agar for plates) were inoculated with single colonies and grown for 3 days at 30°C. Cells were then plated either on YEPD or synthetic complete dextrose medium lacking arginine (SCD-arg) and containing canavanine (60 mg/L) and 5-Fluoroorotic acid (5-FOA; 1 g/L). For the LOH assay, 1 ml YEPD cultures inoculated with single colonies were grown for 3 days at 30°C and plated on YEPD or SCD containing 5-FOA (1 g/L). For determination of gene conversion rates, growth and plating conditions were the same as previously described [33]. For RNH1 overexpression experiment, indicated yeast strains were transformed with pGAL1-RNH1 plasmid or pRS416. Individual Ura+ transformants were used to inoculate 1 ml cultures in SCD-Ura media supplemented with 1% raffinose and 2% galacotse. After 4 days growth at 30°C, appropriate dilutions of the cultures were plated on YEPD or SCD-Ura-Lys. For each strain, 12 to 36 cultures were used to determine rates and 95% confidence intervals using the Lea-Coulson method of median [76], [77]. Where indicated, rates were determined using the p0 method [78].
Characterization of GCR events using pulse field gel electrophoresis (PFGE) and microarray-based comparative genome hybridization (array-CGH) were carried out as previously described [79]. The microarrays used were Agilent custom 8x15k design (AMID 028943), with 14,965 unique 60 nt oligonucleotide probes, and a median genomic spacing of 774 bp. Detailed microarray probe composition and hybridization data are available upon request.
The BglII-digested pAG25 plasmid [80] was integrated at a site proximal to the KanMX4 marker present in the class III clones. Genomic DNA from the clones containing the integrated plasmid was extracted and digested with EcoRV followed by re-ligation and rescue of re-circularized plasmids in E. coli. Restriction analyses of the rescued plasmids were consistent with inverted duplication structures. SacI restriction fragments containing the center of symmetry of the inverted duplicated regions were sub-cloned into a pUC18 plasmid vector and sequenced using primers positioned just outside the vector's multicloning site. The sequences of the inverted duplication breakpoints from clones A7 and A8 are shown in Fig. S4A and S4B. The secondary chromosomal rearrangements in class III had breakpoints at Ty/LTR sequences, and were consistent with the ectopic homologous recombination mechanism most often observed in yeast GCRs [81].
|
10.1371/journal.pgen.1002694 | ELK1 Uses Different DNA Binding Modes to Regulate Functionally Distinct Classes of Target Genes | Eukaryotic transcription factors are grouped into families and, due to their similar DNA binding domains, often have the potential to bind to the same genomic regions. This can lead to redundancy at the level of DNA binding, and mechanisms are required to generate specific functional outcomes that enable distinct gene expression programmes to be controlled by a particular transcription factor. Here we used ChIP–seq to uncover two distinct binding modes for the ETS transcription factor ELK1. In one mode, other ETS transcription factors can bind regulatory regions in a redundant fashion; in the second, ELK1 binds in a unique fashion to another set of genomic targets. Each binding mode is associated with different binding site features and also distinct regulatory outcomes. Furthermore, the type of binding mode also determines the control of functionally distinct subclasses of genes and hence the phenotypic response elicited. This is demonstrated for the unique binding mode where a novel role for ELK1 in controlling cell migration is revealed. We have therefore uncovered an unexpected link between the type of binding mode employed by a transcription factor, the subsequent gene regulatory mechanisms used, and the functional categories of target genes controlled.
| One of the major outstanding questions in eukaryotic gene regulation is how transcription factors with seemingly very similar DNA binding specificities elicit specific biological responses. The ETS transcription factor family provides a paradigm for investigating this phenomenon. Here, we have focused on the ETS transcription factor ELK1, and by combining genome-wide binding analysis coupled with gene expression analysis we have dissected two distinct gene regulatory activities for this transcription factor. In each of these regulatory modes, ELK1 exhibits distinct DNA binding characteristics which correlate with either positive or negative transcriptional activities and give rise to functionally distinct gene expression programmes. We demonstrate a novel function for ELK1 in controlling cell migration through one of these regulatory modes. Thus, we have demonstrated a clear link between the types of regulatory region bound by a transcription factor and its ability to control gene expression (i.e. in a positive or negative manner) and the functional downstream consequences of its target gene cohort. This work has implications for understanding how members of other multi-protein transcription factor families might function to generate different downstream functional consequences through engaging with different types of regulatory regions.
| Eukaryotic transcription factors are grouped into families based on their common DNA binding domains and these families can extend to dozens of members in mammalian cells. This is typified by the ETS-domain containing transcription factors, with human cells encoding 28 different proteins (reviewed in [1]). All these transcription factors contain very similar DNA-binding domains, and at least in vitro, their DNA binding specificities are very similar [2]. It is therefore a challenge to understand how individual family members can control specific gene expression programmes without unwanted crosstalk. Such crosstalk would potentially lead to functional redundancy due to the ability of different family members to bind to the same sites. Indeed, this has been suggested by genome-wide interrogation of DNA binding by different ETS transcription factors, where multiple family members can associate with the same genomic regions [3]–[5]. Such functional redundancy might explain why many of the phenotypes caused by the loss of ETS proteins are relatively subtle, despite the widespread expression of these transcription factors (reviewed in [6]). However, even with this huge potential for functional redundancy, several mouse knockout studies have revealed specific phenotypes for individual ETS transcription factors, suggesting that they function at least partially in a non-redundant fashion. One way for helping to achieve this specificity of action is through functional cooperation with other transcription factors. This is illustrated by the association of ETS1 with RUNX1 [4] and ELK1 with SRF [5], although in both cases, this co-association is only extended to a minority of the binding events identified in vivo.
ELK1 together with ELK4/SAP-1 and ELK3/SAP-2/Net, constitute the ternary complex factor (TCF) subfamily of ETS-domain transcription factors (reviewed in [7]–[8]). Like all ETS proteins, these transcription factors all bind to variants of the GGAA/T motif embedded in a larger 10 bp consensus sequence in vitro, and in the case of ELK1, this binding preference is also recapitulated in vivo [5]. Members of the TCF subfamily can function in a cooperative manner with SRF, and this is driven in part through the close juxta-positioning of their DNA binding sites, but also through direct protein-protein interactions [9]–[11]. Amongst the TCFs, ELK1 is the best studied and it is directly activated by phosphorylation in response to activation of the MAP kinase signalling cascades [reviewed in 7]–[8]. Mouse knockouts show minimal phenotypic changes [12], suggesting that there might be functional redundancy amongst members of this subfamily. This was recently demonstrated to be the case in the context of T-cell differentiation, where the loss of ELK1 caused only subtle effects and there was clear redundancy of function between ELK1 and ELK4 [13]. This redundancy of function was demonstrated at both the level of DNA binding and target gene regulation. Similarly, redundancy at the level of chromatin binding has been shown to occur in HeLa-S3 cells, where depletion of ELK1 led to decreased binding to chromatin, and a concomitant rise in the binding of ELK4 to the same regions in a subset of targets [14]. Thus, understanding the function of ELK1 and other TCFs is complicated by the compounding factors associated with functional redundancy in this transcription factor family. Recently however, a genome-wide RNAi screening study identified ELK1 as a critical factor in promoting cell survival in human breast-derived MCF10A cells [15]. MCF10A cells therefore provide a tractable system for dissecting ELK1 function.
Here, we have combined ChIP-seq and gene expression array analysis to interrogate the ELK1 target gene network in MCF10A cells. We demonstrate that ELK1 binds to its target regions in two distinct ways: uniquely or redundantly with other ETS proteins. The binding regions associated with these two types of interaction show different characteristics concerning the frequency and quality of the ELK1 binding sites and the association with the binding of heterotypic transcription factors. Unexpectedly, the two types of binding regions are associated with different modes of target gene regulation and moreover, this differential regulation also affects distinct functional categories of target genes. This is demonstrated for the unique ELK1 binding mode, where this factor controls cell migration, through acting in a positive manner to activate a set of functionally related target genes.
Functional redundancy amongst members of multi-gene transcription factor families such as the ETS transcription factors is a potential problem when attempting to uncover the function of individual family members (reviewed in [1]). We therefore selected the human breast epithelia-derived MCF10A cells to dissect the role of ELK1 in controlling gene transcription, as ELK1 has been shown to be important for their survival and its deficiency cannot therefore be fully compensated for by other family members [15]. First, we depleted ELK1 levels in MCF10A cells and measured the effect this had on the transcriptome. The effect of ELK1 depletion was assessed in cells grown in the absence of EGF or in cells treated with EGF for 30 mins to activate ERK MAP kinase pathway signalling. ELK1 levels were efficiently reduced following treatment with siRNA (Figure 1A) and ERK activation and ELK1 phosphorylation were rapidly induced by EGF treatment (Figure 1B). Over 6000 genes consistently changed their expression (P-value<0.05; Q-value<0.1) following depletion of ELK1 either in the presence or absence of EGF, and roughly equal numbers of genes were up- and down-regulated in each case (Figure 1C; Table S1). In contrast, far fewer genes consistently changed their expression upon treatment with EGF (Figure S1), suggesting a more pleiotropic cellular response to ELK1 loss. Importantly, the majority of genes whose expression changed upon ELK1 depletion did so irrespective of the activity status of the ERK signalling pathway as only 10% of siELK1-sensitive genes were uniquely altered by ELK1 depletion in the presence on EGF, while 73% were changed both in the presence and absence of EGF (Figure 1D). This suggests a role for ELK1 in controlling gene expression which is largely independent from EGF signalling status.
The large number of gene expression changes elicited in MCF10A cells by ELK1 depletion suggested that a sizeable proportion of the changes likely arise as an indirect consequence of ELK1 loss. Therefore we used ChIP-seq to establish the direct target genes for ELK1. These studies were performed in MCF10A cells grown in the absence of EGF, as ELK1 binding to chromatin is thought to occur irrespective of signalling conditions [16]. A total of 529 genomic binding regions for ELK1 were identified in two independent experiments (FDR<10), and denoted as a “high confidence” dataset (Table S2). By assigning binding regions to the nearest annotated TSS, 516 target genes were identified. A range of binding regions were validated by ChIP followed by qPCR and all the regions tested showed significant enrichment for ELK1 binding compared to a control non-specific antibody (Figure 2A; Figure S2). We also tested several regions from a lower confidence dataset where peaks were only identified in one of the two experiments and although these scored positive in qPCR-based assays, their relative enrichment levels were generally lower than observed for the high confidence data set (Figure 2A; Figure S2). Many of the regions in the high confidence dataset are also bound by ELK1 in a different human breast cell derived line, MDA-MB-231 cells (Figure S3A). Moreover, there is a large overlap of ELK1 binding regions with those identified in two previous studies using ChIP-chip (59%; [5]) and ChIP-seq (43%; [17]) in HeLa cells, with 106 (33%) regions in common between all three studies (Figure S4). This suggests that there is a core set of ELK1 binding regions common to several cell types but also a number of cell type-specific binding events. ELK1 binding regions in MCF10A cells are enriched in promoter -proximal regions with 36% within 10 kb upstream from the TSS of the nearest gene (Figure 2B). Functionally, the ELK1 binding regions were associated with genes grouped under a number of distinct gene ontology classifications by analysis using GREAT [18]. Prominent categories included a number of terms associated with the regulation of gene expression (e.g. “ribosomal subunit” and “mediator complex”) and with the actin cytoskeleton and cell adhesion (Figure S5). The association of ELK1 with genes encoding the core gene expression machinery was expected from previous results [5], but a role in potentially controlling genes involved in cytoskeletal-mediated events is a novel discovery.
Next, we assigned each ELK1 binding region to the nearest gene annotated in the RefSeq database, and compared this list with the ELK1-regulated genes revealed by expression microarray analysis. Over half of the genes associated with ELK1 binding regions (273 out of 516 genes) are also deregulated under at least one of the conditions we tested upon ELK1 depletion (245 were changed in the absence of EGF and 223 in the presence of EGF). Roughly equal numbers of ELK1 binding regions are associated with up- and downregulated genes, irrespective of EGF treatment (Figure 2C). Importantly, comparison of the ChIP-seq-derived ELK1 target genes with randomly selected gene sets demonstrated that the overlaps with the ELK1-regulated gene expression data are highly significant in all cases (z-scores ranging from 3.65 to 4.99). Further analysis of the results demonstrated that the majority of the 273 deregulated genes associated with ELK1 binding regions are the same irrespective of whether EGF was added to the cells or not (Figure 2D). However, amongst genes which are upregulated by EGF, there are 17 that are bound by ELK1, and of these, 13 show diminished expression following ELK1 depletion. This is in keeping with previous findings that ELK1 is associated with EGF/ERK pathway-mediated target gene activation but further illustrates that this appears to be a relatively minor role of ELK1 in this system. Functionally, the direct ELK1 target genes defined by this analysis retained categories identified in the entire ChIP-seq data set such as association with gene expression control and the actin cytoskeleton but in addition, a new category, apoptosis/cell death, was identified as being regulated by ELK1 (Figure S6A–S6C).
Together, these results identify a core set of direct ELK1 target genes, whose regulatory regions are bound by ELK1 and whose expression is perturbed upon depletion of ELK1. This set of genes likely represent directly regulated ELK1 target genes which we subsequently analysed further and are henceforth referred to as “ELK1 target genes”.
Previous ChIP-chip studies using promoter arrays demonstrated that ELK1 binding regions fall into three broad categories; regions that are also bound by SRF, regions that can also be bound by other ETS transcription factors and regions which are apparently uniquely bound by ELK1 independently from these other transcription factors [5], [14]. To establish whether we could detect similar overlaps with ELK1 binding regions on a genome-wide scale, we compared our ChIP-seq data with published ChIP-seq data for SRF and other ETS transcription factors. First, to address the potential for redundant binding of different ETS proteins, we compared the ELK1 ChIP-seq data with that of a divergent ETS factor GABPA [19]. A substantial overlap was observed between ELK1 and GABPA binding regions (43% of ELK1 binding regions) despite the differences in cell types analysed (Figure 3A). This overlap in binding suggests potential redundancy of binding site occupancy by these different ETS factors. We define the group of regions which can potentially be bound by both ELK1 and the divergent ETS protein GABPA as “ELK1 redundantly bound regions”, whereas those bound by ELK1 and not GABPA are termed “ELK1 uniquely bound regions”. To establish whether this subcategorisation held true when other ETS transcription factors were considered, we compared ChIP-seq data for nine other ETS transcription factors performed in a variety of cell lines with ELK1 binding regions which were classified as “unique” and “redundant”. With the exception of the highly related subfamily member ELK4, very little overlap was seen with binding regions in the “unique” subcategory but in 8/10 cases, there was significantly more overlap of these other ETS transcription factors with the “redundant” dataset (Figure S7B). Importantly, 67% of the “unique” ELK1 binding regions were not bound by any of the more divergent ETS proteins in any of these studies, supporting our subcategorisation of these regions. We also compared our ChIP-seq data with ChIP-seq data for SRF from Jurkat cells [19]. Again, a substantial overlap was seen between ELK1 and SRF binding regions despite the different cell types involved (28% of ELK1 binding regions; P<0.001) (Figure S7C). A comparison of ELK1 and SRF bound regions with the binding regions for GABPA permits further subcategorisation of binding events, and reveals a group of regions which are bound by ELK1 alone in the absence of potential redundant binding with GABPA or co-binding with SRF (Figure S7C).
In order to look for potential reasons for the non-identical transcription factor occupancy, we compared the regions bound uniquely by ELK1 (termed “unique”) and those which can also be bound by GABPA (termed “redundant”). First, we examined their location relative to the nearest TSS. The “unique” regions showed a broad distribution with only 26% being located within 2 kb of the TSS, whereas 93% of the “redundant” regions were located in this region and these were largely tightly centered around the TSS (Figure 3B). Next, we implemented a de novo search for over-represented DNA motifs within the binding regions. Three prominent motifs were identified in the “unique” dataset corresponding to ETS, SRF and AP1 binding sites, whereas the “redundant” dataset revealed only a motif corresponding to an ETS transcription factor binding site (Figure 3C). Similar results were obtained when using position weight matrices for all known human transcription factor binding motifs listed in the JASPAR and TRANSFAC databases to search for over-represented motifs in these two datasets (data not shown). The ETS motifs identified in the ELK1 binding peaks closely resembled the motifs identified in a recent high throughput in vitro binding site selection study [2] (Figure S8) and what was identified in a ChIP-chip study for ELK1 performed in HeLa cells [5], although differences in nucleotide preferences at several positions can be observed, generally being more stringent in vitro. Further inspection of the ELK1 binding regions demonstrated that the frequency of the three hexamers comprising the core CCGGAAGT binding motif was much higher in the “redundant” (763 sites in 226 regions) than in the “unique” dataset (403 sites in 303 regions). This also suggests that more than one binding event might be associated with “redundant” regions. Indeed, the number of motifs per region corresponding to hexameric derivatives of CCGGAAGT is significantly higher in the “redundantly” bound regions (commonly three sites per region) than in the “unique” regions (generally only one site per region) (Figure 3D). We investigated whether there were any particular spatial constraints among the multiple ETS sites found in the “redundant” regions but there was nothing obvious detected. In contrast to the differences seen when considering the core CCGGAAGT binding motif, the occurrence of hexameric derivatives of the variant ETS binding motif CAGGATGT is virtually identical in the two datasets (Figure 3D). However, some unique regions have only this relaxed variant, whereas all redundant regions have at least one “strong” consensus site. This suggests that the binding specificity is generally more divergent from the core consensus when unique binding of ELK1 is observed. Part of the reason for the relaxed binding specificity likely relates to the presence of co-occurring SRF binding motifs which are present in 54% of “unique” regions but only 27% of “redundant” regions. Indeed, a significantly greater number (P-value<1×10−6) of SRF motifs appear to form “modules” with adjacent ETS motifs in the “unique” dataset, where the two motifs co-occur within a 50 bp window (Figure 3E). It is possible that co-binding of factors to AP1 binding motifs also influences the binding specificity and functionality of ELK1, as the frequency of FOS binding to the same regions in a different ChIP-seq study [20] is significantly higher with the “unique” binding regions (Figure 3F). Interestingly, there are significantly more overlaps between ELK1 binding and FOS binding in regions associated with genes activated by ELK1 (27/141) rather than in genes where ELK1 acts in a negative manner (13/137) (P-value = 0.021), which is suggestive of a role for AP1 in enhancing the activity of ELK1. In keeping with a potential role for AP1 in modulating ELK1 function, the frequency of occurrence of AP1 binding motifs is also significantly higher in the “unique” binding regions (Figure S7D). Indeed, we can detect binding of the AP1 component FOS to several of the “unique” ELK1 binding regions (Figure S7F).
Together, these results therefore identify two major types of ELK1 binding regions which can be independently characterized by their locations, the types and numbers of DNA binding motifs they contain and the factors which can potentially co-occupy or compete for binding to these sites. Regions which are uniquely bound by ELK1 tend to contain fewer and/or weak ETS binding motifs and are often bound by other partner transcription factors such as SRF and AP1, whereas those redundantly bound by different ETS factors more often contain multiple strong ETS binding motifs, and there is little evidence for the occurrence of common co-binding transcription factors.
Having defined two different types of ELK1 binding regions, we next wished to investigate whether these regions are associated either with different gene regulatory mechanisms, and/or with controlling different cellular processes. First, we used unsupervised k-means cluster analysis to provide an unbiased way to identify groups of genes which responded differently to ELK1 depletion in the presence and absence of concurrent treatment with EGF. Here, we focussed only on the 273 direct target genes identified as being associated with regions bound by ELK1 in the ChIP-seq analysis. Data are depicted relative to average expression of each gene across all conditions to enable common regulatory patterns to be discerned. Eight gene expression clusters were identified which show distinct patterns of responses to these different treatments (Figure 4A, 4B; Figure S9; Table S3). We then examined whether the “unique” or “redundant” ELK1 target genes were associated with any particular cluster(s). Interestingly, we found a clear separation of target gene response associated with different ELK1 binding modes. Clusters 2 and 4 were significantly enriched for “unique” binding whereas clusters 7 and 8 were enriched for “redundant” binding (Figure 4C). Cluster 2 contains genes which are upregulated by EGF treatment but show an attenuated response upon ELK1 depletion, whereas genes in cluster 4 are largely unaffected by EGF but downregulated following ELK1 depletion (Figure 4A and 4B). In contrast, the genes in clusters 7 and 8 are largely non-responsive to EGF treatment but their expression is enhanced upon ELK1 depletion (Figure 4A and 4B). In total, over 75% (205) of the ELK1 target genes are found within the clusters which are strongly correlated with either “unique” or “redundant” binding. We analysed this further by comparing the response of genes bound either by ELK1 alone (“unique”) or by ELK1 and GABPA (“redundant”) (see Figure 3A) to ELK1 depletion. Here, a significant shift in the regulatory mode could clearly be seen, with “uniquely” bound genes being largely downregulated while “redundantly” bound genes were upregulated following ELK1 loss (Figure 4D). Mechanistically, one likely mode of regulation of the latter class of genes, is that ELK1 normally acts repressively at these genes and upon depletion is replaced by a different transcription factor such as GABPA which can potentially provide stronger gene activation. To provide data supporting this prediction, we compared cells depleted of either ELK1 or GABPA and examined the response of three genes predicted to be redundantly bound by ELK1 and GABPA. The expression of all three genes increased upon depletion of ELK1. However, the reciprocal occurred upon depletion of GABPA and decreased expression of all three genes was observed (Figure 4E). Thus, ELK1 and GABPA work antagonistically in controlling gene expression and this is particularly evident in clusters 7 and 8 for targets such as SCNM1 and MDM4. Overall, these results indicate that the regulatory mode attributed to ELK1 strongly correlates with the type of ELK1 binding mode adopted on the target genes.
Next, we wanted to examine whether the type of binding and regulation adopted by ELK1 might also relate to functionally distinct outcomes in terms of the cohorts of genes regulated. We therefore examined whether any of the enriched GO term categories for the direct ELK1 target genes are preferentially associated with any of the different expression clusters. Importantly, we found that genes from particular functional categories are not evenly spread throughout the clusters (Figure 4F). Instead, genes forming the category “cytoskeleton and migration” are enriched in clusters 2, 4, 5 and 6, while genes within the “cell survival”-related terms are enriched in clusters 2 and 6. As one common defining feature of clusters 2, 4, 5 and 6 is downregulation following ELK1 depletion, these data suggest a positive role for ELK1 in driving transcription of genes in these functional categories. Therefore the gene expression clusters associated with particular ELK1 binding and regulatory modes are associated with distinct functional categories of genes. This suggests that the two types of ELK1 binding regions might regulate the expression of genes encoding proteins involved in distinct cellular processes and ultimately lead to distinct functional outcomes. To examine whether this is the case, we separated the direct ELK1 target gene dataset into “unique” and “redundant” groups according to the type of ELK1 binding mode, and screened these datasets for different functional categories of genes using DAVID [21]. We then ranked the resulting enriched GO terms according to statistical significance and saw a clear separation of these GO terms, according to association with either “unique” or “redundant” ELK1 binding with only minimal overlap (Figure 4G; Table S4). Genes “uniquely” bound by ELK1 were generally associated with terms related to the actin cytoskeleton (e.g. actin binding, focal adhesions etc.), whereas “redundantly” bound genes were generally associated with gene expression control (e.g. RNA processing, translation etc.). Some GO term categories related to gene expression were also associated with genes assigned to “uniquely” bound ELK1 regions but these strongly differed from the categories associated with genes assigned to “redundantly” bound regions, demonstrating a further separation of binding mode with respect to the types of genes regulated. Interestingly, terms related to cell survival were generally associated with genes assigned to “uniquely” bound ELK1 regions but also appeared in the “redundantly” bound dataset, albeit with lower significance.
These results therefore reveal an unexpected link between the mode of ELK1 binding to the regulatory region of a gene, the type of regulation of the associated genes and the functional classifications of the target genes controlled. In particular, genes associated with the cytoskeleton and migration, tend to be regulated in a positive manner by ELK1, and also are apparently bound specifically by ELK1 as they are not targets for potential redundant binding by alternative ETS transcription factors like GABPA.
The above results indicate that a key role for ELK1 is mediated through its “unique” binding mode where it controls the activity of genes associated with the actin cytoskeleton and cell migration. To explore the functional significance of this association, we created networks of proteins encoded by “unique” and “redundant” ELK1 target genes using coexpression, textmining, knowledge and experimental data as proximity criteria to identify interconnectivities. These networks were screened for enrichment in GO terms centred on specific cellular functions. The genes associated with “migration/cytoskeleton” formed a prominent subnetwork within the “unique” dataset; more interestingly, a large number of the genes which form the core of this subnetwork were shown by microarray analysis to be misregulated following ELK1 depletion (Figure 5A and Figure S11). The role of ELK1 in controlling key nodes in this subnetwork was confirmed by qPCR. Nine different target genes were selected from the GO term category “migration/cytoskeleton”, whose regulatory regions are bound by ELK1. The majority of these showed decreases in expression upon ELK1 depletion from MCF10A cells in the absence and/or presence of EGF (Figure 5B, and Figure S10A). The majority of these decreases in ELK1 target gene expression were also observed in MDA-MB-231 cells (Figure S3B). Importantly, depletion of GABPA did not cause any significant changes in expression of these genes, in keeping with our designation of these genes as “unique” targets for ELK1 (Figure S10B). SRF has also been associated with controlling genes involved in the actin cytoskeleton and cell migration [22], [23]. We therefore also assessed the role of SRF in regulating the same group of genes. The expression of three genes was significantly reduced upon siRNA depletion (Figure S10C). Two of these, CTGF and ACTB, are also positively regulated by ELK1. Thus ELK1 and SRF appear to have both distinct and common targets amongst the “migration/cytoskeleton” gene network.
A second subnetwork of “cell survival” was also identified which is comprised of genes associated with the “unique” ELK1 dataset (Figure S12), but for the “gene expression” category, two different types of network could be observed (Figure S13A and S13B). In the latter case, again a subnetwork could be identified which is controlled through the “unique” binding of ELK1 and which features clusters of genes encoding signal-responsive transcription factors including immediate-early gene products and several nuclear hormone receptors. In contrast, an additional subnetwork was revealed which is apparently rather controlled through regions that can be redundantly bound by ELK1 and other ETS factors such as GABPA, and is mainly made up of genes encoding ribosomal subunits and mediator components. These findings support the association of different ELK1 binding modes with different functional categories of target genes, but also highlight the importance of ELK1 in controlling key nodes in the networks.
Our results predict that the loss of ELK1 should have profound consequences for the cellular phenotype due to the collapse of the networks involved with controlling the relevant cellular functions, especially in processes associated with the cytoskeleton and downstream effects such as cell migration. To test this, we first investigated the status of the cytoskeleton following ELK1 depletion in MCF10A cells. Following ELK1 depletion, cells became less spread and exhibited altered cytoskeletal characteristics, including a loss of membrane protrusions and enhanced levels of subcortical actin (Figure 6A and 6B). The loss of membrane protrusions suggested that the cells might lose their migratory properties. To test this, we used wound healing assays, and upon ELK1 depletion wound closure was greatly attenuated (Figure 6C and 6D). ELK1 depletion also led to reductions in cell numbers (Figure S14) which could at least in part contribute to the reductions in wound closure efficiency. We therefore used single cell tracking to examine the movement of individual cells associated with cell clusters. Fewer cells detached from these clusters in the siELK1-transfected population (Figure 6E) and those cells which did detach showed greatly impaired migration (Figure 6F). Many of the same phenotypic features could be observed upon ELK1 depletion in MDA-MB-231 cells (Figure S3C–S3E).
These results confirm that ELK1-mediated regulation of a target gene network associated with the actin cytoskeleton and cell migration correlates with the expected phenotypic consequences. To confirm that the key ELK1 target genes in this network play a role in controlling cytoskeletal-related activities such as cell migration, we individually depleted nine of these genes (Figure S15A) and analysed the resulting cellular phenotypes. Defects in the actin cytoskeleton (Figure S15B and S15C) and reduced wound healing (Figure S15D and S15E) were observed in several cases. Importantly, cell migration was defective in the majority of cases (Figure 6G and Figure S16), clearly demonstrating that the ELK1 target genes play a role in controlling this process.
Genome-wide studies are beginning to reveal the target genes for many transcription factors and one of the unanswered questions is how specific responses are generated by a particular transcription factor, especially in the presence of other proteins from the same family that can potentially bind the same sites. Here we have investigated the ETS family transcription factor ELK1 and made the unexpected discovery, that the types of binding mode exhibited by ELK1, correlate not only with the regulatory outcomes at the associated target genes but lead to the control of distinct networks of genes with defined cellular functions.
Our results clearly demonstrate that ELK1 target genes can be functionally classified according to their mode of ELK1 binding and this in turn controls their mode of regulation. Inspection of ELK1 binding regions demonstrates that these can be divided into at least three distinct categories based on the binding of other transcription factors; co-binding with a different transcription factor SRF, competition with GABPA (and potentially other ETS proteins) and binding in a distinct manner independently from these. To simplify our analysis, and concentrate on the potential interplay with other ETS transcription factor family members, we focused on two modes of ELK1 binding; regions which can be occupied by GABPA and ELK1 (“redundant” regions) or regions that can be occupied by ELK1 and apparently not GABPA (“unique regions”). While further analysis of the redundant and unique regions for binding by other ETS proteins from other ChIP-seq studies broadly supports this classification, it is important to emphasise that we cannot definitively conclude that the apparently “unique” regions cannot be bound by another ETS factor under different conditions or in another cell type. Nevertheless, this classification enabled us to ultimately define an important role for ELK1 in controlling cell behaviour where it acts through genes associated with the “unique” regions. These two types of binding regions showed distinct characteristics. “Redundant” regions are generally centered on the TSS and at the sequence motif level, contain multiple sites which closely match the generic consensus ETS protein binding motif CCGGAAGT. Moreover, there is little evidence for the occurrence of common co-binding transcription factors. In contrast, the “unique” regions are located in a wide distribution around the TSS and generally show a relaxed match to the ETS binding motif. Furthermore, these regions contain fewer ETS-like binding motifs than the “redundant” regions. The “unique” regions also show motif enrichment and binding of other classes of transcription factors; AP1 and SRF. The association between ELK1 and SRF has been well established previously (reviewed in [5], [24]) but the association with AP1 is a novel finding. This AP1 connection is related to early studies on the ELK1-regulated FOS promoter which suggested feedback inhibition by FOS [25]. Furthermore, recent genome-wide ChIP-seq studies demonstrate that the ETS proteins ERG, ETV1, and ETV4 exhibit strong co-binding with AP1 in prostate-derived cells whereas others (ETS1 and GABPA) do not [26]. Thus, co-association with AP1 appears to be a common mode of action for a subset of ETS transcription factors, including ELK1. Several other features of ELK1 binding have previously been reported from genome-wide binding analyses for other ETS transcription factors. For example, as observed for ELK1, ETS1 generally shows redundant occupancy of proximal promoters with GABPA but specific interactions with a coregulatory partner, RUNX1, generally take place in more distally located regions where ETS1 binds in a “unique” manner [4]. This study also revealed a putative link between “unique” binding of ETS1 and gene function with an association with T cell-specific genes being observed but this link remained functionally untested. Furthermore, the sequence characteristics of the core ETS1 binding sites differ between “unique” and “redundant” binding sites, as we observed in this study for ELK1. Thus, there are clear similarities in how ETS transcription factors generate specificity in their mechanism of action and biological functions. However, despite these underlying similarities, our study clearly demonstrates that ELK1 and ETS1 function in a different manner. Furthermore, our study was also able to make links to regulatory outcomes (see below) at different classes of target genes which have not yet been established for ETS1.
Importantly, the location of the ELK1 binding sites in proximal promoter (defined as +/−2 kb from the TSS) versus distal regions is not as effective at partitioning of the data. Indeed, the “unique” ELK1 binding sites in the proximal promoter regions share characteristics such as the distribution of binding motifs, over-represented functional categories of targets, and association of common regulatory events, with “unique” ELK1 binding sites located in the distal regions (data not shown). Thus, the mechanistic, regulatory, and functional connections that we observe are not due to binding site location and instead reflect whether ELK1 is likely working in a “unique” manner. However, when we partitioned the data according to a different definition of proximal binding (ie −7 kb to +2 kb from the TSS), we could readily detect SRF and AP1 motifs in the distal regions but ETS motifs were less obvious; the reciprocal was true in promoter-proximal regions (data not shown). Moreover, there were significantly more overlaps between distal ELK1 binding motifs with FOS ChIP-seq data [19] (58/229) than with promoter-proximal regions (20/300) (P-value<1×10−6). This might reflect a more combinatorial mode of action of ELK1 when it is associated with distal binding regions.
Once the two types of ELK1 binding mode had been established, we were then able to correlate binding with regulatory outcomes. “Uniquely” bound regions are generally associated with genes where ELK1 functions in an activating capacity, whereas the converse is true of “redundantly” bound regions. Direct ELK1 target genes which are upregulated by EGF treatment are generally associated with “unique” ELK1 binding regions, consistent with a role for ELK1 as an EGF-responsive transcriptional activator. Strikingly, the majority of ELK1 target genes were barely affected by EGF treatment, despite the fact that ELK1 plays a regulatory role at a large proportion of these. Thus although ELK1 is generally thought to function by acting as a direct recipient of ERK pathway signaling, this appears to be the exception rather than the rule when considering its broader cellular function (reviewed in [7]–[8], [24]). It should, however, be emphasised that even in the absence of EGF, basal ERK activation is present in MCF10A cells (Figure 1C) and thus we cannot rule out a role for constitutive low level signaling working through ELK1 to its target genes.
The lack of EGF response is particularly prominent among the “redundantly” bound ELK1 target genes. In these cases it appears likely that other ETS transcription factors such as GABPA can substitute for ELK1 binding, and the function of ELK1 is to compete for DNA occupancy with those ETS transcription factors. Indeed, depletion of ELK1 or GABPA has a reciprocal effect on the expression of “redundantly” bound target genes, suggesting that a dynamic equilibrium between these transcription factors exists in the cell to maintain target gene expression at the correct level. In this scenario, ELK1 would either be repressive in nature and GABPA activating, or alternatively, both might activate but with GABPA activating to a higher level than ELK1. Similar effects could occur amongst different family members as has been suggested previously, at least at the level of DNA binding [3]. Importantly, our data also suggest that while redundancy at the level of binding is observed, this binding might not translate into functional redundancy.
Given the different binding modes identified for ELK1 and their correlation with different regulatory outcomes for the associated target genes, we then wished to examine whether this translated into specific phenotypic responses. Unexpectedly, we uncovered a clear distinction between the type of ELK1 binding mode employed and the categories of target genes regulated. The “unique” ELK1 binding regions were closely associated with genes encoding proteins involved in the actin cytoskeleton and related processes, whereas the “redundantly” bound regions were generally associated with genes encoding proteins controlling aspects of gene expression. The latter observation suggests a general role for ETS proteins in maintaining the levels of genes encoding the gene expression machinery at the correct levels. A similar conclusion was reached based on overlapping binding of different ETS-proteins in a ChIP-chip study, where commonly bound target genes were often designated as “housekeeping” [3]. However, our results suggested a unique non-redundant function for ELK1 in controlling the expression of proteins associated with the actin cytoskeleton and we subsequently demonstrated that this was indeed the case. Upon ELK1 depletion, cytoskeletal defects were observed and these led to defects in cell migration. A previous study provided hints at such a connection with MMP9 identified as an ELK1 target gene relevant to cell migration [27]. However, a direct link was not made between ELK1 and cell migration control and this gene was not identified as a direct target for ELK1 in our study.
Interestingly, SRF has also been associated with controlling the actin cytoskeleton and cell migration [22], [23] but the role of ELK1 as a potential partner protein in that process has not been addressed. Indeed, it is thought that the alternative SRF partner protein MAL/MRTF plays the major role in this process (reviewed in [28]). More recently, two different types of SRF binding sites have been shown to control cytoskeletal gene expression, where SRF binds either together or independently with the ETS transcription factor PU.1 and these binding events correlate with activation through ubiquitous promoter- and cell type-specific enhancer driven mechanisms, respectively [29]. Thus, different SRF binding modes are associated with the same biological process and type of regulation, while here we reveal that ELK1 binding modes dictate different downstream outcomes. Many of the target genes that are directly regulated by ELK1 differ from the SRF-regulated genes encoding proteins associated with the actin cytoskeleton. Indeed, of the 57 genes in the actin/migration network associated with ELK1 binding, only 14 were previously shown to be direct SRF targets in HeLa cells [5]. This suggests that these two transcription factors may control the expression of different components of the network rather than acting in a more general coordinated fashion. Indeed, we have shown that only a subset of the ELK1 target genes which rely on ELK1 for their expression, are also positively regulated by SRF. One of these, ACTB, encodes β-actin, and is a well established SRF target gene [e.g. 22], [ 23]. Importantly, we have confirmed the role of ELK1 in controlling cell migration in several human breast-derived epithelial cell lines, and it is possible that ELK1 may play a role in cellular metastasis in the context of breast cancer. The other major categories of ELK1-regulated target genes are associated with cell survival and apoptosis. However both “uniquely” and “redundantly” bound target genes appear to be involved. This association was expected due to the original observation that ELK1 is required for the survival of MCF10A cells [15].
In summary, this study has identified an intriguing link between different modes of transcription factor binding and the control of different functional categories of target genes. This permits a specific phenotypic response to be achieved depending on the intrinsic properties and the co-regulatory activities occurring at different transcription factor binding sites. In the case of ELK1, we uncover a specific role in controlling genes associated with the actin cytoskeleton and also determine that it has a second function in which it acts in a dynamic fashion with other ETS transcription factors to maintain the correct expression of components of the gene expression machinery. Future studies will attempt to uncover the mechanistic differences utilized by ELK1 in the different regulatory scenarios.
MCF10A cells were maintained in DMEM/F12 containing 5% horse serum, 20 ng/ml EGF, 10 µg/ml insulin, 100 ng/ml cholera toxin and 0.5 µg/ml hydrocortisone. MDA-MB-231 cells were grown in DMEM containing 10% FBS.
For RNAi, MCF10A and MDA-MB-231 cells were plated out into a mixture of 83% growth medium, 17% OptiMEM (Invitrogen), 20 nM siRNA and 0.33% Lipofectamine2000 (Invitrogen) followed by replacement of the transfection mix with appropriate growth media after 12 hours. All siRNA constructs were ON-TARGETplus SMART pools from Dharmacon except for ELK1, where a custom-designed siRNA duplex corresponding to the sequence GGCAATGGCCACATCATCT was used, and for GABPA where additional duplexes were used from Santa-Cruz. For experiments involving the culture of siRNA-transfected cells longer than 64 hours, siRNA transfections were repeated at t = 48 hrs. For SRF knockdown followed by RT-PCR analysis, two repeats were done with an ON-TARGETplus SMART pool (Dharmacon) and one was performed with an alternative siRNA duplex (Santa Cruz).
Real time RT-PCR was carried out as described previously [30]. The primer pairs used for RT-PCR experiments are listed in Table S5.
MCF10A cells were seeded into 6-well plates (520,000/well), transfected with siELK1 or siGAPDH (control) and maintained in media depleted of EGF for 48 hours. Cells were then stimulated with complete media containing EGF for 30 minutes (with non-stimulated populations used as control). Three biological replicates were performed for each condition. RNA was isolated using the RNeasy kit (QIAgene) according to the manufacturer's protocol and quantifiied with a Nanodrop ND-1000 spectrophotometer (Nanodrop Technologies). Sample labeling and hybridization to Affymetrix GeneChip Human Genome U133 Plus 2.0 arrays were performed according to manufacturer's instructions.
Upon collection of signal, technical quality control was performed with dChip (V2005) [31] using default settings. Background correction, quantile normalization, and gene expression analysis were performed using RMA in Bioconductor [32]. Principal component analysis (PCA) was performed with Partek Genomics Solution (version 6.5, Copyright 2010, Partek Inc., St. Charles, MO, USA). Differential expression analysis between samples was performed using Limma with the functions lmFit and eBayes [33]. A two-factor ANOVA model was used with batch pairing, since batch pairing was evident in the PCA analysis. Gene lists of differentially expressed genes were controlled for false discovery rate (FDR) errors using the method of QVALUE. All probesets with signals lower than or equal to the background level in the EGF treated control samples were discarded. For fold changes of signal between any two conditions, only probesets with Q values lower than 0.1 and P values lower than 0.05 were retained. For probesets complementary to multiple HUGO official gene symbols, only one gene name was retained. Probesets associated with gene names showing changes of signal in opposite directions were excluded from further analysis and if multiple probesets were associated with one official gene symbol, only the probeset exhibiting the largest change of expression was retained for further analysis. The microarray expression data have been submitted to array express (accession number E-MEXP-3407).
For clustering, signal intensities assigned to each probeset were log10-transformed and z-scores were calculated [34]. These transformed signal intensities were subsequently clustered in MultiExperimentViewer (MeV) [35] using the k-means algorithm with a preset of 8 clusters. These were subsequently organised using the hierarchical clustering algorithm.
Cells seeded in duplicate into 12-well plates were transfected with appropriate types of siRNA in media depleted of EGF (MCF10A cells) or FBS (MDA-MB-231 cells). After 48 hours wounds were created in each well using sterile pipette tips; wounds were scanned visually to ensure similar width. Cells were washed twice with 1× PBS and media containing either 20 ng/ml EGF (MCF10A cells) or 10% FBS (MDA-MB-231 cells) was added. For MDA-MB-231 cells, after 18 hours cells were stained with crystal violet. Four images were taken for each of the tested conditions, the area unoccupied by cells in each image was measured. Average values for each biological repeat were normalised to control (siGAPDH transfection). For MCF10A cells, upon stimulation plates were transferred to a chamber heated to 37°C and aerated with 5% CO2 and imaged for 15 hours. The area unoccupied by cells was measured at one hour intervals and normalised to values specific for the siGAPDH-transfected control. Alternatively, after 15 hours cells were fixed and stained with crystal violet. For single cell tracking, sparsely seeded cultured of MCF10A cells were treated identically as in wound healing assays (without the formation of wounds). All images were processed and analysed using ImageJ and Adobe Photoshop CS2.
Fixed cells were visualised using a Leica DMIL upright microscope coupled with a diagnostic instruments HRP045-NIK camera and a 20x/0.30 CPlan Ph1 – air objective. Images were acquired through the SPOT Basic software (Diagnostic Instruments) and processed in ImageJ and Adobe Photoshop CS2. Live cells were imaged in a heated, CO2-enriched chamber of a Leica DM IRE microscope equipped with motorised XYZ stages (Mauhauser) and a 20x/0.50 HC PlanFluotar (Ph2) objective. Images were acquired with a Coolsnap HQ CCD camera (Photometrics) through the Image Pro 6.3 software (Media Cybernetics LtD) and later processed using ImageJ.
For immunofluoresence experiments, cells were fixed with 3.7% paraformaldehyde in 1×PBS, permeabilised with 0.01% Triton X-100 in 1×PBS, washed twice with 1× PBS, stained for 20 min with a solution of AlexaFluor488 phalloidin (Invitrogen), then washed three times with PBS. For detecting DNA, cells were co-stained with a 1 µg/ml solution of the Hoechst dye. Cells were imaged using Olympus BX51 upright microscopes equipped with 20x/0.50 UPlanFln – air objectives and captured with a Coolsnap HQ camera through MetaVue Software (Molecular Devices).
Western blotting was carried out with the primary antibodies; ELK1 (Epitomics, #1277-1), cFOS (SantaCruz, sc7202), ERK2 (SantaCruz, sc154), and phospho-ERK (Cell Signalling, 9106S). The proteins were detected either by chemiluminescence with SuperSignal West Dura Substrate (Pierce) and visualised with a Fluor-S MultiImager (Bio-Rad) or for infrared dye-conjugated antibodies, signal was collected with a Li-cor Odyssey Infrared Imager.
ChIP assays using control IgG (Millipore) or antibodies specific to ELK1 (Epitomics) or cFOS (Santa Cruz) were performed essentially as described previously (O'Donnell et al. 2008) using 3×106 MCF10A cells for a standard ChIP and 5.4×107 cells for a ChIP-seq experiment, grown in media depleted for EGF for 48 hrs for ELK1 ChIP or following stimulation with EGF re-addition for 2 hours for FOS. Bound promoters were detected by quantitative PCR using primers listed in Table S5. Quantitative PCR was performed at least in duplicate, from at least three independent experiments, and analysed as described previously [5].
Samples for ChIP-seq were prepared as described above and libraries were generated and sequencing was performed on a Life Technologies SOLiD 3.5 System (Applied Biosystems) according to the manufacturer's protocols. Two repeats were performed.
SOLiD (Applied Biosystems) *.csfasta sequencer reads were truncated to 32 nt using a perl script: SOLiD_preprocess_filter_v1.pl, available at http://hts.rutgers.edu/filter/. Reads were then aligned to build 18 of human genome from March 2006 (hereafter referred to as hg18) using Corona-Lite 4.2.2 (Life Technologies) and allowing from 0 to 3 mismatches. Aligned reads were used for identification of peaks by MACS v. 1.3.7.1 [36]. For each of the two repeats, the ELK1 signal was compared sequentially to the IgG and input signals and the overlaps of the resulting sets of regions were carried out using Galaxy [37], [38]. The resulting sets of ELK1-specific peaks for each repeat experiment were once more overlapped and a high confidence dataset was defined as regions identified in repeat 2 that overlapped those identified in repeat 1. Regions identified only in repeat 2 are considered to be low-confidence targets. In both cases, only regions characterised by a False Discovery Rate (FDR) lower than 10% were retained. The ChIP-seq data have been submitted to array express (accession number E-MTAB-830).
For association of ELK1 binding regions with potential regulated genes, each ChIP-seq region was assigned to its nearest gene annotated in the refGene table (release 41, May 2010) of the RefSeq Genes track, downloaded from the UCSC Table Browser. This was based on the position of summit of each peak (identified by MACS) and the position of the TSS of the relevant RefSeq gene. Overlaps between ChIP-seq-associated genes and microarray-generated gene expression data were determined using an online tool (http://jura.wi.mit.edu/bioc/tools/compare.php).
The genomic distribution of peaks was determined using CEAS [39]. De novo motif discovery was carried out using Weeder v. 1.4.2 [40]. The resulting top scoring position weight matrices (PWMs) were parsed against known transcription factor PWMs from the JASPAR 2010 database [41] using STAMP [42].
Word-based motif searches were performed using two PERL scripts: CountRegexGFF_IUPAC_1input_nosummary.pl (identifies number of occurrences of a particular IUPAC string on one or both strands of one or more sequences provided in FASTA format) and CountRegexGFF_IUPAC_1input_simple_output.pl (returns positions of all occurrences of a given IUPAC string within given FASTA sequences).
For constructing networks, lists of gene names assigned to ChIP-seq regions were uploaded into STRING [43]. The resulting networks were saved as *.txt files and then uploaded into Cytoscape (v. 2.7.0) with cumulative scores of coexpression, textmining, knowledge and experimental data as proximity criteria. yFiles→organic network layouts were applied and the positioning and graphic representation of nodes were adjusted manually for increased clarity.
For functional profiling of ELK1 binding regions identified by ChIP-seq, Gene Ontology was performed using GREAT v. 1.2.6 [18] with the default (basal plus extension) gene regulatory definition option. Here, each ChIP-seq region is assigned to all genes whose domains it overlaps. Significance was determined by FDR<0.05, and terms significant by both the region-based binomial test and gene name-based hypergeometric test are reported. Gene Ontology analysis of lists of gene names was performed using DAVID v. 6.7 [21] and terms with P-values<0.05 were retained. In each case, Official Gene Symbol was used. Gene Ontology of networks was carried out using the Cytoscape plugin BiNGO (v. 2.42) [44] with default settings.
Statistical analysis for qRT-PCR studies and ChIP assays were performed using paired, 2-tailed Student's t test. The error bars in all graphs represent standard deviation. Fisher's Exact tests and Chi-square tests were used to compare two different distributions of pairs of datasets with respect to a particular condition. Z-score analysis of overlaps between two gene name lists was performed using a PERL script using 10000 background lists of Official Gene Names of size of one of the comparator datasets. Average overlap with standard deviation was recorded and z-score was calculated as the ratio of difference between actual overlap and this average overlap over the standard deviation. Kolomogorov-Smirnov tests were used to determine the statistical significance of differences in shape and range between two distributions. The Genomic HyperBrowser was used to determine the statistical significance of overlaps of sets DNA regions with the following settings: regions of dataset 1 fixed, region sizes and number of dataset 2 fixed, tested hypothesis: overlaps more than expected, overlap tested for the whole genome, with each chromosome treated as a separate bin (maximum possible bin size).
The ChIP-seq and microarray expression data have been submitted to array express (accession numbers E-MTAB-830 and E-MEXP-3407) and the data will be released upon publication.
|
10.1371/journal.pcbi.1006238 | Tox_(R)CNN: Deep learning-based nuclei profiling tool for drug toxicity screening | Toxicity is an important factor in failed drug development, and its efficient identification and prediction is a major challenge in drug discovery. We have explored the potential of microscopy images of fluorescently labeled nuclei for the prediction of toxicity based on nucleus pattern recognition. Deep learning algorithms obtain abstract representations of images through an automated process, allowing them to efficiently classify complex patterns, and have become the state-of-the art in machine learning for computer vision. Here, deep convolutional neural networks (CNN) were trained to predict toxicity from images of DAPI-stained cells pre-treated with a set of drugs with differing toxicity mechanisms. Different cropping strategies were used for training CNN models, the nuclei-cropping-based Tox_CNN model outperformed other models classifying cells according to health status. Tox_CNN allowed automated extraction of feature maps that clustered compounds according to mechanism of action. Moreover, fully automated region-based CNNs (RCNN) were implemented to detect and classify nuclei, providing per-cell toxicity prediction from raw screening images. We validated both Tox_(R)CNN models for detection of pre-lethal toxicity from nuclei images, which proved to be more sensitive and have broader specificity than established toxicity readouts. These models predicted toxicity of drugs with mechanisms of action other than those they had been trained for and were successfully transferred to other cell assays. The Tox_(R)CNN models thus provide robust, sensitive, and cost-effective tools for in vitro screening of drug-induced toxicity. These models can be adopted for compound prioritization in drug screening campaigns, and could thereby increase the efficiency of drug discovery.
| Visualization of nuclei using different microscopic approaches has for decades allowed the identification of cells undergoing cell death, based on changes in morphology, nuclear density, etc. However, this human-based visual analysis has not been translated into quantitative tools able to objectively measure cytotoxicity in drug-exposed cells. We asked ourselves if it would be possible to train machines to detect cytotoxicity from microscopy images of fluorescently stained nuclei, without using specific toxicity labeling. Deep learning is the most powerful supervised machine learning methodology available, with exceptional abilities to solve computer vision tasks, and was thus selected for the development of a toxicity quantification tool. Two convolutional neural networks (CNN) were developed to classify cells based on health status: Tox_CNN, relying on prior cell segmentation and cropping of nuclei images, and Tox_RCNN which carries out fully-automated cell detection and classification. Both Tox_(R)CNN classification outputs provided sensitive screening readouts that detected pre-lethal toxicity and were validated for a broad array of toxicity pathways and cell assays. Tox_(R)CNN approaches excel in affordability and applicability to other in vitro toxicity readouts and constitute a robust screening tool for drug discovery.
| Toxicity is a major cause of failure in drug development and causes costly withdrawals of drugs from the market. Drug development productivity would be greatly improved if cytotoxic compounds were identified during early in vitro screening [1–4]. Drug-induced cytotoxic effects lead to changes in cell and nuclear morphology which are characteristic of the specific cell-death pathway involved, the best characterized being apoptosis, necrosis, and autophagy [5–7]. The field has advanced with the establishment of high content screening (HCS) techniques and the emergence of toxicity reporters revealing specific biochemical pathways triggered during cell death programs or measuring metabolic cell function [8–10]. However, toxicity reporters are often limited to assess the specific biochemical pathways for which they were designed [11–13], and they are thus unlikely to capture the wide variety of toxic effects that can be triggered by different drugs in screening campaigns. Toxicity-screening approaches have combined multi-parametric image analysis of fluorescently labeled nuclei with the use of toxicity reporters in advanced machine learning pipelines [14–18]. However, toxicity reporters increase experimental complexity, thus reducing throughput and increasing screening costs. There is therefore an urgent need to develop broad specificity, cost-effective in vitro toxicity assays for incorporation in the primary screening phases of drug development. Cytotoxic effects have classically been visually identified from cell and nuclear morphology [5–7]. However, the complexity and variability of toxicity-associated morphological patterns has so far hindered their systematic and quantitative analysis and thus prevented their use as standalone toxicity screening endpoints. Although nuclear fluorescence staining forms the basis of most high content cell-based assays, its use is normally limited to image segmentation and nuclei counting to score cell-loss due to lethal toxicity [19,20], thus disregarding the wealth of information contained in images of fluorescently labeled nuclei. In an effort to exploit this information for the quantification of pre-lethal toxicity, we have explored state-of the-art machine learning tools for automated pattern recognition. The success of classical learning-based computer vision methods relies heavily on extraction and selection of a reasonable set of relevant features that are highly discriminative of the phenotypes being studied. Feature extraction and selection requires in-depth knowledge of the phenotype under study, which is hindered in the current application by the complexity and great variety of drug-induced toxicity-associated nuclei organization patterns. The most recent major advance in machine learning is deep convolutional neural networks (CNN) which, similar to a brain, have multiple layers of interconnected artificial neurons [21,22]. Through an automated process, deep neural networks learn abstract representations of raw images from pixel information as a progressive hierarchy of sub-images, from which they extract features that can be used to classify complex patterns in a supervised manner. CNNs can thus automate the critical steps of feature extraction and selection by learning to extract high-level features based on spatial relationships, which has enabled them to outperform other machine learning methods in computer vision tasks, as demonstrated for several challenging biomedical applications [23–28]. Thus, deep technology seemed well suited to the analysis and prediction of drug toxicity in images of fluorescently labeled nuclei.
Here, we present novel deep-learning approaches for in vitro cell-based toxicity assessment. The Tox_(R)CNN approaches proved to efficiently predict a broad spectrum of toxicity mechanisms from different drugs, nuclear stains and cell lines. The main strength of these tools is their unique ability to predict toxicity based exclusively on nuclei staining; this offers the advantage of improved affordability and applicability of toxicity prediction. The attraction of the Tox_(R)CNN tools relies on their high potential to enable sensitive and efficient compound prioritization based on detection of pre-lethal toxicity in primary screening campaigns.
To test if CNNs can predict cell toxicity based exclusively on nuclear staining, we designed an experimental assay in which HL1 cells were treated with reference compounds at different concentrations. The included reference compounds cover a range of cytotoxic effects: the DNA targeting genotoxic drugs cyclophosphamide (Ciclo), 5-fluorouracil (5Fluo), and doxorubicin (Doxo); the apoptosis-triggering drug staurosporine (Staur); the enzyme inhibitors acetaminophen (Aceta) and sunitinib (Sunit), which are not associated with a specific toxicity mechanism; the uncoupler of mitochondrial oxidative metabolism FCCP; and the microtubule stabilizer Taxol, which inhibits mitosis. To guarantee varying degrees of toxicity outcome we used previously empirically established dose-curves by selecting concentrations of drugs ranging between having no effect up to significantly reducing cell number due to moderate cell death. Cells were labeled with the DNA-specific fluorescent probe DAPI and imaged with an automated confocal microscope, revealing a great variety of nuclear patterns induced by the different compounds (Fig 1A). The standard toxicity readout of nucleus count (Num Nuc) revealed cell loss due to drug-induced lethal effects, but did not reveal the great variety of toxic effects associated with the reference compounds assayed (Fig 1B). As reference toxicity readouts, we analyzed Caspase 3/7 nuclear translocation and Mitotracker cytoplasmic intensity (Fig 1C and 1D), both of which evidenced dose-dependent toxic effects promoted by the compounds tested. To assess toxicity independently of cell density and to enable detection of pre-lethal toxicity, we implemented a deep CNN architecture for the estimation of cell health status from microscopy images of DAPI-stained nuclei (Fig 1E). Since we are aiming at a cell-based toxicity assessment, we used standard image analysis procedures to segment nuclei and cytoplasm according to the DAPI signal and used the segmentation to crop images (see Materials and methods). Different image cropping strategies were designed based on the regions of interest included in the resulting image crop; nuclei (Nuc), nuclei and cytoplasm (Cell), nuclei and 3 adjacent pixels (Nuc_Ring), and nuclei, cytoplasm, and background (CNN All) (Fig 1F). Cropped images, each one containing a mass-centered cell, were used to train independent CNN models for making toxicity prediction (CNN Nuc, CNN Cell, CNN Nuc_Ring and CNN All). An additional model was trained using the combination of images obtained using the different cropping strategies (CNN 4crops). We repeated the training of the models (CNN Nuc, Nuc_Ring, Cell, All, and 4crops) 5 times each, in order to increase confidence in model comparison results and evaluate reproducibility. CNN models deliver a “health status” score as output for each cell, which determines a binary classification: healthy or toxicity affected. As a supervised learning technique, the CNN required images labeled according to the expected output (healthy or toxicity affected) as ground-truth images for training. However, the reference standards analyzed for this purpose, Caspase 3/7 and Mitotracker, did not qualify as general toxicity labels because neither efficiently captured all the drug-induced cytotoxic effects included in our experimental assay: both yielded poor resolution of Taxol-induced cytotoxicity and of the kinetic effects produced by Doxo and Aceta. As an alternative strategy, we produced a training dataset by labeling image crops according to the treatment exposure; cells from untreated wells were labeled healthy, whereas those from wells treated with the highest drug concentrations were labeled toxicity affected (S1A Fig). The percentage of cells classified as healthy by the different CNN models served as the per-well measurement of “general” toxicity. In spite of the high degree of uncertainty introduced into the impure training set, the CNN healthy predictions efficiently revealed dose-response toxicity curves for all the drugs tested in the assay (Fig 1G and S1B–S1E Fig), providing better resolution than those obtained from Num Nuc or the Caspase or Mitotracker readouts (Fig 1B–1D). At high doses of several toxicants, CNN models trained with information from nuclei only (CNN Nuc) underwent an unexpected drop in toxicity prediction. This was not observed in CNN models trained with cytoplasmic information (CNN Cell/Nuc_Ring/All/4crops), which displayed a steady increase in toxicity prediction due to their enhanced ability to reveal DNA release into the cytoplasm during necrosis induced by high drug concentrations.
As screening readout, Z-scores were obtained from the percentage of cells classified as healthy by the CNN model for each compound-treated test well by normalizing to the DMSO-treated control cells (DMSO). Untreated cells were not used as negative controls since these are not commonly included as controls in screenings. To allow comparative evaluation of the different toxicity readouts and benchmarking the different cropping strategies, readouts are displayed so that Z-score positive values depict toxic effects. CNN readouts were more efficient at predicting toxicity than the classical Num Nuc readout (Fig 1H), with CNN Nuc displaying the highest Z-scores. Model performance was assessed on cells treated with the apoptosis-inducer staurosporine, which allowed the use of Caspase readout for ground truth labelling (i.e., Caspase-negative cells from untreated wells were labelled healthy whereas Caspase-positive cells from staurosporine-treated wells were labeled as toxicity affected). This test set was used for computing Receiver Operating Characteristic (ROC) curves to evaluate model prediction accuracies using the Area Under the ROC Curve (AUC) measurement. Results show high performance (above 0.9AUC) for all cropping strategies (Fig 1I), being the CNN Cell the most accurate. Importantly, CNN Nuc predictions were less correlated with the number of cells in the field than the other cropping strategies (Fig 1J), which displayed non-negligible correlations, suggesting that predictions of the health status of a single cell are influenced by surrounding information, including the presence of neighboring cells. As consequence, CNN Nuc reveals to be more independent of experimental errors due to cell density inconsistencies and also more indicated to detect pre-lethal toxicity. Accordingly, the CNN Nuc model proved to be the best readout for early prediction of toxicity, since it outperformed other cropping strategies at sub-lethal drug concentrations, where there is no reduction in Num Nuc indicating significant cell loss. Consistent with this pattern, plotting treatments by Z-scores to reveal above-threshold “toxic hits” confirmed CNN Nuc to be the most sensitive method for detecting early toxicity yielding 100 toxic hits (rounded mean of results from the 5 CNN Nuc models) out of 184 treated wells (Fig 2A), thus outperforming other CNN models. CNN Cell, Nuc_Ring, All and 4crops readouts yielded 82, 74, 77 and 83 hits, respectively (Fig 2B and 2C and S1A and S1B Fig). The standard toxicity readouts Num Nuc, Caspase and Mitotracker detected 27, 81, and 6 toxic hits, respectively (Fig 2D–2F). All CNN models yielded significant Z-scores for Taxol at 0.1μM, demonstrating broader application than established readouts; Caspase detected toxicity effects only at 2μM Taxol, whereas Mitotracker and Num Nuc did not detect significant Taxol toxicity. The higher sensitivity of CNN Nuc compared with the other CNN models and established readouts was further evidenced by the lower half-maximal toxic concentration values (EC50) obtained for the mild toxicant 5Fluo (Fig 2 and S2 Fig). Based on these results, nuclear crops were used as inputs for CNN models in all subsequent studies, and from here on CNN refers to these CNN Nuc models.
To avoid relying on external image segmentation and cropping procedures while providing a per-cell toxicity prediction, we undertook an alternative cell-based deep-learning approach incorporating automated object detection. An RCNN was implemented for the automated localization and classification of individual nuclei using raw images as input, instead of crops. The framework, based on the Faster RCNN algorithm [29], includes a region proposal network (RPN) that uses features extracted from the last convolutional layer of a CNN to detect bounding boxes around individual candidate cell, which are then classified as healthy, toxicity-affected, or background (Fig 3A and 3B). We trained the RCNN model with cell bounding box coordinates obtained with the standard segmentation procedure used in the CNN approach (see Materials and methods). A set of 7 independent experiments in HL1 cells treated with the eight reference drugs, including two experiments in which untreated cells were cultured at different confluencies, was used for in-parallel training of the RCNN and CNN mixed models for toxicity prediction using balanced datasets of healthy and toxicity affected labeled nuclei (Tox_CNN and Tox_RCNN_balanced). Resulting CNN predictions showed that the Tox_RCNN_balanced model, unlike Tox_CNN, erroneously predicted toxicity of untreated cells grown at low densities (Fig 3C). To prevent the models from learning to predict toxicity as a reduction in cell number due to drug-induced lethal effects, we trained an additional mixed model that included extra images of untreated cells cultured at different densities (120 extra healthy training wells), hereafter referred to as Tox_RCNN. The unbalanced and balanced Tox_RCNN models (Num Nuc RCNN and Num Nuc RCNN_balanced) detected a similar number of objects (nuclei); moreover, the number was consistent with the number obtained by the standard segmentation procedure (Num Nuc) at regular cell densities (Fig 3C–3E), demonstrating the efficiency of automated detection by RCNN. Tox_RCNN slightly overestimated cell number at low confluencies, probably due to the recognition of cellular debris that are discarded by the image processing procedure. Both Tox_CNN and Tox_RCNN mixed models successfully classified untreated cells at very low densities as healthy and efficiently predicted the toxic effects of drugs and high DMSO concentrations (Fig 3C), further demonstrating their independence from cell-density fluctuations. Tox_(R)CNN models performed efficiently in the test wells from an experiment used for training (Fig 3D) and in one independent experiment with HL1 cells at higher confluency (S3A Fig). Overall, Tox_(R)CNN models were consistently more sensitive than Num Nuc at predicting drug toxicity, with Tox_CNN outperforming Tox_RCNN classification in most cases. Even though these models were trained in HL1 cells, they suitably predicted toxicity from these drugs in two other cell lines, EAHY926 (Fig 3E) and MEVEC (S3B Fig), thus confirming the applicability of Tox_(R)CNN models for the prediction of toxicity in different cell types. In addition, these models successfully predicted toxicity of HL1 cells labelled with Hoechst 3342 (S4 Fig). Together, these results demonstrate the robustness of deep-learning-based toxicity prediction with regard to inter-experimental and intra-experimental variability, thus confirming Tox_(R)CNN as powerful screening tools.
To demonstrate the value of these Tox_(R)CNN models as tools for broad toxicity prediction, we performed a new screening in which HL1 cells were treated with a panel of 24 drugs, including those used to train the CNN model and additional drugs acting through several mechanisms. Tox_(R)CNN mixed models sensitively predicted the outcome of toxic compound-treatments, thus proving their ability to reveal the toxicity of compounds for which they have not been trained (Fig 4A). Conveniently, Tox_CNN enabled the automated extraction of features from pixel intensity values, which were used for unsupervised hierarchical clustering of compounds (see Materials and methods). Tox_CNN features clustered compounds with known mechanisms of toxicity associated in a biologically meaningful manner, even though the models were not trained for this purpose (Fig 4B). The ionophores FCCP and monensin, which produce ROS and mitochondrial toxicity clustered together. The DNA synthesis inhibitors 5Fluo, gemcitabine, and mitomicine also group in the same cluster. Apoptotic death inducers (staurosporine, thapsigargin, bortezomib, and imatinib) were clustered closely together with other drugs of unknown mechanism. The DNA intercalating anthracyclines epirubicin and doxorubicin and the topoisomerase II inhibitor Etoposide, all of which promote double strand breaks also clustered together. Other drugs included, such as microtubule modulators Taxol and Vinblastine did not cluster together. Statins (lovastatin and simvastatin) with yet unknown mechanism of toxicity, but expected to be similar because belonging to the same family of proteins, were clustered together. These findings confirm not only that deep CNN are able to perceive general toxic effects, but also that their ability to learn feature representations provides useful knowledge for comparing drug-induced mechanisms of toxicity.
To further evaluate the Tox_(R)CNN deep-learning models as screening tools for prioritizing compounds based on their toxicity potential, we re-analyzed a pre-accomplished HCS of primary pancreatic cancer associated fibroblasts (pan-CAFs). Among several assay-specific labels, most of which are irrelevant here, this HCS included DAPI staining in the assay for both image segmentation and nuclei counting as a toxicity endpoint. A transfer learning strategy was applied to the Tox_(R)CNN models delivering Tr_Tox_(R)CNN models. Training strategy was designed to allow the use of pre-run screens lacking reference toxicity-inducing treatments (see Materials and methods). In brief, the training dataset was produced from images from drug-treated wells with a significantly reduced cell number, which were labeled toxicity affected, while cells from DMSO treated wells were labeled healthy, since no untreated cells were available for training. Interestingly, compounds #19 and #33 (anagrelide and quercetin) were negligibly lethal according to nuclei counting, but were predicted by Tox_CNN to be toxic at high concentrations (Fig 5A and 5B), demonstrating the greater sensitivity of deep-learning-based prediction compared with Num Nuc. Toxicity of these compounds was confirmed by re-testing in primary cardiac fibroblasts (S5 Fig). Even though Tr_Tox_CNN performed better that the Tox_CNN model, the latter was still a more sensitive predictor of toxic effects than nuclei counting, demonstrating the value of these tools. Nuclei counting by both the transferred and original Tox_RCNN models (Tr_Tox_RCNN and Tox RCNN) was consistent with the standard procedure (Num Nuc), revealing that object detection was performing adequately. However, the original Tox_RCNN model was a poor toxicity predictor in pan-CAFs compared with the transferred RCNN model (Tr_Tox_RCNN), further evidencing the need for a transfer-learning approach for RCNN toxicity predictions in cell lines different from those used for training. The screening comprised 60 drugs at 8 concentrations (480 treatments) and yielded 102 toxicity hits (mean Z-score > 3) based on nuclei counting (Fig 5C). In contrast Tr_Tox_CNN and Tr_Tox_RCNN identified 127 and 126 toxic wells, respectively (Fig 5D and 5E). Z-scores computed for the transferred Tr_(R)CNN models were plotted independently for all compounds screened (S6 Fig). These results further demonstrate the superior performance and sensitivity of deep Tox_(R)CNN toxicity predictions over classical screening endpoints based on nuclei staining.
There is a need to incorporate highly predictive toxicity assays into primary in vitro high-throughput screening in order to reduce the attrition of drug candidates at later phases of the drug discovery pipeline. In vitro cytotoxicity assessment is normally limited to measuring the number of viable cells per well. However, single-cell readouts make better outputs that avoid sources of experimental error such as non-homogeneities in cell-dispensing, drug-induced proliferative effects, and heterogeneous responses of different cell sub-populations, which could be misinterpreted if only well averages are examined. The toxicity research field has therefore been directed towards finding novel cell labels and readouts that distinguish between different cytotoxicity mechanisms [8–13]. Nevertheless, the use of toxicity reporters has not gained broad acceptance because it adds experimental complexity, thus reducing throughput and increasing screening costs. Here, we have established tools that predict cell toxicity based on the analysis of fluorescently labeled nuclei. These tools outperformed outputs that rely on toxicity reporters or cell counting. Since nuclei staining is common to most high content cell based assays, this tool has broad applicability for toxicity prediction in HCS, even in pre-accomplished screens, as demonstrated here.
Over recent years, deep learning approaches have been successfully deployed in computer vision tasks and constitute the state-of-the-art tools for supervised machine learning. Several key features give deep learning approaches an advantage over other machine learning methodologies for toxicity analysis. First, by learning to represent data with multiple levels of abstraction in an unsupervised manner (i.e. without human-based programming) it avoids cumbersome knowledge-based feature engineering. This makes deep learning approaches independent of prior in depth knowledge of the target phenotypes, and therefore more suitable for broad toxicity prediction of multiple drugs with differing mechanisms by HCS. Second, their ability to learn intricate patterns improves recognition, feature extraction, and classification from noisy images, providing accuracy. This has led to deep technology outperforming other machine learning methods and human-based analysis, as demonstrated for several challenging biomedical applications [23–28]. Third, using transfer learning methods, networks can be updated to classify new datasets with limited training data [26,27,30], and thus makes them suitable for predicting toxicity in pre-accomplished screenings. Accordingly, the deep-learning approaches presented here successfully predict toxicity of a broad spectrum of toxicity-associated patterns in HCS images of fluorescently labeled nuclei, proving to be more sensitive than established toxicity reporters and readouts for the recognition of pre-lethal toxicity. The Tox_(R)CNN models suitably predicted toxicity in cell lines, nuclear stains and for compounds different from the ones used for training. Moreover, for very different cell lines, pre-trained Tox_(R)CNN models can be subject to a simple transfer-learning approach that does not require toxicity controls, increasing the performance of toxicity predictions; this was successfully achieved here in the pre-accomplished screening with primary pancreatic cancer associated fibroblasts (panCAF). Although CNN models were more sensitive and transferable and performed better, the use of RCNN models is justified by their independence from external segmentation and image cropping. We also demonstrate the utility of CNNs for extracting knowledge (as feature maps) that allows comparative analysis of different drugs in a screen. Previous cell-based toxicity screening approaches have combined multi-parametric image analysis of fluorescently labeled nuclei with the use of toxicity reporters in advanced machine learning pipelines [14–18]. The main strength of the tools presented here is their unique ability to predict toxicity based exclusively on nuclei staining, which offers the advantage of improving affordability and applicability of toxicity prediction. HCS is now an established tool for phenotypic drug discovery; in this setting, the deep-learning approaches presented here will promote a better use of HCS technology for toxicity assessment. The relevance of the deep learning approaches presented here relies on their high potential to enable sensitive and efficient compound prioritization based on detection of pre-lethal toxicity. They thus provide affordable cytotoxicity counter-screens for high throughput primary campaigns and could allow academic screening centers and pharma companies to discard cytotoxic compounds during primary screening and hit-to-lead drug development campaigns, thereby increasing the efficiency of drug discovery.
Mouse cardiac muscle HL1 cells purchased from Merck Millipore were grown on fibronectin(25μg/ml)/gelatin(1mg/ml) coated dishes with 10% fetal bobine serum (FBS) in Claycomb medium (Sigma-Aldrich). Mouse embryonic ventricular endocardial cells (MEVEC) were kindly provided by Dr. de la Pompa [31] and cultured on 0.1% gelatin coated flasks in 10% FBS supplemented DMEM. EAHY926 cells were kindly provided by Dr. Edgell and maintained in 10% FBS supplemented DMEM. PanCAF were obtained from Dr. Hidalgo and cultured in RPMI with 20% FBS. Primary pig cardiac fibroblasts were isolated from fresh surgical samples by collagenase tissue digestion as described in [32]. Fibroblasts were grown to 80% confluency in flasks containing supplemented DMEM. All culture media were supplemented with 10% FBS (except PanCAFs, which had 20% FBS), 100 U/ml penicillin, 100 μg/ml streptomycin and 2 mM L-glutamine and refreshed every 2–3 days. Mycoplasma tests were performed bimonthly for all cell line cultures.
Fluorescence staining reagents including DAPI (4′, 6-diamidine-2-fenilindol), Hoechst 33342 (H42), Mitotracker Orange and Cell Event 3/7 Caspase Green, were purchased from Invitrogen. Dimethyl sulfoxide vehicle (DMSO); Acetaminophen (ACETA); Doxorubicin (DOXO); carbonylcyanide p-trifluoromethoxyphenylhydrazone (FCCP); Sunitinib; Staurosporine (STAUR); Paclitaxel (TAXOL); Imatinib; Thapsigargin; Gemcitabine; Quercetin; Atenolol; Simvastatin; Genistein; Vinblastine; Monensin; Anagrelide; Epirubicin; Etoposide and Lovastatin were from Sigma. Ciclophosphamide (CICLO), 5-Fluorouracil (5FLUO)and Sunitinib malate (SUNIT) were from Tocris. Indacaterol and Bortezomib were kinldy provided by Dr. Blanco, Experimental Therapeutics Programme at CNIO.
Cells were seeded on 384-well plates (5000 cells/well, otherwise specified), after 24h compounds were added to wells at N serial concentrations in 4 replicate wells. Plate was designed to avoid well distribution effects by using layout shown in S1A Fig, which places negative controls oriented asimetrically in distant positions within the plate (usually center and right). Additionally, in experiments designed for training deep learning models, double of wells with extreme doses were included in the plate simetrically distributed at top and bottom locations. To avoid evaporation-related edge effects external rows were filled with PBS and not used for the assay. Compounds were dissolved in DMSO (final concentration of 1% across the entire assay, otherwise specified). Cells were maintained in culture with compounds for 24h, and then stained for Caspase 3/7 and/or MitoTracker prior fixation with 4% PFA for 10 min at RT and final nuclei staining. Imaging of fluorescently labelled nuclei and toxicity probes was performed with an Opera automated confocal microscope (Perkin Elmer) fitted with an NA = 0.7 water immersion objective at a magnification of 20×. S1 Table summarizes all experiments used in this work, including information about cell lines, stainings, treatments, and number of images and cells.
Image processing algorithm was developed using Definiens Developer version XD2.4 (Definiens AG, Germany). Nuclei and cytoplasmic regions were first segmented based on differential contrast of DAPI/Hoechst 33342 intensity. Cells with a nuclear size bellow 0.3 times or over twice the mean of nuclear sizes per field were considered as debris and reclassified as non-cellular regions. An ID was univocally assigned to each cell the number of cells per well was computed. Toxicity readouts based on fluorescent reporters were extracted as Caspase 3/7 nucleus:cytoplasm intensity ratio (Casp nuc/cyto), and Mitotracker mean intensity in the cytoplasmic region (Mito), where specified.
Images from DAPI/Hoechst 33342 stained nuclei were cropped and saved individually assuring one cropped image per single mass-centered cell, thus conserving univocal cell ID. Four different strategies of cropping images were established (Fig 1F):
The four strategies were set to a fixed size of 50x50 pixel crops, thus guaranteeing a proper inclusion of complete nuclei based on nuclear sizes and image pixel sizes. To avoid off-centered nuclei crops, those nuclei with a distance to field image boundaries of less than 50 pixels were excluded from the cropping extraction and further analysis. Additionally, bounding box coordinates from segmented cells were also extracted for training automatic object detection by RCNN.
We designed the toxicity convolutional neural network with an architecture and parameters adapted from a state-of-the-art CNN model, VGG [33], with high performance in image classification and well-known layers, activations, and initializations, including several 3x3 convolutions before a max-pooling layer. However, the limited size of our image crops (50x50 instead of 224x224 used by VGG) required a reduction in the depth of the network to prevent overfitting. Inspired by state-of-the-art architecture LeNet-5 [34], which was designed to work with 32x32 images, we constrained the network to 2 convolutional + max pooling groups of layers. Most of parameters were inherited from VGG model (batch size, initial learning rate, weight initialization…), since the lack of a general reference standard did not allow us to perform systematic optimization of parameters. Number of epochs was set up to 120 since performance of the network reached stability (hold-out validation using 33% of training data during 300 epochs with batch size of 256) while maintaining a reasonable computing time and preventing from potential overfitting. Kernel sizes of 3 and 5 were also tested reaching very similar results (AUC values for caspase-based evaluation were 0.9513±0.0065 and 0.9509±0.0053 for CNN Nuc model with kernel sizes of 3 and 5, respectively; p-val = 0.9), so we fixed them to 3 such as in original VGG architecture, which also speeded up computations x1.63. Final Tox_CNN network architecture comprises 4 convolutional and 2 fully connected layers (Fig 1E) to classify single-cell images as healthy or toxicity affected. The Rectified Linear Unit (ReLU) activation function is applied between each layer except the output dense layer, which uses a softmax activation function to provide a separate probability for each of the classes. We used ReLu as activation function since Sigmoid and Tanh can result in the so-called vanishing gradient problem. Convolutional layers convolve a 3×3 kernel over some input to produce 32, 32, 64, and 64 feature maps, respectively. To reduce the number of features and the computational complexity of the network, we introduced two max-pooling layers with a window size of 2×2 after convolutional layers 2 and 4. Additionally, to avoid overfitting, we included two dropout blocks after convolutional layers 2 and 4 (probability of 25%), and another one next to the first fully connected layer (probability of 50%). Dropout deactivates some neurons randomly with a specified probability during the weight update cycle. The final max-pooling layer is then flattened and followed by two densely connected layers with 512 and 2 features. Finally, we applied a softmax activation function to the output of last fully connected layer to calculate the probability for each class label. The total number of parameters to learn is equal to 4,031,458, most of them belong to the first fully connected layer. We used ADADELTA algorithm [35] to adjust the learning rate automatically. To increase the number of data and avoid overfitting, we augmented images by applying random rotations in the range of [0°, 20°], horizontal shifts in the range of [0, 0.2 × Imagewidth], vertical shifts in the range of [0, 0.2 × Imageheight], and horizontal/vertical flips, where Imagewidth and Imageheight show width and height of input images, respectively. By default, the modifications were applied randomly, so not every image will be changed every time. Images were normalized to zero mean and unit variance before feeding them into the network.
We used state-of-the-art Faster Region-Based CNN [29] (RCNN) to both detect and classify cells as healthy or toxicity affected from entire images. Faster RCNN is composed of two modules: a Regional Proposal Network (RPN) and a RCNN network. RPN is a Fully Convolutional Network (FCN) [36] which proposes square regions within an image that may contain objects of interest without considering their classes while RCNN network classifies the object proposals from RPN into one of the classes (or background), and refines the bounding boxes’ coordinates of the final proposals. We used the original Faster RCNN architecture [29] without any significant modifications except for the number of outputs for the classification and regression layers since we have two classes. Therefore, the classification layer has 18 (9×2) outputs, and the regression layer has 36 (9×4) outputs (coordinates). Number of iterations was fixed to 750,000.
To train the network for single-cell toxicity prediction, a set of cells need to be labelled as healthy and toxicity affected. The uncertainty of the toxicity state of individual cells due to lack of bona-fide toxicity reporters hamper the possibility of creating a pure and clean training dataset. Therefore, cells were labeled according to the known or expected toxic response of different extreme treatments; a set of untreated cells (or cells under harmless treatment) were labelled as healthy, and a set of cells treated with known toxic compounds at the highest concentrations were labelled as toxicity affected. This labelling strategy minimize the amount of manual supervision needed to perform the tedious task of creating a large annotated dataset for training, avoiding also the use of any toxicity labels such as fluorescent reporters that always provide partial information since they are unable to detect all the toxic effects. These healthy and toxicity affected sets are initially conformed by all cells in a selection of wells correspondent to the appropriate extreme treatments. Training sets for both classes (healthy and toxicity affected) were balanced, where indicated, by removing all cells from randomly selected entire images. The final output is a field-based treatment-driven training dataset that represents two groups/classes with the (expected) highest mean difference in terms of toxicity state. A training-strategy avoiding overfitting was undertaken to properly deal with minimized, yet still present mislabeling of cells affected by toxicity in conditions with harmless or no treatment, as well as resilient cells in extreme harmful conditions.
Training of both Tox_CNN and Tox_RCNN models need the outputs obtained from the image processing routine developed in Definiens (see Image processing section). CNN approach used cropped single cell 50 × 50 pixel images; and RCNN approach used full fields (683 x 507 pixel images) and bounding-box coordinates of cells. The initial experiment pursuing the comparison of CNN performance from the different cropping strategies used images from one experimental plate (Experiment #1), where the correspondent trainings were performed in parallel and repeated five times. The creation of central Tox_(R)CNN mixed models involved images from 7 HL1 experiments (Experiments #1–7) including several conditions: untreated cells, cells treated with DMSO, and cells treated with up to 8 drugs with known toxic effects at different concentrations. All drugs used demonstrated toxic effects in previous experiments where dose curves were fixed. Three of these experiments were designed in a way that allows the classifier to learn that healthy cells can also grow at low densities. Training dataset was created as detailed in the previous section, labelling single cells from wells without any treatment as healthy, and cells from treated wells with highest concentrations of the 8 available compounds as toxicity affected. In each plate, only half of these wells per condition were sparsely selected for training, and the rest were bound to test (S1A Fig). In total, Tox_CNN mixed model was trained using 739,727 cells (image crops) covering 8 compounds (toxicity affected) and untreated cells (healthy). Bounding-box coordinates of training cells, together with the correspondent 7,489 (balanced) and 10,883 (unbalanced) entire images (fields) were used to train the Tox_RCNN mixed models. S2 Table summarizes the number of instances (crops or field images) and experiments used for training each model.
For CAF screening, we used deep transfer learning [37] to adapt Tox_(R)CNN mixed models to a different cell line improving toxicity prediction. With this strategy, an existing pre-trained network is fine-tuned, avoiding training an entire neural network from scratch and reusing low level feature-detectors already learned. Therefore, we froze the weights of the first two convolutional layers in both (R)CNN approaches and retrained the rest of the layers to adapt to this new dataset. For this screening, since there is no prior information about the dose-response curve and expected toxicity, we used a different strategy to create the training set, but again following the guidelines detailed before (see Label-free toxicity annotation of images). First, we selected drugs with at least one concentration with a significant toxic effect scored as a significant reduction of the number of cells per well (Z-score>3). Then, for each selected drug, cells treated with the two highest concentrations of that drug were included in the training set, and labelled as toxicity affected. Since untreated wells are not included in regular screenings, cells from half of the DMSO-treated wells were included in the healthy training set, which corresponds to the harmless condition in this specific screening. This resulted in a training set with 150,529 instances (6,057 field-images). S2 Table includes information about the number of instances (crops or field images) and experiments used to create the transferred models. We followed this general strategy for generating training sets, to ensure that it can be used in any assay where no prior information about toxic effects of compounds included in the screening is available. We used 25 epochs to re-train the Tox_CNN model, and 45,000 iterations to update the weights of the Tox_RCNN model; conforming the transferred models for toxicity prediction in PanCAF screening (Tr_Tox_(R)CNN models).
We evaluated the performance of Tox_(R)CNN models on several independent experiments. Cells out of the training set coming from experiments that were partially used for training were also employed for testing purposes: treatments with intermediate concentration of drugs were never included in training set, and not all wells from extreme treatments were selected for training. Tox_CNN model classified crop images at the input as healthy or toxicity affected based on the probabilistic scores obtained at the output for both classes. Tox_RCNN model return object detections which were classified as healthy, toxicity affected or background (considered as non-cell detections and discarded from further analyses) in entire field images. S3 Table summarizes experiments and number of instances (crops or field images) tested with the different models, and references to figures including the corresponding results.
Well-based toxicity measurements were constructed from Tox_(R)CNN predictions by computing the percentage of cells classified as healthy in each well. Standard toxicity measurements obtained from fluorescent reporters (Caspase 3/7 and Mitotracker) were aggregated in well-based values by computing the mean. Nuclei count obtained by image processing and RCNN-derived counting of nuclear detections were also reported for each well. For each type of toxicity measure (x), we computed Z-scores by subtracting the mean (μ) and then dividing by the standard deviation (σ) of negative control (DMSO treated wells):
Z-Score=x-μσ
Finally, we adjust the sign of the outputs to get increasing values for toxic effects, thus obtaining final well-based toxicity readouts that allow direct comparisons.
The previously explained uncertainty of the toxicity status of individual cells under the wide range of different compound treatments hampers the proper evaluation of CNN models for parameter optimization and comparison of models obtained with different cropping strategies (CNN Nuc, CNN Nuc_Ring, CNN Cell, CNN All, and the combination of all the previous, CNN 4crops). For that reason, assessment of model performance was tackled in different ways:
Tox_CNN features from the first fully connected layer (512 features/sample) were extracted for all cells in 4-plate screening of HL-1 cells with 24 drugs. Features obtained from cells that were either untreated, DMSO-treated, or treated with a non-extreme toxic concentration of the drugs (25 μM) (except for Taxol in which a 0.781 μM concentration was used) were analyzed by PCA to select the 50 most informative features for further processing. Mean well-based features were later normalized (Z-score with respect to untreated condition) and aggregated in order to obtain mean values per treatment/condition. Finally, these feature vectors were assembled by hierarchical clustering [44] to generate the hierarchical tree (clustergram function in MATLAB with Euclidean distance metric and weighted average linkage). Clusters along both dimensions (features and treatment/condition) were displayed in a heatmap of feature vector values including dendrograms representing the multilevel hierarchy obtained.
Half maximal effective concentration (EC50) is the concentration of a drug which induces a response halfway between the observed baseline and maximum effect of that drug. For the present work, the drug responses used for the EC50 calculations are the different toxicity effects, as indicated. EC50 were computed by fitting a Hill Equation sigmoid curve to the dose-response values and estimating the correspondent EC50 Hill Equation parameter [45]. Dose-response values and adjusted curves are displayed in Fig 2 and S2 Fig in such way that visual increasing values depict toxic effects for all toxicity measurements in order to allow proper comparison. Response axes are fixed to [1; 0] for CNN predictions, [4,000; 0] for nuclei count, [0.5; 3.5] for the Caspase ratio and [500; 100] for Mitotracker measurements. EC50 values were not calculated in cases where standard sigmoidal curve fitting was inappropriate within the existing range of drug doses (Fig 2E and 2F).
We used Keras [46] on the Theano backend [47] to develop the Tox_CNN model presented here. To construct the Tox_RCNN model we used a python implementation of Faster RCNN [29] which is developed on Caffe [48]. All experiments were run on a standard PC with an Intel Xeon CPU E5-2643 @ 3.30 GHz, 32 GB working memory, and using a 16 GB NVIDIA Quadro K4000 GPU to speed up the computations. Training time for the Tox_CNN model was 11 hours; transfer learning took 2 hours. Training time for the Tox_RCNN mixed model and transfer learning were 187 hours and 9 hours, respectively. Boxplots and well-based dot plots were created with NCSS Statistical Software (version 11), and mean Z-score plots were depicted with Python. MATLAB (R2017a) was used to perform hierarchical clustering and to compute dose-response curve adjustments and EC50 calculations.
All figures displaying (R)CNN-derived information include only results from test sets; otherwise indicated. Figures showing information from initial CNN models associated with different cropping strategies show either the results from 5 independently trained models or the results of a representative model/cropping-strategy performing with the median AUC value (out of five).
|
10.1371/journal.pcbi.1006611 | Top-down inputs drive neuronal network rewiring and context-enhanced sensory processing in olfaction | Much of the computational power of the mammalian brain arises from its extensive top-down projections. To enable neuron-specific information processing these projections have to be precisely targeted. How such a specific connectivity emerges and what functions it supports is still poorly understood. We addressed these questions in silico in the context of the profound structural plasticity of the olfactory system. At the core of this plasticity are the granule cells of the olfactory bulb, which integrate bottom-up sensory inputs and top-down inputs delivered by vast top-down projections from cortical and other brain areas. We developed a biophysically supported computational model for the rewiring of the top-down projections and the intra-bulbar network via adult neurogenesis. The model captures various previous physiological and behavioral observations and makes specific predictions for the cortico-bulbar network connectivity that is learned by odor exposure and environmental contexts. Specifically, it predicts that—after learning—the granule-cell receptive fields with respect to sensory and with respect to cortical inputs are highly correlated. This enables cortical cells that respond to a learned odor to enact disynaptic inhibitory control specifically of bulbar principal cells that respond to that odor. For this the reciprocal nature of the granule cell synapses with the principal cells is essential. Functionally, the model predicts context-enhanced stimulus discrimination in cluttered environments (‘olfactory cocktail parties’) and the ability of the system to adapt to its tasks by rapidly switching between different odor-processing modes. These predictions are experimentally testable. At the same time they provide guidance for future experiments aimed at unraveling the cortico-bulbar connectivity.
| In mammalian sensory processing, extensive top-down feedback from higher brain areas reshapes the feedforward, bottom-up information processing. The structure of the top-down connectivity, the mechanisms leading to its specificity, and the functions it supports are still poorly understood. Using computational modeling, we investigated these issues in the olfactory system. There, the granule cells of the olfactory bulb, which is the first brain area to receive sensory input from the nose, are the key players of extensive structural changes to the network through the addition and also the removal of granule cells as well as through the formation and removal of their connections. This structural plasticity allows the system to learn and to adapt its sensory processing to its odor environment. Crucially, the granule cells combine bottom-up sensory input from the nose with top-down input from higher brain areas, including cortex. Our biophysically supported computational model predicts that, after learning, the granule cells enable cortical neurons that respond to a learned odor to gain inhibitory control of principal neurons of the olfactory bulb, specifically of those that respond to the learned odor. Functionally, this allows top-down input to enhance odor discrimination in cluttered environments and to quickly switch between odor tasks.
| A key property of the mammalian brain that is essential for its vast computational power is the pervasiveness of centrifugal, top-down feedback from higher to lower brain areas [1–5]. The top-down projections can provide the receiving brain area with information that is not available in the feedforward stream [6] and can switch the processing by the lower brain area between different modes, as demonstrated in the visual system [7]. From a theoretical perspective, it has been posited that top-down signals can direct the lower brain area to suppress response to expected inputs or to inputs that have already been recognized by the higher brain area [8–11] and to transmit predominantly information about unexpected or unexplained inputs and the error in the prediction of the inputs [12], focusing on task-relevant information. Such specificity in the processing requires that the top-down projections be precisely targeted [10, 13–16]. The mechanisms that are at work in the formation of these specific connectivities are, however, not well understood. Experiments in the visual system suggest that they are activity-dependent [17]. Here we address this issue in the context of the extensive structural plasticity of the adult olfactory system.
In the olfactory bulb, which is the first brain area receiving sensory input from the nose, structural plasticity is not restricted to early development, but is also pronounced in adult animals. At that point its key players are the granule cells (GCs). They receive sensory input from the bulb’s principal neurons—mitral and tufted cells (MCs)—and are the target of massive top-down projections from the olfactory cortex [2–5]. Not only do the GC dendritic spines exhibit strong and persistent fluctuations [18, 19], but these interneurons themselves, which constitute the dominant neuronal population of the bulb, undergo persistent turnover through adult neurogenesis [20]. Both, spine fluctuations and the survival of the granule cells depend on the sensory environment [18, 21].
Functionally, adult neurogenesis is observed to enable and improve various aspects of learning and memory [22, 23]. A particularly clear example is the perceptual learning of a spontaneous odor discrimination task, which is substantially compromised if adult neurogenesis is suppressed [24]. The survival of GCs seems also to play an important role in retaining the memory of an odor task, with the survival of odor-specific GCs being contingent on the continued relevance of that odor memory [25]. This suggests that GCs receive non-sensory, task-related information via the top-down projections. These projections likely also underlie the GC activation that can be evoked by context alone, without the presence of any odor, if that context has previously been associated with an odor [26]. Importantly, the context-evoked GC activation patterns reflect specifically the GC activation pattern that would be induced by the associated odor.
Taken together, the experiments suggest that sensory input together with non-olfactory information like context or valence may shape the connectivity within the olfactory bulb as well as that of the top-down projections onto the GCs. It is, however, poorly understood how the bottom-up and top-down inputs into the GCs jointly shape the network connectivity, what mechanisms are at work, what kind of connectivities arise, and what kind of functionalities these connectivities support. To address these issues we have employed computational modeling using a biophysically supported framework. This model makes a number of predictions that can be tested with physiological and behavioral experiments and that can guide future experiments aimed at unraveling the cortico-bulbar connectivity.
We have developed a computational network model for the olfactory bulb and its bidirectional communication with a cortical area, which is based on key features of adult neurogenesis as observed experimentally. Details of the model are given in Methods. Briefly, the bulbar component of the model network comprised the two main neuronal populations of the olfactory bulb, the mitral/tufted cells (MCs) and the granule cells (GCs). The MCs are known to project from the olfactory bulb to a host of other brain areas, including the anterior olfactory nucleus, the anterior and the posterior piriform cortex, the olfactory tubercle, the lateral entorhinal cortex, and the cortical amygdaloid nucleus [27–29]. Each of these areas presumably extracts different aspects of the olfactory information from the MC activation patterns. Here we are interested in the top-down projections that reach the olfactory bulb from olfactory cortex. Experiments indicate that parts of olfactory cortex perform pattern completion [30–32] rather than the pattern separation that is observed for the bulbar output [33–35]. This pattern completion is presumably due to excitatory associational fibers. To wit, when presenting multi-odor mixtures the cortical representation of a mixture of 10 odors changed only little when one of the odors was omitted, but changed much more when one odor was replaced [30–32]. This suggests that the cortical network filled in the missing information when one component was omitted. In contrast, in these experiments the bulbar representations changed substantially already when a single odor was omitted from the mixture. Moreover, the pattern completion depended on the training of the animals, reflecting plastic processes [31].
Motivated by the observation of cortical pattern completion we included a minimal model of an associational brain area through a third population of principal, excitatory cells (CCs), which were connected with each other through plastic excitatory connections as well as fixed inhibitory connections. The population of CCs was divided into a larger subpopulation that received olfactory inputs from the MCs and a smaller subpopulation that was driven by non-olfactory, contextual input. This allowed this area to learn the association of specific odors with specific contexts. To capture the experimentally observed excitation of GCs by non-olfactory context information associated with specific learned odors [26], we implemented top-down projections from these CCs to the bulbar GCs. At this point it is not known whether these projections arise directly from the neurons involved in the associative pattern completion or whether additional cortical neuron populations are involved. Thus, the CCs were not meant to model a specific cortical neuron population; instead, they were intended as an effective neuronal population that mimics the observed pattern completion and the non-olfactory excitation of GCs. For simplicity, we call these CCs ‘cortical cells’.
In Fig 1A each neuron population is indicated by a circle and the connections between the individual neurons of the various populations are shown in terms of connectivity matrices (black = connection, white = no connection between the respective neurons as illustrated in Fig 2 below). The MCs received excitatory input from the sensory neurons (OSNs) and formed excitatory projections to the CCs as well as reciprocal synapses with the inhibitory GCs. This reciprocal nature of the MC-GC synapses is essential for the function of the system. To aid visualization of the results, in most computations each CC received input from only one MC, resulting in a diagonal connectivity matrix. In the Supporting Information S6 Fig we also show results for the more realistic case of an expansion from the bulb into the cortex, which resulted in a sparse cortical odor representation. The CCs formed all-to-all recurrent excitatory connections among themselves, which were endowed with Hebbian plasticity to give this network autoassociative properties. In addition, they inhibited each other via an unmodeled interneuron population. The CCs’ top-down projections formed excitatory synapses onto the GCs.
All neurons were described using nonlinear firing rate models. For simplicity, the MCs and GCs had threshold-linear firing-rate functions. The MCs exhibited some spontaneous firing in the absence of odor input, while the GCs responded only when their combined input from MCs and CCs surpassed a positive threshold. The saturation of the CC-response needed for their associative property was implemented with a sigmoidal nonlinearity.
The key aspect, network restructuring by adult neurogenesis, was implemented by persistently adding and removing GCs, as sketched in Fig 2. Newly added GCs formed randomly chosen connections with a subset of the MCs and of the CCs. Note that Fig 2 depicts only the reciprocal connections between GCs and MCs, overlaid onto the corresponding connectivity matrix. GCs were removed with a probability that decreased sigmoidally with their ‘resilience’. The resilience of a GC was taken to be the sum of its thresholded activities, which were driven by MCs and CCs in response to a set of training stimuli. This activity-enhanced survival of GCs was motivated by a host of experimental observations. Odor deprivation up-regulates GC cell death [21, 36], while odor enrichment increases the number of GCs [37]. More direct, physiological evidence is provided by genetic experiments in which the excitability of GCs was enhanced and reduced by modifying Na- and K-channels, respectively [38]. This enhanced and reduced the survival of GCs, correspondingly. The enhancement of activity could even rescue GCs from cell death induced by sensory deprivation [38].
The network structure arising in this model for a range of suitable parameter values is illustrated in Fig 1A and 1B for a simple set of two training stimuli that activated partially overlapping sets of MCs. We mostly used such simplified odor-evoked patterns rather than natural stimuli [19, 39], because they facilitated the visualization and understanding of the emerging network connectivities. Results using natural stimuli based on odor maps obtained by Johnson and Leon (cf. [40]) are presented in Supporting Information S3 Fig. Since all connections were learned based on activities, the spatial arrangement of the various neurons played no role in the model. Exposing the network to these odors lead to a strengthening of the recurrent connections among the CCs that were co-activated by the odors, implementing an associative memory of these odors. Due to the non-zero threshold for GC activation, for a GC to survive it needed to be co-activated by multiple MCs and CCs at least for some of the training stimuli. Thus, the surviving GCs were connected predominantly to MCs and CCs that had similar receptive fields.
To wit, in Fig 1A one population of GCs was mostly driven by MCs and CCs responding to the left training stimulus, while the other population responded to the one on the right. Due to the reciprocal nature of the MC-GC synapses [41] this induced not only mutual disynaptic inhibition of MCs that are co-activated by some of the training stimuli [19, 39] but also inhibition of MCs by cortical CCs that shared that receptive field. The effective connectivity matrices for the disynaptic inhibition among MCs and of MCs by CCs are shown in Fig 1B. Due to the persistent addition and removal of GCs these connectivities fluctuated in time leading to trial-to-trial variability, as discussed in Supporting Information S4 Fig. This trial-to-trial variability is also apparent in the temporal fluctuations of the number of GCs and the odor response and odor discriminability shown in Fig 3C, 3D and 3E below.
Thus, the emerging network structure is characterized by subnetworks or network modules, each of which is associated with a learned odor and provides a bidirectional projection between the bulbar and cortical representation of that odor. This is idealized in Fig 1C, where the circles again represent neuronal populations and the thickness of the lines indicates the number of neurons involved in that type of connection. Through this cortico-bulbar feedback structure cortical cells have inhibitory control specifically over those MCs that provide their dominant sensory input. Of course, depending on the similarity of the odors in the training set their associated subnetworks are more or less overlapping or intertwined.
The simulation experiments presented in the body of this paper are for one set of model parameters. Extensive further analysis has shown, however, that our results, particularly for the network structure, are robust with respect to parameter changes (cf. Supporting Information S5, S7, S8, S9 Figs).
What ramifications does this odor-specific network structure have for stimulus processing? In the following we use simulation experiments to address this question in a number of behaviorally relevant settings.
Behavioral experiments on spontaneous odor discrimination using a habituation protocol have shown that exposure to an odor related to the odors used in the discrimination task can induce a perceptual learning of that task; however, this learning was compromised when adult neurogenesis was suppressed [24]. A parsimonious interpretation of this finding is that the restructuring of the bulbar network by adult neurogenesis enhances differences in the bulbar representations of the similar odors rendering them more discriminable. To assess the impact of the network structure on odor discriminability we envisioned a read-out of the bulbar output that consists of the sum of the suitably weighted outputs of all MCs. Discriminability can then be characterized by Fisher’s linear discriminant F given by the square of the difference between the trial-averaged read-outs corresponding to the two odors divided by the trial-to-trial variability of the read-outs. Our firing-rate framework did not include any trial-to-trial variability. We therefore took as a proxy for it the firing rate, which would be proportional to the variability if the rates arose from Poisson-like spike trains. We considered here the optimal value F o p t that is obtained if the weights of the outputs to the read-out are chosen to maximize F for the stimuli in question. Such optimal weights could be the result of the animal learning the task. For similar odors F o p t typically increased in our model as the network structure evolved in response to these odors, typically in parallel with a reduction in the Pearson correlation of the MC activity patterns, capturing the observed perceptive learning [24] (cf. our previous results for a purely bulbar model [39]).
We assessed network performance based on the MC activity patterns for two reasons. As mentioned above, through the MC projections the information flows from the bulb to a variety of other brain areas [27–29], each of which making use of that information for different purposes. The MC activity patterns constitute therefore the central basis for multiple types of odor processing, and enhancements of the bulbar odor representations will affect all of them. Moreover, the CCs included in our model represented only a single aspect of such processing: associating odors with contexts [26] and associative pattern completion [30–32]. Thus, the CC-activity had a very strong contextual excitatory component, which was independent of the odor stimuli. The impact of contexts on CCs was therefore opposite to that on MCs and reduced the discriminability of these CC-patterns (cf. Supporting Information S10 Fig). These CCs were therefore not suitable as a read-out that aims to discriminate very similar odors. We envision that the modeled CCs effectively represent only a subpopulation of cortical cells, while other cortical cells, which were not included in our model, are engaged in tasks like fine discrimination of odors. This interpretation is in part motivated by the fact that the balance between feedforward sensory input and recurrent associative input varies along the anterior-posterior axis of piriform cortex [42] and between its types of principal cells [43, 44] as well as by recent observations indicating that different populations of cells in anterior piriform cortex encode odors differently [45].
The survival of the model GCs depended on the sensory input they received from the MCs as well as on the top-down input from the CCs. The top-down input was enhanced if the presented odor had previously been ‘memorized’, i.e. if the corresponding recurrent excitatory connections among the CCs had been strengthened. What happens if the cortical memory of one of the learned odors is erased, i.e. if the recurrent connections of the CCs representing that odor are removed? In our simulations removal of the memory of odor pair 1 (Fig 3B) significantly enhanced cell death and quickly reduced the total number of GCs (Fig 3C). The GCs that were removed during this phase were predominantly GCs that had previously responded to the odors in that pair and whose top-down inputs had been enhanced by the cortical memory of that odor pair (Fig 3D and Supporting Information S1 Fig). In parallel, the network’s ability to discriminate between the odors in that pair was substantially reduced (Fig 3E). This degradation of the performance did not occur if the removal of GCs was blocked in the simulation.
These results capture essential features of experiments in mice in which the extinction of an odor memory enhanced the apoptosis of GCs, particularly of those GCs that had been responsive to that odor. However, fewer of the GCs died and the mice did not forget the task when apoptosis was blocked during the extinction of the odor memory [25].
Experimentally it has been found that specific GC activity patterns could be evoked even in the absence of odors, if the animal was placed in an environment that previously had been associated with an odor [26]. In fact, the GC activation pattern that was induced by this environment had substantial overlap with the pattern evoked by the associated odor. This was not the case for a different environment. A natural interpretation of these observations is that the odorless GC activation patterns were driven by top-down projections onto the GCs from higher brain areas that have access to non-olfactory, e.g. visual, information [26]. To capture this aspect we extended our cortical model to include CCs that did not receive direct input from the olfactory bulb but were driven by non-olfactory, contextual information. This could, for instance, represent information from other sensory modalities, information about the task the animal is to perform, or an expectation by the animal. We introduced excitatory associational connections with Hebbian plasticity between these cells and the CCs that received MC input and extended the global inhibition to those cells.
Presenting two odors to the network, each in the presence of a different context, established associational connections in the cortical network between the CCs that received odor input (cell indices 1 to 110) and the CCs that received the corresponding contextual input (marked ‘context 1’ and ‘context 2’ in Fig 4B) This enabled the contextual input to excite GCs via top-down projections (Supporting Information S2 Fig) even in the absence of any odor stimulation. Due to the specificity of the cortico-bulbar network the resulting odorless GC activity patterns were highly correlated with the patterns induced by the associated odor, but not with those of the other odor (Fig 4C and 4D), recapitulating the experimental observation. In contrast, the context-evoked, odorless MC patterns were anti-correlated with the odor-evoked patterns, akin to representing a novel ‘negative’ odor, which may evoke a very different percept than the odor itself. This may account for the enhanced response of the animals observed in the familiar, but odorless context, but not in the unfamiliar context [26].
What functionality is enabled by the learned network structure, which allows CCs to inhibit specific MC? To assess this question we considered two scenarios: i) the detection and discrimination of odors in a cluttered environment and ii) rapid switching between different odor tasks.
We considered cluttered environments in which the presence of additional odors may or may not occlude (mask) the odors of interest, an olfactory analog of the ‘cocktail party’ problem [46]. As an illustrative example we considered the detection of a weak target odor (Fig 5). In the presence of a strong odor that activated a large number of MCs and occluded the target odor this required the discrimination between the occluder alone and the occluder with the target (Fig 5E). This was difficult because the MCs carrying the information about the target were also driven by the occluder, rendering the relative difference between the overall pattern with target (red, solid symbols) and that without target (black, open symbols) small. If the activation by an occluding odor could be reduced without suppressing the contribution from the target odor, detection of the target should be significantly enhanced. Indeed, in our model such an odor-specific inhibition was possible if the occluding odor was familiar, i.e. if it had been one of the training stimuli.
In these simulation experiments we associated the occluding odor during the training with context 1 (Fig 5D). The odor-specific inhibition then had two contributions. The intra-bulbar component did not require cortical activation and suppressed the familiar occluder on its own (Fig 5F left panel). In the presence of the context the detectability of the target was even further enhanced by the cortical excitation of the GCs (Fig 5G left panel), increasing the Fisher discriminant F o p t(Fig 5H). However, the same context was detrimental for the detection of the target odor if the occluding odor was not present: the strong context-enhanced inhibition almost eliminated the response to the target odor (Fig 5F and 5G right panels). The flexibility in the control afforded by top-down inputs therefore substantially enhanced performance.
The activity of the CCs and their inputs to the bulb play dual roles: they shape the cortical-bulbar connectivity during the network development and they provide—via that connectivity—input to the bulb during the task performance. To illustrate the impact of the established connections from the CCs to the GCs, we include in Fig 5H the outcome when the top-down input was blocked (wGC = 0) after the network had been established. Even in the absence of the context signal the CCs representing the familiar, occluding odor were excited, albeit less strongly. Blocking their input to the bulb therefore disinhibited the MCs representing the occluding odor and reduced the Fisher discriminant. Blocking the top-down input in the absence of the occluder avoided the excessive suppression of the target odor by the context, leading to a larger Fisher discriminant. However, the discrimination turned out not as good as in the presence of top-down input but without the context signal, since that top-down input also reduced the background activity of the MCs.
If the odors of interest are not occluded by any other odors in the environment, read-out cells can, in principle, adapt the weights of their input synapses so as to focus only on the relevant odors and ‘ignore’ the cluttered environment. However, this weight optimization is often not possible, since the animal may not know yet which odors are to be discriminated. This is, for instance, likely the case in the early phase of learning a new discrimination task [35]. During this phase it is reasonable to envision that animals rely on a large number of read-out cells each of which receives inputs from different combinations of MCs and is therefore sensitive to different aspects of the MC activity patterns. The activity of many of these read-out cells will be dominated by the uninformative components of the odor environment (‘distractor’), making it difficult to discriminate between two weak target odors, even if they are not occluded by the environment (Fig 6E, right panel illustrating optimal and random read-out). To analyze such a situation we employed a non-optimal Fisher discriminant F n o n o p t based on a large number of random read-outs of the MCs (cf. [10] in Methods).
We considered the discrimination between two similar, novel target odors in the presence of a strong, familiar odor that did not occlude the targets but served as a distractor (Fig 6E). It was associated with context 2. In addition, the network was familiarized with odors 1 and 2, which were associated with context 1 and partially overlapped with the novel target odors (Fig 6A–6D). Even in the absence of any contextual signal the learned intra-bulbar connectivity was able to suppress to some extent the distracting, familiar odor relative to the novel odors (Fig 6F). In the presence of context 2, which was associated with the distractor (‘correct’ context), this suppression was substantially enhanced through the cortical feedback driven by that context, leading to much better discriminability of the two novel odors (Fig 6G and 6H).
As in the case of discrimination via an optimal read-out (Fig 5) it was not beneficial to have indiscriminate strong cortical feedback for all familiar odors, even if these odors were not present. For instance, the feedback driven by context 1 (‘incorrect’ context) was detrimental to the discrimination of the target odors if neither of the familiar odors 1 and 2, which were associated with context 1, were part of the odor scene. While the target odors were not occluded by these familiar odors, they had significant overlap with them. Therefore the connectivity that was learned through the training included a sizable number of inhibitory projections—via the GCs—from the CCs representing context 1 to the MCs representing the target odors. This lead to a strong suppression of the MC response to the target odors, resulting in poor discriminability (Fig 6G and 6H). If the top-down input was blocked (wGC = 0), neither the beneficial nor the detrimental impact of the two contexts arose.
Thus, the ability of top-down inputs to induce specific inhibition in a flexible manner substantially enhanced olfactory processing.
The neurogenic evolution of the structure of the cortico-bulbar network occurs on a time scale of days, particularly since the apical reciprocal MC-GC synapses form only days after the proximal synapses of the top-down projections [47]. Synaptic plasticity based on changes in the synaptic weights can act on much shorter time scales, allowing cortical associational connections to adapt faster to changes in the tasks that the animal needs to perform. This could modify the top-down signals to the bulb, altering bulbar processing.
As an example we considered a situation in which the network was trained on two pairs of odors, O1,2 and O3,4. The two odors in each pair were very similar, but the pairs were dissimilar from each other (Fig 7A). As expected, the training increased the discriminability of the odors within a pair. However, for the discrimination of the mixture M1 = 0.55O1 + 0.45O3 from the mixture M2 = 0.45O1 + 0.55O3, obtained by combining odors from the two pairs, the training on the individual components O1,2 and O3,4 was detrimental. While it established mutual inhibition of MCs that were activated by the same mixture component (O1 or O3), it provided only little mutual inhibition between MCs that were activated by different mixture components: the number of connections between MCs representing odor O1 (MCs with index near 30 in Fig 7B) and MCs representing odor O3 (MC index near 80) was small. However, these MCs are activated simultaneously in the mixtures and mutual inhibition of these MCs is needed to enhance the discrimination between the mixtures [19, 39, 48, 49]. As a result the inhibition significantly reduced the relative difference between the two mixtures and with it their Fisher discriminant (Fig 7F, left panel, and Fig 7G). Inhibition between the MCs representing O1 and those representing O3 can be effected by establishing associational excitatory connections between the CCs that disynaptically inhibit these two groups. To do so we exploited the Hebbian plasticity of the cortical synapses and trained the cortical network briefly on the mixture 0.5O1 + 0.5O3 (Fig 7E, right panel). This enhanced the discriminability of the mixtures substantially (Fig 7F, right panel, and Fig 7G). The role of the top-down inputs before and after learning are qualitatively different. Before learning, the intra-bulbar as well as the cortically driven inhibition of mixture component O1 increases with O1-activation, which reduces the difference in the O1-activity in the two mixtures, and analogously for component O3. This reduces the discriminability of the mixtures. After learning, the cortical contribution to the inhibition of each component is driven by the combined activity of both components and is therefore identical for both mixtures. This preserves their difference. In S11 Fig (Supporting Information) we show that this improvement is not due to the overall reduction in MC-pattern amplitude.
Thus, cortical projections can exploit the learned, odor-specific subnetwork structure (Fig 1C) to switch between different cortico-bulbar processing modes, adapting to the odor objects at hand.
Since the top-down input is at the core of the bidirectional mapping between the bulbar and cortical odor representations, we considered the influence of its strength wGC on the connectivity and function of the bulbar-cortical network in some detail. Here we focus on the context-enhanced processing of a stimulus in the presence of a distractor. The results for an occluded stimulus are presented in S9 Fig in Supporting Information.
The top-down inputs are instrumental during the network development as well as during the processing of stimuli. The survival of GCs depends on their overall activity, which results from bulbar and from cortical inputs. Therefore, increasing the strength wGC of the top-down projections shifts the balance from the bulbar inputs determining the survival and with it the network connectivity to the cortical inputs dominating the development. This is demonstrated in Fig 8 for the cluttered environment investigated in Fig 6. For very small wGC training with the stimuli shown in Fig 6A resulted in a highly selective intra-bulbar connectivity in which the mutual inhibition was essentially restricted to MCs that were co-active for one of the training stimuli. The MCs received, however, disynaptic top-down inhibition that only slightly reflected the receptive fields of the MCs and the CCs, and most CCs induced inhibition on most MCs that responded to any of the training stimuli (Fig 8B). In the opposite limit of large wGC only few CCs had projections to the bulb, but the inhibition they induced was only weakly targeted to specific MCs. Moreover, the intra-bulbar connectivity was essentially homogeneous. As a result, the inhibition induced by the top-down projections did not allow different CCs to inhibit specific different sets of MCs. Thus, to obtain a subnetwork structure of the type sketched in Fig 1C, wGC had to be in an intermediate range in which the bulbar and cortical inputs to the GCs were of similar magnitude (1 ≲ wGC ≲ 3.5 in Fig 8).
We probed the performance of the resulting networks with two tasks. In both, the stimuli involved a distractor that is very different from the target odors (Fig 8D) and we considered the non-optimal read-out discussed in Fig 6. For the context associated with the distractor to enhance the processing, the top-down input has to suppress the MC-activity driven by the distractor but not that due to the target odors. This was the case for intermediate values of wGC and resulted for both types of target odors in an increase in F n o n o p t in the presence of the correct context, but in a decrease for the incorrect context (Fig 8C and 8E). For larger wGC, however, the performance deteriorated, both with the correct context and without any context. This was caused by the decrease in the specificity of the connectivity, which is quantified in S7 Fig. As a result even the correct context suppressed the activity of the MCs responding to the target odors, reducing F n o n o p t. This is shown explicitly in S8 Fig in Supporting Information. A similar dependence on wGC was found for the optimal detection of an occluded target (S9 Fig in Supporting Information, cf. Fig 5).
The key anatomical feature of the model network resulting from learning is its connectivity. Specifically, the projections that the GCs receive from the MCs and from the CCs are predicted to be matched: a given GC receives cortical inputs predominantly from those CCs that respond to the same odors as the MCs projecting to that GC. This matching of the GCs’ receptive fields can be tested experimentally. One possibility is to express—after suitable training—ChR2 conditionally (e.g., via c-Fos [50]) in those principal cells of piriform cortex that are activated by the training odor, combined with expression of a calcium-indicator in the GCs. Note that the training needs to activate the neurogenic plasticity of the bulb [24], which is not the case if the odor exposure is only passive [51]. The model predicts that optical stimulation of cortical cells in the absence of an odor will then lead to excitation patterns of the GCs that are strongly correlated with the patterns excited by the training odor (Fig 4C and 4D). Previous experiments in which odor-evoked and context-evoked GC activation patterns were found to be correlated in different animals are suggestive of this outcome [26].
On a behavioral level the model makes specific predictions for the learning of odor discrimination or detection in cluttered environments. Recent experiments have shown that the detection of an odor in a go/no-go task is particularly difficult if that odor is masked or occluded by an odor [46]. Our model predicts that the performance in such a detection task would be enhanced if the animal is first familiarized with the occluding odor over an extended period of time [24]. The resulting restructuring of the bulbar network would lead to a reduction in the response to that familiar odor, partially unmasking the task-relevant odor. If, in addition, the occluding familiar odor has been associated with a non-olfactory context [26], the model predicts that the performance is further enhanced if the task is performed in that context, but not in a different, novel context (Fig 5A).
Even if the cluttered odor environment does not occlude the task-relevant odors, it is expected that the learning speed in an odor-discrimination task [35] is reduced by strong distracting odors. The model predicts that sufficient familiarization [24] with the distracting odors will reduce their uninformative contributions to the overall MC activation pattern. This is expected to increase the signal-to-noise ratio and with it the learning speed by reducing the contributions from the uninformative MCs to the variability of the read-out. If, in addition, the distracting, uninformative odors are associated with a context, the learning speed is predicted to increase further in the presence of that context (Fig 6B). If the cluttered environment makes the task too hard to learn for naive animals, our model suggests that prior familiarization with the distracting odors, preferably in a specific context, may enable the animals to master this difficult task.
Adult neurogenesis is a striking mechanism of structural plasticity that has the potential to rewire a network extensively. In mammals it arises predominantly in two brain areas. In the dentate gyrus it involves excitatory granule cells; their role in the network has been studied in detail [52, 53], also in terms of computational models [54]. In the olfactory system, where adult neurogenesis involves inhibitory rather than excitatory granule cells, there is also a substantial body of experimental work [20], but only few modeling studies are available [39, 55]. Here we have developed a computational model for the neurogenic evolution of the network connectivity with an emphasis on the possible role of the pervasive top-down projections from cortical areas. The model is based on a number of experimental observations: GC survival depends on GC activity [21, 38], GC activity can be induced in the absence of odor stimulation [26], and piriform cortex exhibits extensive recurrent excitation, which can support associational memory [30, 31]. The model captures qualitatively the experimentally observed perceptual learning afforded by neurogenesis [24] as well as the enhanced apoptosis of specific GCs and the reduced odor discriminability after the extinction of memories [25].
Without theoretical guidance, it is difficult in experiments to identify the functional structure of the cortico-bulbar connectivity, in particular, because odor representations in the olfactory system do not reflect detailed spatial maps like those of other sensory systems [56, 57]. An important contribution of the model is therefore its prediction that through the structural plasticity the network develops a structure that reflects the learned odors and provides enhanced inhibition that is specific to these odors. This inhibition is in part intra-bulbar and in part driven by top-down (cortical) inputs. The latter reflects the formation of a bidirectional connection between the bulbar and the cortical representation of the familiar odors, which allows the cortical cells associated with such an odor to inhibit specifically those MCs that are excited by that odor. This inhibition is mediated by GCs. For this connectivity to arise the reciprocal nature of the MC-GC synapses is essential. Our results therefore suggest that a key function of the reciprocity of these synapses may be to guide the wiring of the cortico-bulbar network connectivity. The predicted matching of the GC receptive fields for olfactory and for cortical input can be tested experimentally. Moreover, the model can guide future experiments aimed at elucidating the cortico-bulbar connectivity.
Functionally, the model predicts, in particular, that the learned connectivity improves the detection and discrimination of novel odors in cluttered environments, if the occluding or distracting odors are familiar. This is achieved by a reorganization of the cortical-bulbar network so as to enhance the inhibition of familiar odors. Processing of odors in cluttered environments can also be enhanced by adaptation of sensory neurons or adaptation further downstream [58]. For those mechanisms to be successful, the occluding/distracting stimuli need to be present before the novel stimulus. In contrast, the neurogenically formed structured network suppresses familiar distractors and occluders independent of their relative onset times, even if the novel odor precedes the occluding or distracting odor.
Moreover, the processing of cluttered environments can be enhanced by top-down input when the occluding or distracting odor activates cortical memory. This may indicate that a higher brain area has recognized parts of the odor scene and may allow something akin to the ‘explaining away’ of components of a complex odor mixture that is theoretically predicted for the optimal processing of stimuli [8–10, 16]. Recent work has identified networks that demix familiar odors employing approximate optimal Bayesian inference; the anatomical structure of these networks is very close to that emerging naturally in our neurogenic model [10, 11]. By providing a biophysically supported mechanism through which the system can learn the required network structure our model complements this abstract normative approach.
Going beyond the purely olfactory aspect, the top-down input could encode task-related expectations or contextual information originating from other sensory modalities. Thus, it may implement a predictive coding in the bulb that reflects the context or task at hand [59]. Our model demonstrates how such contextual information can enhance performance.
The formation of the bulbar and cortico-bulbar network via structural plasticity is a relatively slow process. However, our model predicts that the network structure emerging from it can be exploited by faster synaptic learning processes in cortex, which allow the system to switch relatively quickly between different discrimination tasks. This is reminiscent of the task-dependent switching of neuronal responses observed in V1 [7].
Focusing on the slow evolution of the network structure our model is intentionally minimal with respect to the dynamics of the individual neurons. We describe the neurons in terms of their firing rate and focus on neuronal steady-state activities. So far, the knowledge about the biophysical mechanisms controlling GC survival and their dependence on neuronal activity is not sufficient to guide the development of a detailed model of that process, which would, e.g., connect GC spiking or calcium-levels with GC survival [38, 60, 61].
To assess the discriminability of MC activity patterns we assumed that the firing rates result from an irregular firing in which the variance of the spike number is proportional, albeit not necessarily equal, to the mean spike number. Thus, we have not taken the widely observed rhythmic aspects of the bulbar activity into account [62], nor the possibility that animals may also be able to make use of spike-timing-, correlation-, or synchrony-information in odor processing [9, 63–66]. The modular network structure and the associated specific disynaptic inhibition of MCs by top-down inputs that are predicted by our model are likely to have impact also on spike timing and synchrony [67]. A study of the resulting dynamics and functional consequences is, however, beyond the scope of this paper and will be left to future work.
Not much is known experimentally about the parameter values for the model. There have been recent efforts to constrain various coupling strengths in bulbar-cortical rate models based on experimental measurements of firing rate correlations [68]. The two types of networks investigated there have, however, different types of connections than the network investigated here. For instance, in their models the bulbar inhibitory cells receive either input directly from the sensory neurons, which would identify them as periglomerular rather than granule cells (e.g. [69]), or only by the cortical cells but not from the mitral cells. The observed reciprocal nature of the synapses between GCs and MCs plays, however, a central role in our model. It is therefore not clear to what extent the results in [68] directly provide constraints for the strength of the connections included in our work. It would be of interest to apply their approach to the class of networks obtained here to obtain some additional guidance for the choice of parameters. We have checked that our key results are quite robust under parameter changes as long as the cortical network does not become bistable, i.e. as long as it does not sustain activity without any bulbar or contextual input.
We have used a minimal model to focus on the dominant aspects associated with neurogenic network restructuring. Thus, in the bulb we have omitted the odor processing performed in the glomerular layer, which may also contribute to pattern separation and normalization [70, 71]. Our cortical model aimed to capture only two aspects of cortical processing: the formation of memories and the association of odors with contexts, both through excitatory lateral connections with Hebbian plasticity and all-to-all recurrent inhibition. We have not included multiple layers of principal cells nor feedforward inhibition from the bulb [72].
The olfactory system most likely gains additional richness through the nonlinear response properties of the GCs, which include local dendritic depolarization that can provide reciprocal inhibition to the MCs even without spiking [73], as well as wide-spread dendritic activation associated with calcium- or sodium-spikes that may drive wide-spread lateral inhibition [74, 75]. Thus, it is likely that inhibition can operate in multiple regimes [76], which may further enhance the system’s ability to switch between different tasks.
Adult neurogenesis is not the only mechanism contributing to the plasticity of the olfactory bulb, which is, for instance, reflected in the temporal evolution of the receptive fields of adult-born GCs [77] and in changes in the connectivity between MCs and GCs that occur during learning [78]. Spike-timing dependent plasticity is known to be enhanced in the proximal synapses of young adult-born GCs [4, 79]. At the reciprocal synapses between MCs and GCs long-term depression [80, 81] and bidirectional plasticity driven by single bursts [82] have been observed. In addition, the reciprocal synapses exhibit substantial structural plasticity in the form of spine fluctuations in adult-born as well as neonatally born GCs [18, 19]. In computational models of the olfactory bulb without top-down inputs spine fluctuations and adult neurogenesis lead to very similar network connectivities [19, 39]. We therefore expect that a cortico-bulbar model based on spine fluctuations would lead to results that are very similar to the ones based on adult neurogenesis described here. How the two types of structural plasticity complement each other and how the sequential formation of proximal and apical synapses [47] affects the emerging connectivity are interesting questions, which are, however, beyond the scope of this paper.
Our model consisted of three populations of neurons, mitral/tufted cells (M), granule cells (G), and ‘cortical’ cells (C). Although the differences between mitral and tufted cells in terms of their properties and function are becoming increasingly known [83–85], the model did not distinguish between them. The neuronal activities were described by the nonlinear rate equations
τ M d M i d t = - M i + S i + M s p - - g ∑ j W i j ( M G ) [ G j - G t h ] + , (1) τ G d G i d t = - G i + ∑ j W i j ( G M ) [ M j ] + + + w G C ∑ j W i j ( G C ) σ ( C j ) , (2) τ C d C i d t = - C i + w C M ∑ j W i j ( C M ) [ M j ] + + + ∑ j ≠ i ( α W i j ( C C ) - w i n h ) σ ( C j ) . (3)
The odor stimuli were given by Si and the spontaneous MC activity by Msp. A minimal, rectifying nonlinearity was chosen for the MCs and GCs,
[ M ] + = { M for M > 0 0 for M ≤ 0 .
To allow for pattern completion in the cortical area the nonlinearity for the CCs was chosen to be sigmoidal
σ ( C ) = C m a x 1 + e - γ C ( C - C t h ) .
The bulbar connectivity matrices satisfied
W i j ( M G ) = W j i ( G M ) ,
reflecting the important reciprocity of the MC-GC synapses. The entries of W i j ( G M ) were given by 0 or 1. Without loss of generality the strength of the excitatory synapses from the MCs to the GCs was set to 1, while the strength of the reciprocal inhibition was given by g.
The CCs received excitatory inputs of strength wCM from the MCs via the connectivity matrix W i j ( C M ), i.e. the entries of W i j ( C M ) were 0 or 1. In all of our computations except for those investigating the impact of sparse cortical representations (cf. S6 Fig in Supporting Information) W i j ( C M ) was taken to be the identity matrix. The excitatory cortical synapses were taken to be plastic. In all the simulations presented in this paper the cortical connectivity was learned at the beginning of the simulations using the Hebbian plasticity rule
W i j ( C C ) → W i j ( C C ) + η { ( σ ( C i ) - Ω ) σ ( C j ) - κ } (4)
with hard limits given by 0 ≤ W i j ( C C ) ≤ W m a x ( C C ). Here, the cortical activities Ci were given by the steady state reached for the current values of W i j ( C C ). The learning rate was given by η, the threshold for potentiation by Ω. There was an overall slow decay of the weights given by κ. The parameter α in Eq (3) allowed a reduction of the recurrent excitation during learning to reduce interference with previously learned memories [86] by switching between αlearn and αrecall. During the neurogenic network evolution the strengths of the recurrent cortical connections were held fixed. There was also all-to-all inhibition among the cortical cells with strength winh.
The connectivities W i j ( M M ) and W i j ( M C ) of the effective disynaptic inhibition between MCs and of MCs by CCs, examples of which are shown in Fig 1B, are given by
W i j ( M M ) = ∑ k = 1 N G W i k ( M G ) W k j ( G M ) , W i j ( M C ) = ∑ k = 1 N G W i j ( M G ) W k j ( G C ) , (5)
where NG is the number of GCs. Note that the diagonal elements of W i j ( M M ) are given by the number of non-zero elements in the corresponding rows of W i j ( M G ) and are typically quite large. In the connectivity figures we have therefore replaced these large values by 0 to reveal the structure of the remaining connections. In the computations these diagonal elements were, of course, not set to 0.
In each time step of the neurogenic network evolution N n e w ( G ) new GCs were added to the network, each making N c o n n ( M G ) randomly chosen connections with MCs and N c o n n ( G C ) randomly chosen connections with CCs. Then the steady-state values of M, G, and C for each odor S i ( β ), β = 1…Ns, in the odor environment were determined by solving the evolution Eqs (1), (2) and (3) for the corresponding odor until a steady-state was reached. Since only the steady-state values were desired we set τG = 0, which allowed a drastic reduction of the number of differential equations that had to be solved. Then the resilience of each GC was determined as the sum of its thresholded activity over all Ns training stimuli,
R i = ∑ β = 1 N s [ G i ( β ) - G t h ] + , (6)
and GCs were removed depending on their survival probability Pi given by (cf. Fig 2)
P i = 1 2 ( tanh ( γ R ( R i - R 0 ) ) + 1 ) . (7)
This completed the neurogenic time step. For all results, except in Fig 3, sufficiently many such steps were taken to reach a statistically steady state in the network connectivity and in the various quantities assessing the features of the system. By using sufficiently many GCs the network fluctuations in this steady state were kept small enough to allow reliable measurements of the various relevant quantities.
The number of GCs was not fixed in this neurogenic model; instead, it depended on the strength of the inputs S i ( β ) and the strength g of the inhibitory connections. For strong inhibition g the MC activities were low, leading to low activities of the GCs and a correspondingly low survival probability. As a result, the number of GCs was low. Conversely, choosing weak inhibition g lead to a large number of GCs. When the number of GCs was low, adding and removing a single GC had a significant effect on the MC and GC activities, resulting in strong fluctuations in the output of the network. To balance computational effort with the need to keep the fluctuations sufficiently low, we therefore adjusted in the simulations the inhibitory strength g during the network evolution to keep the number of GCs within a predetermined target range centered at N G C ( a i m ). Only in Fig 3, where the focus was the temporal evolution of the network, we kept g fixed after t = 18, 750 before the memory was extinguished at t = 25, 000.
Each model odor stimulus was given by Nh combinations of Gaussian activity profiles of the form
S i = ∑ k = 1 N h A k e - ( i - x k ) 2 w k 2 . (8)
Contextual inputs drove the corresponding contextual cortical cells Ci with a cell-independent amplitude Acontext. In total there were N C C ( c o n t e x t ) CCs representing the contexts.
To assess the discriminability of two stimuli S i ( 1 ) and S i ( 2 ) we used the optimal Fisher discriminant Fopt F o p t = ∑ i = 1 N M C ( [ M i ( 1 ) ] + - [ M i ( 2 ) ] + ) 2 [ M i ( 1 ) ] + + [ M i ( 2 ) ] + (9)
and a non-optimal Fisher discriminant Fnonopt, which was obtained as the mean of 5,000 Fisher discriminants Frandom,
F n o n o p t = 〈 F r a n d o m 〉 ,
each of which was based on a random read-out of the MC activities,
F r a n d o m = ( ∑ i = 1 N M C w i ( r a n d o m ) ( [ M i ( 1 ) ] + - [ M i ( 2 ) ] + ) )2 ∑ i = 1 N M C ( w i ( r a n d o m ) )2 ( [ M i ( 1 ) ] + + [ M i ( 2 ) ] + ) . (10)
Here the weights w i ( r a n d o m ) were real numbers between -1 and +1 drawn from a uniform distribution. By including negative weights we allowed for the possibility that the hypothetical read-out cell could also receive disynaptic inhibition from the MCs.
In the simulation experiments presented in the body of this paper we used one set of parameters. We have, however, tested that our results are robust under parameter changes (cf. S5 Fig in Supporting Information). The parameters used here are as follows:
τ M = 1 τ G = 0 τ C = 1 , M s p = 0 . 2 G t h = 3 w G C = 3 , α l e a r n = 0 . 05 α r e c a l l = 1 , γ C = 3 C t h = 0 . 2 C m a x = 1 w i n h = 0 . 05 , Ω = 0 . 2 κ = 0 . 1 η = 10 W m a x ( C C ) = 0 . 06 , N c o n n ( M G ) = 16 N c o n n ( G P ) = 2 N n e w ( G ) = 0 . 1 · N G C ( a i m ) N C C ( c o n t e x t ) = 48 γ R = 5 R 0 = 3 N G C ( a i m ) = { 2 , 000 Fig.1, Fig. 3,Figs.5,6 4 , 000 Fig.4 16 , 000 Fig.7
The parameter values for the training stimuli stimuli were
A k = 2 - M s p w k = 12 A c o n t e x t = 2 . (11)
The same parameters were used for the probe stimuli in Figs 3 and 7. In Fig 5 the parameters of the target in the probe stimulus were
A k = 0 . 72 w k = 2 ,
while in Fig 6 the parameters for the targets were
A k = 1 . 08 w k = 12 .
The parameters for the occluder and the distractor in Figs 5 and 6 were also given by Eq (11).
The Matlab code implementing this model is available from the ModelDB web site senselab.med.yale.edu/modeldb under access number 247188.
|
10.1371/journal.pbio.2005970 | CellProfiler 3.0: Next-generation image processing for biology | CellProfiler has enabled the scientific research community to create flexible, modular image analysis pipelines since its release in 2005. Here, we describe CellProfiler 3.0, a new version of the software supporting both whole-volume and plane-wise analysis of three-dimensional (3D) image stacks, increasingly common in biomedical research. CellProfiler’s infrastructure is greatly improved, and we provide a protocol for cloud-based, large-scale image processing. New plugins enable running pretrained deep learning models on images. Designed by and for biologists, CellProfiler equips researchers with powerful computational tools via a well-documented user interface, empowering biologists in all fields to create quantitative, reproducible image analysis workflows.
| The “big-data revolution” has struck biology: it is now common for robots to prepare cell samples and take thousands of microscopy images. Looking at the resulting images by eye would be extremely tedious, not to mention subjective. Thus, many biologists find they need software to analyze images easily and accurately. The third major release of our free open-source software CellProfiler is designed to help biologists working with images, whether a few or thousands. Researchers can download an online example workflow (that is, a “pipeline”) or create their own from scratch. Pipelines are easy to save, reuse, and share, helping improve scientific reproducibility. In this release, we’ve added the capability to find and measure objects in three-dimensional (3D) images. We’ve also made changes to CellProfiler’s underlying code to make it faster to run and easier to install, and we’ve added the ability to process images in the cloud and using neural networks (deep learning). We’ve also added more explanations to CellProfiler’s settings to help new users get started. We hope these changes will make CellProfiler an even better tool for current users and will provide new users better ways to get started doing quantitative image analysis.
| Image analysis software is now used throughout biomedical research in order to reduce subjective bias and quantify subtle phenotypes when working with microscopy images. Automated microscopes are further transforming modern research. Experiments testing chemical compounds or genetic perturbations can reach a scale of many thousands of perturbations, and multidimensional imaging (time-lapse and three-dimensional [3D]) also produces enormous data sets that require automated analysis. In light of this data scale, computer algorithms must deliver accurate identification of cells, subcompartments, or organisms and extract necessary descriptive features (metrics) for each identified object.
Racing to keep up with the advancement of automated microscopy are several classes of biologist-focused image analysis software, such as companion packages bundled with imaging instruments (e.g., MetaMorph—Molecular Devices, Elements—Nikon), stand-alone commercial image processing tools (e.g., Imaris—Bitplane), and free open-source packages (e.g., ImageJ/Fiji, CellProfiler, Icy, KNIME). Commercial software is often convenient to use, especially when bundled with a microscope. Although cost and lack of flexibility may limit adoption, there is a focus on usability, particularly for applications of interest to the pharmaceutical industry. Still, the proprietary nature of the code in commercial software limits researchers from knowing how their data is being analyzed or modifying the strategy of a given algorithm, if desired.
The open-source biological image analysis software ecosystem is thriving [1]. ImageJ [2] was the first and is still the most widely used package for bioimage analysis; several other packages are based on its codebase (most notably, Fiji). ImageJ excels at the analysis of individual images, with a user interface analogous to Adobe Photoshop. Its major strength is its community of users and developers who contribute plugins, although an associated drawback is the sheer number of plugins, with varying degrees of functional overlap, usability, and documentation. Multitasking toolboxes like KNIME [3] offer a more modular approach, which is better suited to automated workflows. KNIME equips users with a wide breadth of powerful utility, from performing image analysis to data analytics.
CellProfiler, our open-source software for measuring and analyzing cell images, has been cited more than 6,000 times, currently at a rate of more than 1,000 per year. The first version of CellProfiler was introduced in 2005 and published in 2006 [4]. It is widely adopted worldwide, enabling biologists without training in computer vision or programming to quantitatively measure phenotypes robustly from thousands of images. A second major version of CellProfiler, rewritten in Python from its original MATLAB implementation, was published in 2011 [5] and included methods for tracking cells in movies and measuring neurons, worms, and tissue samples. In 2015, a laboratory unaffiliated with our team rigorously compared 15 free software tools for biological image analysis: CellProfiler was ranked first for both usability and functionality [6].
CellProfiler provides advanced algorithms for image analysis, organized as individual modules that can be placed in sequential order to form a pipeline. This pipeline is then used to identify and measure cells or other biological objects and their morphological features. CellProfiler’s modular design and carefully curated library of image processing and analysis modules benefits biologists in several ways:
Reproducibility at scale: CellProfiler is designed to produce high-content information for each cell or other object of interest in each image and to apply the same objective analysis in high-throughput, e.g., across thousands or millions of images.
Flexible feature extraction: Individual modules measure standard morphological features such as size, shape, intensity, and texture. Customized combinations of modules can extract even more complex information. As such, CellProfiler is commonly used for morphological profiling experiments such as Cell Painting [7,8], which is being adopted in pharmaceutical companies to speed several steps in drug discovery [9].
Easy to learn: Each of the 70+ modules includes carefully crafted documentation, curated by both imaging and biology experts, to make image processing more approachable and understandable for the average scientist. Further, each individual setting is explained in practical terms to aid researchers in configuring it. The number of modules and settings is carefully limited to avoid overwhelming users, while a plugin system allows the flexibility of a larger array of contributed modules.
Community: CellProfiler has an active community of more than 3,000 people on its online question and answer forum. With more than 15,000 posts, users provide feedback that fuels improvements to CellProfiler, find pipelines related to their area of research, interact with developers, get input on challenging problems, and improve image analysis skills and knowledge by helping other users design solutions.
In the CellProfiler 3.0 release, we introduced methods for analyzing 3D images, using deep learning architectures and cloud computing resources, and other improvements to CellProfiler’s usability and capabilities.
This new version of CellProfiler has support for analysis of 3D images in many of its modules (S1 Fig). Although open-source software tuned to 3D problems exists (e.g., Vaa3D, BioImageXD, Slicer) [10], it often emphasizes visualization and rendering; these new 3D capabilities of CellProfiler meet the community’s demand for modular high-throughput 3D analysis. CellProfiler 3.0 can apply image processing, segmentation, and feature extraction algorithms to entire image volumes (volumetric analysis), in addition to the more typical iterative and separate analysis of two-dimensional slices from a 3D volume (“plane-wise” analysis). Whole-volume algorithms consider 3D neighborhoods and incorporate information from surrounding planes, yielding more accurate results, but require more available memory, particularly for large files. CellProfiler’s volumetric algorithms can be configured to account for anisotropic data (in which the distance between Z planes does not match the distance between pixels in the X and Y dimensions). While we focused on adding 3D capability to most of our image processing and feature extraction modules, we will continue increasing the number of CellProfiler modules that support image volumes for situations in which it is not computationally prohibitive.
We developed 3D pipelines to identify cells and subcompartments of cells for a number of experimental situations and sample types across a number of laboratories. We identified nuclei based on a DNA stain (Fig 1A) in 3D image stacks of human induced pluripotent stem cells (hiPSCs). After processing by several CellProfiler modules (Fig 1C), the final results agree well with manually annotated nuclei (Fig 1D). Results for a variety of images with a range of complexity are shown in Fig 2, with more detailed views in S2–S5 Figs. We characterized CellProfiler’s segmentation accuracy in two ways: in the first, we used real microscopy images (Fig 1A, Fig 2A, Fig 2B) whose ground truth was manually annotated by an expert image analyst; such images are realistic, but the manual annotation introduces some subjectivity. We therefore also used synthetic images (Fig 2C, Fig 2D)[11,12], which, depending on the model used to create them, may not perfectly represent real microscopy images but whose ground truth can be unambiguously known.
To determine how well the segmented objects agreed with ground truth, CellProfiler’s “MeasureImageOverlap” module was used to calculate the plane-wise Rand index [13], a performance metric of accuracy (Fig 1B, Fig 2E). Rand index values showed good agreement (0.919–0.976) between each tested image and its ground truth. The results produced by CellProfiler 3.0 were comparable to results produced by the commonly used Fiji plugin MorphoLibJ (0.930–0.977) (Fig 1B, Fig 2E and S2–S5 Figs; the MorphoLibJ macro codes are provided in S1 Table). We demonstrate several kinds of analysis, including analyses of cell count in a time series that was synthetically generated [11,14](S5 Fig); identification and quantification of children objects inside parent objects, such as speckles of transcripts within cells (Fig 3); and measurement of various features of hiPSCs located at the center and the edge of the cell colony (Fig 4).
All pipelines, annotated with notes to understand the function of each module, are provided at https://github.com/carpenterlab/2018_mcquin_PLOSBio. All raw images, together with ground truth annotations used to test CellProfiler 3.0 performance, are publicly available for further community algorithm development in the Broad Bioimage Benchmark Collection [15], as indicated in the legends for Fig 1 and S2–S5 Figs.
Convolutional neural networks (CNNs) are a type of deep learning model that transforms input images into outputs specified by the problem type [16]. For instance, image classification models transform images into categorical labels [17], while image segmentation models transform images into segmentation masks [18]. CNNs are now widely used to solve many computer vision tasks, given their ability to produce accurate outputs after learning from examples. CellProfiler now can be configured to make use of cutting-edge CNNs to analyze biomedical images. While CellProfiler does not yet incorporate user-friendly functionalities to train neural networks, various models that have been already trained by researchers can be run inside CellProfiler.
Running neural network models requires the installation of certain deep learning frameworks that are distributed separately, such as TensorFlow or Caffe. TensorFlow [19] is an open-source software library for machine learning that interfaces with Python and is compatible with CellProfiler when installed from source on Linux, Mac, and more recently, Windows. Caffe [20] is a deep learning framework designed for high-performance neural networks and is primarily available for Linux systems. Some network models may need special graphics processing units (GPUs) installed and configured in the system to run the computations efficiently, but this is not always required. Fortunately, both TensorFlow and Caffe can easily switch between running on GPUs and traditional central processing units (CPUs) just by changing the corresponding configuration.
We created the CellProfiler 3.0 module ClassifyPixels-Unet to segment nuclei in images stained with DNA labels (https://github.com/CellProfiler/CellProfiler-plugins). This plugin implements a U-Net[18] model using TensorFlow and can be run on CPUs. We have also provided the network architecture with training routines in case users have their own annotated images to learn a segmentation model for different images and objects of interest (https://github.com/carpenterlab/unet4nuclei). The ClassifyPixels-Unet module classifies pixels into one of three classes: background, nucleus interior, or nuclear boundary (S7 Fig). A pretrained network for nuclei segmentation is available for download and is automatically loaded by the plugin; a pipeline and image to run this are available as S4 File.
We also created a CellProfiler 3.0 module, MeasureImageFocus, in collaboration with Google Accelerated Science, who trained a model to detect focus in images [21]. The module displays a table with the predicted focus score and certainty for the whole image, as well as a figure with the focus scores and corresponding certainties of individual 84 × 84 patches represented by color and opaqueness. It uses TensorFlow as its underlying deep learning framework. Independently, Sadanandan and colleagues created a CellProfiler 2.2.0 module—CellProfiler-Caffe bridge—that enables running a pretrained model for cell segmentation within a CellProfiler pipeline [22].
We created Distributed-CellProfiler (https://github.com/CellProfiler/Distributed-CellProfiler), a script-based interface that allows running thousands of batches of images through CellProfiler in parallel on Amazon Web Services (AWS; S8 Fig). While Distributed-CellProfiler does require basic knowledge of AWS and interaction with the command line, it is well documented and has been successfully run by biologists without formal computational training. The script handles infrastructure creation and removal as well as creation and storage of logs, allowing users without access to a local cluster computing environment to analyze large data sets with only minimal time devoted to having to set up those resources. Sample pipelines and configuration files are available as S5 File.
Plug-ins: CellProfiler-plugins is a new repository for the community to share and distribute new CellProfiler modules (https://github.com/CellProfiler/CellProfiler-plugins).
Documentation: All of CellProfiler’s documentation was updated for content and readability; detailed help is available for 100% of module configuration options (excluding plugins).
New image processing features: CellProfiler 3.0 introduces an extended suite of modules for feature detection, feature extraction, filtering and noise reduction, image processing, image segmentation, and mathematical morphology operations.
Infrastructure improvements: The project team reengineered major core components of CellProfiler. CellProfiler’s codebase was trimmed down, in part because of better integration with Python’s scientific community. We have adopted and contributed to the standard libraries of the scientific Python community, including NumPy, SciPy, and scikit-image. CellProfiler’s code is now 100% Python, which improves interoperability with the robust Python scientific ecosystem and simplifies third-party contributions. As well, we upgraded support to 64-bit on Linux, MacOS, and Windows, and a continuous integration process ensures the software is well tested on a variety of platforms.
We made substantial progress simplifying CellProfiler’s installation. In addition to our previously existing Mac and Windows builds, a Python wheel is now available from the Python Package Index, and a Docker image is now available from Docker Hub. In an effort to expand CellProfiler’s flexibility, we made CellProfiler much simpler to compile on a variety of familiar and unusual platforms by requiring fewer dependencies and only using ubiquitous build systems.
Educational resources: CellProfiler’s many examples and tutorials are now publicly available on GitHub (https://github.com/CellProfiler/examples and https://github.com/CellProfiler/tutorials) and have been updated for compatibility with CellProfiler 3.0.
Speed: CellProfiler 3.0’s processing speed is faster than version 2.2 on the most common types of pipelines; the degree of difference depends on the exact modules involved: CellProfiler 3.0 ran at a comparable or faster speed than CellProfiler 2.2 for 11 of 16 example pipelines tested (S9 Fig). While the total amount of time needed to run the five pipelines shown in S9 Fig was comparable between CellProfiler and MorphoLibJ (482 versus 542 seconds), the relative speed was highly specific to the individual pipeline (S6 File), ranging from 2× faster in CellProfiler to 6× faster in MorphoLibJ (S2 Table). In addition, CellProfiler can run multiple images in parallel, depending on the individual’s number of threads, computing power, and access to cloud computing resources, making it suited to large-scale experiments. As well, CellProfiler’s modules enable more readily configurable complex analyses than MorphoLibJ, such as associating cytoplasm regions (as in Fig 3), transcripts (as in Fig 3), and other entities to nuclei and measuring a wide variety of morphological properties of each, including intensities, shapes, textures, colocalization metrics, and neighborhood relationships (as in Fig 4).
CellProfiler is mature software serving a large community and making an impact through its thousands of users’ biological discoveries. It has been involved in the discovery of potential life-saving drugs for infectious diseases, leukemia, and cerebral cavernous malformation [23–27] and in clinical trials for hematological malignancies [28] and will continue to fuel basic and applied research around the world.
CellProfiler can readily generate a large amount of morphological information for each biological entity that is measured. We see advancements in data mining, downstream and apart from CellProfiler, as blossoming in the coming years. Already, 20 laboratories in the field of morphological profiling have gathered for two annual meetings/hackathons (now called CytoData) [29], collaborated to outline best practices [30], and begun a community library (Cytominer, https://github.com/cytomining/cytominer). In addition to our user-friendly tool for classical machine learning based on measured features, CellProfiler Analyst [31], we have begun creating Deepometry (http://github.com/broadinstitute/deepometry), a tool that enables scientists without training in machine learning to perform single-cell phenotype classification using deep learning and other advanced downstream data analytics. Interoperability of CellProfiler with popular notebook tools like Jupyter would allow seamless workflows involving other complementary software tools.
Finally, deep learning has revolutionized computer vision and other fields in the past few years [16,32], and bioimaging will be no exception. As noted, already some models trained for specific tasks can be used via CellProfiler, and we expect that over time, more generalizable models will be created that can accomplish useful tasks such as detecting common cellular structures across diverse types of images and experimental setups, as in, for example, the 2018 Data Science Bowl challenge. Community-driven collections of images and ground truth, as well as “model zoos,” will be instrumental for this. We have also begun creating libraries (Keras-ResNet [https://github.com/broadinstitute/keras-resnet] and Keras-RCNN [https://github.com/broadinstitute/keras-rcnn]) that will provide the foundation for interfaces that allow biologists to annotate, train, and use deep learning models. We expect that over time, these models will reduce the amount of time biologists spend tuning classical image processing algorithms to identify biological entities of interest in images.
Images were kindly provided by Javier Frias Aldeguer and Nicolas Rivron of Hubrecht Institute for Developmental Biology and Stem Cell Research and Li Linfeng of MERLN Institute for Technology-Inspired Regenerative Medicine. As per Rivron and colleagues [33], mouse embryos (3.5 dpc) were fixed right after isolation from the mother’s uterus. Fixation was performed using 4% PFA in RNAse-free PBS containing 1% acetic acid. ViewRNA ISH Cell Assay kit (cat# QVC0001) was used for performing smFISH on the embryos. The protocol includes steps of permeabilization and protease treatment as well as probes, preamplifier, amplifier, and label hybridizations. Embryos were then mounted in Slowfade reagent (Thermofisher cat# S36937) and directly imaged in a PerkinElmer Ultraview VoX spinning disk microscope in confocal mode by using a 63×/1.40 NA oil immersion lens.
Images were acquired by collaborators from the Allen Institute for Cell Science, Seattle, as per Roberts and colleagues [34]. Briefly, wild-type C (WTC) hiPSCs were cultured in a feeder-free system on tissue culture dishes or plates coated with GFR Matrigel (Corning) diluted 1:30 in cold DMEM/F12 (Gibco). Undifferentiated cells were maintained with phenol red containing mTeSR1 media (85850, STEMCELL Technologies) supplemented with 1% (v/v) penicillin-streptomycin (P/S; Gibco). Cells were not allowed to reach confluency greater than 85% and are passaged every 3–4 days by dissociation into single-cell suspension using StemPro Accutase (Gibco). When in single-cell suspension, cells were counted using a Vi-CELL Series Cell Viability Analyzer (Beckman Coulter). After passaging, cells were replated in mTeSR1 supplemented with 1% P/S and 10 μM ROCK inhibitor (Stemolecule Y-27632, Stemgent) for 24 hours. Media is replenished with fresh mTeSR1 media supplemented with 1% P/S daily. Cells were maintained at 37°C and 5% CO2. Cells were maintained with phenol red–free mTeSR1 media (05876, STEMCELL Technologies) 1 day prior to live cell imaging.
Three to four days after cells are plated and mature and healthy colonies are observed on 96- and 24-well imaging plates, the cells are stained with NucBlue Live ready probe reagent (R37605, ThermoFisher) and CellMask Deep Red plasma membrane stain (C10046, ThermoFisher) to visualize DNA and plasma membrane, respectively. The protocol is available online: http://www.allencell.org/uploads/8/1/9/9/81996008/sop_for_cellmask-and-nucblue_v1.0_1.pdf. Phenol red–free mTeSR1 is preequilibrated to 37°C and 5% CO2. 1X NucBlue solution made in preequilibrated phenol red–free mTeSR1 is spun for 60 minutes at 20,000 g. The 2X and 10X working stocks of CellMask Deep Red lot #1730970 and #1813792, respectively, are made in 1X NucBlue solution. All solutions are kept at 37°C and 5% CO2 until used. The 100 μL and 400 μL of NucBlue solution are added per well of 96-well imaging plates and 24-well imaging plates, respectively, and incubated at 37°C and 5% CO2 for 20 minutes. An equal amount of CellMask Deep Red working stock is added to the wells containing NucBlue solution. Final dye concentrations in the wells are 1X NucBlue and 1X and 5X CellMask Deep Red lots #1730970 and #1813792, respectively. Cells are incubated at 37°C and 5% CO2 for 10 minutes and gently washed with preequilibrated phenol red–free mTeSR1. Fields of view as shown in Fig 4 that are acquired near the edge (and the center as a control) of hiPSC colonies receive an additional photoprotective cocktail treatment which serves to minimize singlet oxygen and free radical formation. The photoprotective cocktail is used at a working concentration of 0.3 U/ml (1:100) OxyFluor as defined by the OxyFluor product insert, with the addition of 10 mM sodium lactate and 1 mM ascorbic acid (OxyFluor OF-0005, Oxyrase).
As per Roberts and colleagues [34], cells were imaged on a Carl Zeiss spinning disk microscope with a Carl Zeiss 20×/0.8 NA plan APOCHROMAT or 100×/1.25 W C-APOCHROMAT Korr UV Vis IR objective, a CSU-X1 Yokogawa spinning disk head, and Hamamatsu Orca Flash 4.0 camera. Microscopes were outfitted with a humidified environmental chamber to maintain cells at 37°C with 5% CO2 during imaging. Cells are imaged immediately following the wash step and for up to 2.5 hours after dye addition on a Zeiss spinning disk microscope at 100× with the following general settings: 405 nm at 0.28 mW, 200 ms exposure; 638 nm at 2.4 mW, 200 ms exposure; acquiring each channel at each z-step.
Experienced bioimage analysts drew outlines around nuclear boundaries on each slice of the 3D images and labeled background regions in a different color with GIMP (https://www.gimp.org), an open-source drawing and annotation software. These annotated layers were then exported from GIMP as an image. This outline image is converted to 3D objects via a CellProfiler pipeline (https://github.com/CellProfiler/tutorials/tree/master/Annotation), and an object label matrix image is exported, in which each object’s voxels are assigned a unique integer value. These label images are referenced as ground truth.
|
10.1371/journal.ppat.1005658 | Cysteine Peptidase B Regulates Leishmania mexicana Virulence through the Modulation of GP63 Expression | Cysteine peptidases play a central role in the biology of Leishmania. In this work, we sought to further elucidate the mechanism(s) by which the cysteine peptidase CPB contributes to L. mexicana virulence and whether CPB participates in the formation of large communal parasitophorous vacuoles induced by these parasites. We initially examined the impact of L. mexicana infection on the trafficking of VAMP3 and VAMP8, two endocytic SNARE proteins associated with phagolysosome biogenesis and function. Using a CPB-deficient mutant, we found that both VAMP3 and VAMP8 were down-modulated in a CPB-dependent manner. We also discovered that expression of the virulence-associated GPI-anchored metalloprotease GP63 was inhibited in the absence of CPB. Expression of GP63 in the CPB-deficient mutant was sufficient to down-modulate VAMP3 and VAMP8. Similarly, episomal expression of GP63 enabled the CPB-deficient mutant to establish infection in macrophages, induce the formation of large communal parasitophorous vacuoles, and cause lesions in mice. These findings implicate CPB in the regulation of GP63 expression and provide evidence that both GP63 and CPB are key virulence factors in L. mexicana.
| The parasite Leishmania mexicana expresses several cysteine peptidases of the papain family that are involved in processes such as virulence and evasion of host immune responses. The cysteine peptidase CPB is required for survival within macrophages and for lesion formation in susceptible mice. Upon their internalization by macrophages, parasites of the L. mexicana complex induce the formation of large communal parasitophorous vacuoles in which they replicate, and expansion of those large vacuoles correlates with the ability of the parasites to survive inside macrophages. Here, we found that CPB contributes to L. mexicana virulence (macrophage survival, formation and expansion of communal parasitophorous vacuoles, lesion formation in mice) through the regulation of the virulence factor GP63, a Leishmania zinc-metalloprotease that acts by cleaving key host cell proteins. This work thus elucidates a novel Leishmania virulence regulatory mechanism whereby CPB controls the expression of GP63.
| The protozoan Leishmania parasitizes macrophages and causes a spectrum of human diseases ranging from self-healing cutaneous lesions to a progressive visceral infection that can be fatal if left untreated. Infection is initiated when promastigote forms of the parasite are inoculated into the mammalian host by infected sand flies and are internalized by phagocytes. Inside macrophages, promastigotes differentiate into amastigotes to replicate within phagolysosomal compartments also known as parasitophorous vacuoles (PVs). Upon their internalization, L. donovani and L. major promastigotes arrest phagolysosomal biogenesis and create an intracellular niche favorable to the establishment of infection and to the evasion of the immune system [1, 2]. Disruption of the macrophage membrane fusion machinery through the action of virulence factors plays an critical role in this PV remodeling. Hence, insertion of the promastigote surface glycolipid lipophosphoglycan (LPG) into the PV membrane destabilizes lipid microdomains and causes exclusion of the membrane fusion regulator synaptotagmin V from the PV [2–4]. Similarly, the parasite GPI-anchored metalloprotease GP63 [5, 6] redistributes within the infected cells and cleaves key Soluble NSF Attachment Protein Receptors (SNAREs) and synaptotagmins to impair phagosome functions [1, 7].
Whereas L. major and L. donovani multiply in tight individual PVs, parasites of the L. mexicana complex (L. mexicana, L. amazonensis) replicate within large communal PVs. Relatively little is known about the host and parasite factors involved in the biogenesis and expansion of those communal PVs. Studies with L. amazonensis revealed that phagosomes containing promastigotes fuse extensively with late endosomes/lysosomes within 30 minutes post-infection [8]. At that stage, parasites are located within small individual compartments and by 18 to 24 hours large PVs containing several parasites are observed. The rapid increase in the size of those PVs requires extensive fusion with secondary lysosomes and correlates with the depletion of those organelles in infected cells [9–11]. Homotypic fusion between L. amazonensis-containing PVs also occurs, but its contribution to PV enlargement remains to be further investigated [12]. These studies highlighted the contribution of the host cell membrane fusion machinery in the biogenesis and expansion of large communal PVs and are consistent with a role for endocytic SNAREs in this process [13]. Interestingly, communal PVs interact with the host cell’s endoplasmic reticulum (ER) and disruption of the fusion machinery associated with the ER and Golgi inhibits parasite replication and PV enlargement [14–16].
The Leishmania-derived molecules involved in the expansion of the communal PVs remains to be identified. LPG and other phosphoglycans do not play a significant role in L. mexicana promastigote virulence and PV formation [17], in contrast to L. major and L. donovani [2]. Cysteine peptidases (CP) are a large family of papain-like enzymes that play important roles in the biology of Leishmania [18]. Three members of these papain-like proteases are expressed by L. mexicana and the generation of CP-deficient mutants revealed that CPB contributes to the ability to infect macrophages and to induce lesions in BALB/c mice [19–21]. The precise mechanism(s) by which CPB participates in the virulence of L. mexicana is poorly understood. Previous studies revealed that CPB traffics within and outside infected macrophages [18]. In infected macrophages, CPB alters signal transduction and gene expression through the activation of the protein tyrosine phosphatase PTP-1B and the cleavage of transcription factors responsible for the expression of genes involved in host defense and immunity [20, 22]. The observation that CPs interfere with the host immune response through the degradation of MHC class II molecules and invariant chains present in PVs housing L. amazonensis [23], raises the possibility that CPB participates in the modulation of PV composition and function.
In this study, we sought to gain insight into the mechanism by which CPB contributes to L. mexicana virulence, with a focus on the PV. We provide evidence that CPB participates in PV biogenesis and virulence through the regulation of GP63 expression.
Formation and expansion of communal PVs hosting L. mexicana involve fusion between PVs and endocytic organelles, as well as homotypic fusion among PVs [10–12]. To identify the host and parasite factors involved in this process, we embarked on a study to elucidate the fate of endosomal SNAREs during infection of macrophages with L. mexicana. Given the requirement of CPB for L. mexicana to replicate normally inside macrophages [19], we included a L. mexicana CPB-deficient mutant (Δcpb) in our investigation. We infected BMM with either WT or Δcpb L. mexicana promastigotes for 2 h and we assessed the distribution of the endosomal SNAREs VAMP3 and VAMP8 by confocal immunofluorescence microscopy. As previously observed during infection with L. major promastigotes [1], we found a notable reduction in the staining intensity for both VAMP3 (Fig 1A) and VAMP8 (Fig 1B) in BMM infected with WT L. mexicana, but this was not observed with Δcpb. This reduction in staining intensity correlated with a down-modulation of VAMP3 and VAMP8 proteins in BMM infected with WT L. mexicana, compared to cells infected with Δcpb (Fig 1C). These results suggested that L. mexicana causes the reduction of VAMP3 and VAMP8 levels in infected BMM through the action of CPB. However, we considered the possibility that CPB acted indirectly on VAMP3 and VAMP8 because we previously found that GP63 targets those SNAREs in L. major-infected BMM [1]. We therefore ensured that similar levels of GP63 were present in lysates of BMM infected with WT and Δcpb L. mexicana promastigotes. As shown in Fig 2, GP63 was detected in lysates of BMM infected with WT L. mexicana up to 72 h post-infection, when the parasites replicate as amastigotes. The important reduction in GP63 levels at this time point is consistent with previously published data showing a 90% reduction in the amount of GP63 detected in amastigotes with respect to promastigotes [24, 25]. Surprisingly, we found that GP63 was barely detectable in BMM infected with Δcpb at all time points tested. This observation raised the possibility that the lack of VAMP3 and VAMP8 down-regulation in Δcpb-infected BMM was due to defective expression of GP63.
To address the issue of GP63 down-regulation in L. mexicana Δcpb, we first determined whether complementation of Δcpb with the CPB gene array (Δcpb+CPB) restores wild type GP63 levels. As shown in Fig 3A, GP63 levels and activity are down-modulated in the Δcpb mutant, and complementation with the CPB array restored GP63 levels and activity similar to those observed in WT parasites. It was previously reported that expression of the cell surface glycolipid LPG and of GP63 may share common biosynthetic steps [26–29]. We therefore evaluated the levels of LPG in lysates of WT, Δcpb, Δcpb+CPB, and Δcpb+GP63 parasites by Western blot analysis. Strikingly, similar to GP63, LPG levels were also down-modulated in the Δcpb mutant and complementation with either the CPB array or GP63 restored wild type LPG levels. To further investigate the possible role of CPB in the regulation of GP63 expression, we determined the levels of GP63 mRNA in WT, Δcpb, Δcpb+CPB, and Δcpb+GP63 parasites by RT-PCR. As shown in Fig 3B, GP63 mRNA levels were highly down-regulated in Δcpb and complementation with the CPB array restored wild type levels of GP63 mRNA. Interestingly, complementation of Δcpb with L. major GP63 did not increase endogenous GP63 mRNAs. RT-PCR using L. major GP63-specific primers showed that this gene is expressed only in Δcpb+GP63. Together, these results suggest that CPB controls GP63 mRNA levels at the post-transcriptional level. Clearly, additional studies will be required to elucidate the underlying mechanism(s). Our results also raised the possibility that down-modulation of GP63 in the Δcpb mutant may have accounted for the inability of Δcpb to down-regulate VAMP3 and VAMP8. The finding that expression of GP63 in Δcpb restored LPG levels was unexpected and suggested a role for GP63 in the expression of LPG in L. mexicana. As it is estimated that at least 25 genes are required for the synthesis, assembly, and transport of the various components of LPG [30], it may be difficult to determine whether GP63 acts on the expression of a LPG biosynthetic gene or on a biosynthetic step. Assessment of LPG2 gene expression revealed that it was equally expressed WT, Δcpb, Δcpb+CPB, and Δcpb+GP63 parasites. Further studies will be necessary to understand how GP63 expression restores LPG synthesis in Δcpb. Since LPG does not play a major role in the virulence of L. mexicana [17], the Δcpb mutant expressing exogenous GP63 provides a unique opportunity to address the impact of GP63 on SNARE cleavage, as well as on the in vitro and in vivo virulence of L. mexicana.
We next assessed the impact of GP63 on VAMP3 and VAMP8 during L. mexicana infection. To this end, we infected BMM with either WT, Δcpb, Δcpb+CPB, or Δcpb+GP63 L. mexicana promastigotes for various time points, and we assessed VAMP3 and VAMP8 levels and intracellular distribution. Fig 4A shows that GP63 is present at high levels in lysates of BMM infected for 2 h with WT, Δcpb+CPB, and Δcpb+GP63 promastigotes (compared to lysates of BMM infected with Δcpb). At 72 h post-infection, GP63 levels are strongly reduced in BMM infected with WT and Δcpb+CPB, whereas they remain elevated in BMM infected with the Δcpb+GP63 (Fig 4A) [25]. The high levels of GP63 present in BMM infected with Δcpb+GP63 for 72 h may be related to the fact that expression of the L. major GP63 gene from the pLEXNeo episomal vector [31] is not under the control of endogenous GP63 3' untranslated regions. Western blot analyses revealed that down-regulation of VAMP3 and VAMP8 correlated with GP63 levels expressed by the parasites. Consistently, the staining intensity of VAMP3 and VAMP8 was reduced in BMM infected with GP63-expressing parasites, as assessed by confocal immunofluorescence microscopy (Fig 4D and 4E). These results suggest that GP63 is responsible for the down-modulation of the endosomal SNAREs VAMP3 and VAMP8 in L. mexicana-infected BMM. Interestingly, we observed recruitment of VAMP3 to PVs containing L. mexicana parasites at later time points, when promastigotes have differentiated into amastigotes, with the exception of Δcpb+GP63 L. mexicana promastigotes (Fig 4B). In contrast, we found that VAMP8 is excluded from L. mexicana-containing PVs both at early and later time points post-infection, independently of GP63 levels, suggesting that additional mechanisms regulate VAMP8 recruitment to L. mexicana PVs.
Since GP63 was shown to contribute to L. major virulence [32], we next sought to determine whether expression of GP63 is sufficient to restore the ability of Δcpb to replicate inside macrophages and to cause lesions in mice [19]. To this end, we first infected BMM with either WT, Δcpb, Δcpb+CPB, or Δcpb+GP63 stationary phase promastigotes and we assessed parasite burden and PV surface area at various time points post-infection. We found that Δcpb had an impaired capacity to replicate inside macrophages and to induce the formation of large communal PVs compared to WT and Δcpb+CPB parasites (Fig 5A, 5B and 5C). Strikingly, expression of GP63 in Δcpb restored its ability to replicate in macrophages and to induce large communal PVs up to 72 h post-infection. These results underline the role of GP63 in the ability of L. mexicana to infect and replicate in macrophages, even in the absence of CPB. Following inoculation inside the mammalian host, promastigotes are exposed to complement and both GP63 and LPG confer resistance to complement-mediated lysis [32, 33]. L. mexicana promastigotes were therefore analyzed for their sensitivity to complement-mediated lysis in the presence of fresh human serum. As shown in Fig 6A, over 40% of Δcpb was killed after 30 min in the presence of 20% serum, whereas Δcpb+CPB, and Δcpb+GP63 were more resistant to serum lysis at 14% and 10%, respectively. Absence of both GP63 and LPG may be responsible for the serum sensitivity of Δcpb. Finally, to assess the impact of GP63 on the ability of Δcpb to cause lesions, we used a mouse model of cutaneous leishmaniasis. Susceptible BALB/c mice were infected in the hind footpad with either WT, Δcpb, Δcpb+CPB, or Δcpb+GP63 promastigotes and disease progression was monitored for 9 weeks. Consistent with its reduced capacity to replicate inside macrophages, Δcpb failed to cause significant lesions compared to WT parasites [19] and Δcpb complemented with CPB (Fig 6B). Remarkably, expression of GP63 in Δcpb restored its capacity to cause lesions, albeit to a lower level than Δcpb complemented with CPB. Lesion size correlated with parasite burden, as measured at 9 weeks post-infection (Fig 6C). Collectively, these results indicate that expression of GP63 is sufficient to restore virulence of Δcpb.
This study aimed at investigating the mechanism(s) by which CBP contributes to L. mexicana virulence. To this end, we initially examined PV biogenesis by assessing the impact of L. mexicana infection on the trafficking of VAMP3 and VAMP8, two endocytic SNAREs associated with phagosome biogenesis and function [1, 34]. We found that both SNAREs were down-modulated in a CPB-dependent manner, which hampered VAMP3 recruitment to PVs. We also discovered that expression of GP63, which we previously showed to be responsible for cleaving SNAREs in L. major-infected macrophages [1], was down-modulated in the L. mexicana Δcpb. Strikingly, restoration of GP63 expression in Δcpb bypassed the need for CPB for SNARE cleavage. Similarly, episomal expression of GP63 enabled the Δcpb mutant to establish infection in macrophages, induce larger PVs and cause lesions in mice. These findings imply that CPB contributes to L. mexicana virulence in part through the regulation of GP63 expression, and provide evidence that GP63 is a key virulence factor for L. mexicana.
The observation that CPB regulates GP63 expression at the mRNA levels was both unexpected and intriguing. Insight into the possible mechanism(s) may be deduced from a recent study on the role of cathepsin B in L. donovani, which is homologous to the L. mexicana CPC [35]. Similar to L. mexicana Δcpb, L. donovani ΔcatB displays reduced virulence in macrophages. To investigate the role of cathepsin B in virulence, the authors performed quantitative proteome profiling of WT and ΔcatB parasites and identified 83 proteins whose expression is altered in the absence of cathepsin B, with the majority being down-modulated [35]. Among those were a group of proteins involved in post-transcriptional regulation of gene expression (RNA stability, processing, translation) [35]. Whether this is the case in Δcpb deserves further investigation. Clearly, a detailed analysis of wild-type and Δcpb parasites may provide the information required to understand the extent of the impact of CBP on the expression and synthesis of virulence factors and the exact role of CPB in L. mexicana virulence. The observation that episomal expression of GP63 in Δcpb restored LPG synthesis is an intriguing issue, as it suggests that GP63 acts on a LPG biosynthetic step. This role for GP63 is likely redundant, since L. major Δgp63 promastigotes express LPG levels similar to that of wild type parasites (S1 Fig).
It has been proposed that expansion of the PVs hosting parasites of the L. mexicana complex leads to the dilution of the microbicidal effectors to which the parasites are exposed, thereby contributing to parasite survival [36]. Both host and parasite factors may be involved in the control of PV enlargement. On the host side, it has been previously reported that L. amazonensis cannot survive in cells overexpressing LYST, a host gene that restricts Leishmania growth by counteracting PV expansion [37]. Similarly, disrupting the fusion between PVs housing L. amazonensis and the endoplasmic reticulum resulted in limited PV expansion and inhibition of parasite replication [15, 16]. On the parasite side, virulence of L. amazonensis isolates was shown to correlate with the ability to induce larger PVs [38]. Our results indicate that the inability of Δcpb to multiply inside macrophages is associated with smaller PV size, and that expression of GP63 is sufficient to restore the capacity of Δcpb to survive within macrophages and to induce PV expansion. How does GP63 modulate L. mexicana virulence and PV expansion? In addition to the numerous macrophage proteins known to be targeted by GP63, it is possible that SNARE cleavage is one of the factors associated with L. mexicana virulence and PV expansion. For instance, we previously reported that VAMP8 is required for phagosomal oxidative activity [1]. One may envision that its degradation by GP63 is part of a strategy used by L. mexicana to establish infection in an environment devoid of oxidants, thereby contributing to parasite survival. The LYST protein is a regulator of lysosome size and its absence leads to further PV expansion and enhanced L. amazonensis replication [37]. It is interesting to note that LYST was proposed to function as an adaptor protein that juxtaposes proteins such as SNAREs that mediate intracellular membrane fusion reactions [39]. In this context, cleavage of SNAREs that interact with LYST may interfere with its function and promote PV expansion. Further studies will be necessary to clarify these issues, including the potential role of VAMP3 and VAMP8 in PV biogenesis and expansion.
Previous studies using Δcpb parasites led to the conclusion that CPB enables L. mexicana to alter host cell signaling and gene expression through the cleavage of various host proteins [20, 22]. Hence, CPB-dependent cleavage of PTP-1B, NF-κB, STAT1, and AP1 in L. mexicana-infected macrophages was associated with the inhibition of IL-12 expression and generation of nitric oxide, both of which are important for initiation of a host immune response and parasite killing, respectively. Our finding that GP63 expression is down-modulated in the Δcpb mutant raises the possibility that cleavage of those transcription factors may actually be mediated by GP63. Indeed, GP63 cleaves numerous host macrophage effectors, including PTP-1B, NF-κB, STAT1, and AP1 [40]. Revisiting the role of CPB in the context of GP63 expression will be necessary to elucidate whether, and to which extent, CPB is acting directly on the host cell proteome.
In sum, we discovered that CPB contributes to L. mexicana virulence in part through the regulation of GP63 expression. Complementation of Δcpb revealed the importance of GP63 for the virulence of L. mexicana, as it participates in the survival of intracellular parasites, PV expansion, and the formation of cutaneous lesions.
Experiments involving mice were done as prescribed by protocol 1406–02, which was approved by the Comité Institutionnel de Protection des Animaux of the INRS-Institut Armand-Frappier. In vivo infections were performed as per Animal Use Protocol #4859, which was approved by the Institutional Animal Care and Use Committees at McGill University. These protocols respect procedures on good animal practice provided by the Canadian Council on Animal Care (CCAC).
The mouse anti-GP63 monoclonal antibody was previously described [41]. The mouse anti-phosphoglycans CA7AE monoclonal antibody [42] was from Cedarlane and the rabbit polyclonal anti-aldolase was a gift from Dr. A. Jardim (McGill University). Rabbit polyclonal antibodies for VAMP3 and VAMP8 were obtained from Synaptic Systems.
Bone marrow-derived macrophages (BMM) were differentiated from the bone marrow of 6- to 8-week-old female 129XB6 mice (Charles River Laboratories) as previously described [43]. Cells were cultured for 7 days in complete medium (DMEM [Life Technologies] supplemented with L-glutamine [Life Technologies], 10% heat-inactivated FBS [PAA Laboratories], 10 mM HEPES at pH 7.4, and antibiotics) containing 15% v/v L929 cell–conditioned medium as a source of M-CSF. Macrophages were kept at 37°C in a humidified incubator with 5% CO2. To render BMM quiescent prior to experiments, cells were transferred to 6- or 24-well tissue culture microplates (TrueLine) and kept for 16 h in complete DMEM without L929 cell–conditioned medium. Promastigotes of L. mexicana wild-type strain (MNYC/BZ/62/M379) and of L. major NIH S (MHOM/SN/74/Seidman) clone A2 were grown at 26°C in Leishmania medium (Medium 199 supplemented with 10% heat-inactivated FBS, 40 mM HEPES pH 7.4, 100 μM hypoxanthine, 5 μM hemin, 3 μM biopterin, 1 μM biotin, and antibiotics). The isogenic L. mexicana CPB-deficient mutant Δcpbpac (thereafter referred to as Δcpb) and its complemented counterpart Δcpbpac[pGL263] (thereafter referred to as Δcpb+CPB) were described previously [21]. L. mexicana Δcpb promastigotes were electroporated as described [44] with the pLEXNeoGP63.1 plasmid [32] to generate Δcpb+GP63 parasites. L. mexicana Δcpb+CPB and Δcpb+GP63 promastigotes were grown in the presence of 50 μg/ml hygromycin or 50 μg/ml G418, respectively. The L. major NIH clone A2 isogenic Δgp63 mutant and its complemented counterpart Δgp63+gp63 have been previously described [32]. Cultures of Δgp63+gp63 promastigotes were supplemented with 50 μg/ml G418.
Prior to macrophage infections, promastigotes in late stationary phase were opsonized with DBA/2 mouse serum. For synchronized phagocytosis using parasites, macrophages and parasites were incubated at 4°C for 10 min and spun at 167 g for 1 min, and internalization was triggered by transferring cells to 34°C. Macrophages were washed twice after 2h with complete DMEM to remove the non-internalized parasites and were further incubated at 34°C for the required times. Cells were then washed with PBS and stained using the Hema 3 Fixative and Solutions kit (Fisher HealthCare), or prepared for confocal immunofluorescence microscopy.
Macrophages on coverslips were fixed with 2% paraformaldehyde (Canemco and Mirvac) for 40 min and blocked/permeabilized for 17 min with a solution of 0.05% saponin, 1% BSA, 6% skim milk, 2% goat serum, and 50% FBS. This was followed by a 2 h incubation with primary antibodies diluted in PBS. Macrophages were then incubated with a suitable combination of secondary antibodies (anti-rabbit Alexa Fluor 488 and anti-rat 568; Molecular Probes) and DAPI (Life technologies). Coverslips were washed three times with PBS after every step. After the final washes, Fluoromount-G (Southern Biotechnology Associates) was used to mount coverslips on glass slides, and coverslips were sealed with nail polish (Sally Hansen). Macrophages were imaged with the 63X objective of an LSM780 microscope (Carl Zeiss Microimaging), and images were taken in sequential scanning mode. Image analysis and vacuole size measurements were performed with the ZEN 2012 software.
Prior to lysis, macrophages were placed on ice and washed with PBS containing 1 mM sodium orthovanadate and 5 mM 1,10-phenanthroline (Roche). Cells were scraped in the presence of lysis buffer containing 1% Nonidet P-40 (Caledon), 50 mM Tris-HCl (pH 7.5) (Bioshop), 150 mM NaCl, 1 mM EDTA (pH 8), 5 mM 1,10-phenanthroline, and phosphatase and protease inhibitors (Roche). Parasites were washed twice with PBS and lysed in the presence of lysis buffer containing 0.5% Nonidet P-40 (Caledon), 100mM Tris-HCl (Bioshop) and 150 mM NaCl at -70°C. Lysates were thawed on ice and centrifuged for 10 min to remove insoluble matter. After protein quantification, protein samples were boiled (100°C) for 6 min in SDS sample buffer and migrated in 12% or 15% SDS-PAGE gels. Three micrograms and 15 μg of protein were loaded for parasite and infected macrophage lysates, respectively. Proteins were transferred onto Hybond-ECL membranes (Amersham Biosciences), blocked for 1 h in TBS 1X-0.1% Tween containing 5% skim milk, incubated with primary antibodies (diluted in TBS 1X-0.1% Tween containing 5% BSA) overnight at 4°C, and thence with appropriate HRP-conjugated secondary antibodies for 1 h at room temperature. Then, membranes were incubated in ECL (GE Healthcare), and immunodetection was achieved via chemiluminescence. Membranes were washed 3 times between each step. For zymography, 2 μg of lysate were incubated at RT for 6 min in sample buffer without DTT and then migrated in 12% SDS-PAGE gels with 0.2% gelatin (Sigma). Gels were incubated for 1 h in the presence of 50 mM Tris pH 7.4, 2,5% Triton X-100, 5 mM CaCl2 and 1 μM ZnCl2 and incubated overnight in the presence of 50 mM Tris pH 7.4, 5 mM CaCl2, 1μM ZnCl2 and 0,01% NaN3 at 37°C [45].
Late stationary phase promastigotes were incubated for 30 min in complete DMEM medium with 20% human serum from healthy donors. Parasites were then incubated in LIVE/DEAD Fixable Violet Dead Cell Stain Kit (Life technologies) and fixed in 2% paraformaldehyde. Flow cytometric analysis was carried out using the LSRFortessa cytometer (Special Order Research Product; BD Biosciences), and the BD FACSDiva Software (version 6.2) was used for data acquisition and analysis.
Male BALB/c mice (6 to 8 weeks old) were purchased from Charles River Laboratories and infected in the right hind footpad with 5x106 stationary phase L. mexicana promastigotes as described [46]. Disease progression was monitored by measuring footpad swelling weekly with a metric caliper, for up to 9 weeks post-infection. Footpads were then processed to calculate parasite burden using the limiting dilution assay.
After 9 weeks of infection, mice were euthanized under CO2 asphyxiation and subsequently by cervical dislocation. The infected footpads were surface-sterilized with a chlorine dioxide based disinfectant followed by ethanol 70% for 5 min. Footpads were washed in PBS, lightly sliced, transferred to a glass tissue homogenizer containing 6 ml of PBS, and manually homogenized. The last step was repeated two to three times, until complete tissue disruption was achieved. Final homogenate was then centrifuged at 3,000 x g for 5 min and resuspended in the appropriate volume of PBS. 100 μl of homogenate were added in duplicates to 96-well plates containing 100 μl of complete Schneider’s medium in each well (twenty-four 2-fold dilutions for each duplicate). Plates were incubated at 28°C. After 7–10 days, the number of viable parasites was determined from the highest dilutions at which promastigotes were observed using an inverted microscope [47].
Total RNA was extracted from promastigotes using the TRIzol reagent (Invitrogen Life Technology, Carlsbad, CA) and reverse transcribed. One-tenth of the resulting cDNA was amplified by PCR on a DNA thermal cycler, version 2.3 (Perkin-Elmer Corporation, Norwalk, CT), with the following primer pairs: for the L. mexicana GP63 C-1 5'-ACCGTCTGAGAGTCGGAACT-3' (forward), 5'-GTAGTCCAGGAATGGCGAGT-3' (reverse); the L. major GP63-1 5'-TCTGAGGCACATGCTTCGTT-3' (forward), 5'-GTCAGTTGCCTTCGGTCTGA-3' (reverse), the L. mexicana LPG2 5'CATTTGGTATCCTGGTGCTG-3' (forward), 5'-GAGGAAGCCACTGTTAGCC-3' (reverse), and the L. mexicana α-tubulin 5'-CTATCTGCATCCACATTGGC-3' (forward), 5'-ACTTGTCAGAGGGCATGGA-3' (reverse). The PCR products were analyzed by electrophoresis on a 3% (wt/vol) agarose gel, and the pictures were taken using AlphaImager 3400 imaging software (Alpha Innotech Corporation, San Leandro, CA).
Statistically significant differences were analyzed by ANOVA followed by the Tukey post-hoc test using the Graphpad Prism program (version 5.0). For the limiting dilution assay, the non-parametric Mann-Whitney or Kruskal-Wallis test was used. Values starting at P<0.05 were considered statistically significant. All data are presented as mean ± SEM.
|
10.1371/journal.ppat.1002827 | IL-10-Producing Th1 Cells and Disease Progression Are Regulated by Distinct CD11c+ Cell Populations during Visceral Leishmaniasis | IL-10 is a critical regulatory cytokine involved in the pathogenesis of visceral leishmaniasis caused by Leishmania donovani and clinical and experimental data indicate that disease progression is associated with expanded numbers of CD4+ IFNγ+ T cells committed to IL-10 production. Here, combining conditional cell-specific depletion with adoptive transfer, we demonstrate that only conventional CD11chi DCs that produce both IL-10 and IL-27 are capable of inducing IL-10-producing Th1 cells in vivo. In contrast, CD11chi as well as CD11cint/lo cells isolated from infected mice were capable of reversing the host protective effect of diphtheria toxin-mediated CD11c+ cell depletion. This was reflected by increased splenomegaly, inhibition of NO production and increased parasite burden. Thus during chronic infection, multiple CD11c+ cell populations can actively suppress host resistance and enhance immunopathology, through mechanisms that do not necessarily involve IL-10-producing Th1 cells.
| Dendritic cells are well known as myeloid cells that bridge innate and adaptive immunity, and play an important role in the induction of cell-mediated immunity to a variety of pathogens. However, very little is known about the function of dendritic cells after infection has become established. In this study, we have examined the role of dendritic cells during the later phases of experimental visceral leishmaniasis, caused by infection with Leishmania donovani. We show for the first time that dendritic cells are responsible for promoting the development of splenic pathology, and that their removal during established infection leads to improved host resistance. Furthermore, our studies provide the first formal evidence in vivo that dendritic cells making IL-27 can induce production of the regulatory cytokine IL-10 by effector Th1-like CD4+ T cells. Surprisingly, we also found that other populations of CD11c+ cells were able to induce pathology and suppress host resistance, yet did not stimulate IL-10 production in CD4+ T cells, suggesting that the latter T cell population may not play an essential role in disease progression. Our studies provide new insights into dendritic cell function in chronic parasite infection and suggest potential new avenues for immunotherapy against visceral leishmaniasis.
| Dendritic cells (DCs) are widely recognized as being the most important myeloid cell involved in antigen presentation and the initiation and regulation of CD4+ T cell-dependent protective immunity against a variety of intracellular parasites (reviewed in [1], [2]), and show promise for the development of new approaches in vaccination and immunotherapy [3], [4]. Initially based largely on in vitro studies, the key role of DCs in antigen presentation has been borne out in recent years through the availability of mice in which DCs can be ablated in a conditional manner [5]. Hence, diphtheria toxin (DTx)-mediated ablation of DCs results in a significant reduction in T cell priming following various infectious challenges, including with Mycobacterium tuberculosis, Plasmodium yoelli, Listeria monocytogenes, Streptococcus pyogenes and LCMV [6], [7], [8], [9]. In contrast, the role of DCs during later stages of infection and their contribution to the immune imbalance that is often associated with chronic infection are less well understood, in spite of the known ability of DCs to induce tolerogenic or regulatory responses [4], [10], [11], [12].
CD11c+ DCs play multiple roles in the pathogenesis of leishmaniasis, including experimental visceral leishmaniasis (EVL) caused by Leishmania donovani (reviewed in [13]). Dermal DC [14] and Langerhans cells [15] have been implicated in the early stages of L. major infection, and as this infection progresses, many parasites are found in the draining LN within CD11c+ cells that resemble TipDCs [16]. Expression of MHCII on DCs is both necessary and sufficient for the induction of effective immunity to L. major, suggesting macrophage antigen presentation may not be required for effector T cell function [17]. In EVL, splenic DCs belonging to the CD8α subset of conventional DCs (cDCs) are the first cells to produce IL-12 within the splenic microenvironment [18], and are activated for heightened expression of a variety of costimulatory molecules through both direct interactions with Leishmania parasites and through inflammatory signals [19]. In chronic EVL, however, cDC cytokine production is modulated in a subset-specific manner [18] and migration through lymphoid tissue is disrupted [20]. In addition, CD11c expression is found on other cells known to contribute to anti-leishmanial resistance, including NK cells [21], and inflammatory monocytes/TipDCs [16]. However, the relative contribution of these different CD11c+ cell populations to disease progression and the regulation of T cell effector and regulatory function is poorly understood.
Visceral leishmaniasis is also noted for the production of the immunoregulatory cytokine IL-10, and targeting of IL-10 signaling has been identified as a potential therapeutic strategy [22]. Although multiple cellular sources of IL-10 have been identified in VL, the identification of a population of IFNγ-producing CD4+ T cells that also produces IL-10 and its association with progressive disease in both mice [23], [24] and in humans [25] has drawn particular attention. The co-production of IL-10 by IFNγ-producing CD4+ T cells is not novel for leishmaniasis, however, and is now a recognized feature of Th1 cell differentiation. Considerable attention has been focused, therefore, on dissecting the molecular signals required for expression of this mixed effector/regulatory phenotype. In vitro studies using transgenic CD4+ T cells and repeated exposure to antigen and APCs have suggested that the induction of IL-10 is a consequence of sustained antigen presentation, requiring the presence of high levels of IL-12 [26]. The cytokine IL-27 is also implicated in the generation of IL-10-producing CD4+ T cells in vitro [27], [28], [29], [30], [31], via signaling pathways dependent on STAT3 [29], and optimal generation of CD4+IL-10+ cells by IL-27 requires the co-ordinate initiation of c-Maf, ICOS and IL-21 expression [32], [33]. In addition, emerging evidence suggests that IL-27 may directly alter methylation patterns around the il10 promoter in CD4+ T cells, thus allowing greater IL-10 expression [34]. IL-27 also favors the production of IL-10 by IFNγ-producing Th1 cells through an alternate signaling pathway that involves STAT1, STAT4 and Notch [35], [36]. In spite of these advances, the cellular sources of IL-27 in vivo have been poorly defined. A direct role for DC-derived IL-27 in the generation of IL-10+ T cells has been described in vitro, where production of this cytokine in response to galectin-1, and during ovalbumin-induced oral tolerance, both favored the differentiation of IL-10-producing T cells with potent regulatory capacity [37], [38]. Nevertheless, the cellular requirements for generating CD4+IFNγ+IL-10+ T cells in vivo remain obscure and no studies to date have addressed this question in the context of chronic infection.
We therefore sought to address two inter-related questions: i) what is the role of CD11c+ cells during chronic L. donovani infection and ii) do these cells contribute to the emergence of IL-10-producing Th1 cells. Given recent concerns over off target effects induced by DTx treatment of mice [8], [39], we used a functional complementation approach to independently examine the role of CD11chi cDCs and CD11cint/lo cells in determining host resistance and Th1 cytokine production. We show that CD11chi cDCs, as well as a mixed CD11cint/lo cell population, are capable of inhibiting host resistance and promoting disease-associated pathology. Our study also provides the first formal evidence that IL-10+IL-27+ cDCs are able to promote IL-10 production by Th1 cells in vivo. Our data therefore highlight CD11c+ cells as potential targets for immunotherapy and also demonstrate an important discordance between disease progression and the emergence of IL-10-producing Th1 cells.
C57BL/6 mice infected with L. donovani amastigotes developed pronounced splenomegaly from day 21 post infection (p.i.) (Figure 1A), associated with an increasing tissue parasite burden (Figure 1B). As assessed by polyclonal activation ex vivo (Figure 1C and D), the frequency of splenic CD3ε+CD4+ IFNγ+ T cells increased from 2.4±0.4% in naïve mice to 44.4±3.0% (p<0.001) by day 28 of infection which, when taking into account splenomegaly, reflected a >500-fold increase in absolute number of cells committed to IFNγ production in the spleen. Infection was also associated with emergence of a population of splenic CD3ε+CD4+ T cells capable of producing both IFNγ and IL-10. This population increased in frequency ∼50 fold, comprising 0.1±0.02% of CD3ε+CD4+ cells in naïve mice and 5.0±0.9% at day 28 of infection (p<0.001), equating to a ∼170 fold expansion in the total number of splenic T cells capable of simultaneous production of IFNγ and IL-10 (7.6±1.2×103 vs 1.3×106±1.7×105 in naïve and day 28 infected mice, respectively; p<0.001). Whilst the frequency and number of CD3ε+CD4+ T cells producing IL-10 alone also increased, this population of IL-10-producing cells remained modest compared to those that also made IFNγ. Similar results were also obtained following in vitro re-stimulation of splenic CD4+ T cells using Leishmania-antigen pulsed BMDCs (Figure 1E and F).
In order to determine the lineage origin of these IFNγ+IL-10+ CD4+ T cells, cytokine producing cells (Figure 1G) were examined for intracellular expression of the Th1-associated transcription factor Tbx21 (T-bet), the regulatory T cell-associated transcription factor forkhead box transcription factor 3 (Foxp3), and for surface expression of the IL-7 receptor alpha chain (CD127). CD3ε+CD4+ T cells capable of simultaneous production of IFNγ and IL-10 were exclusively T-bet+, CD127− and Foxp3− (Figure 1H). In contrast to the expansion of IL-10-producing CD4+ T cells, the frequency of splenic Foxp3+ Treg did not increase during infection (Figure S1), but instead decreased as previously reported [23]. Therefore, chronic infection with L. donovani was associated with the expansion of antigen-specific CD4+ T-bet+CD127− Foxp3− ‘Th1-like’ cells capable of simultaneously producing IFNγ and IL-10.
To address the cellular mechanisms underlying the expansion of CD4+IFNγ+IL-10+ T cells, we focused on alterations within the splenic CD11chiMHCIIhi cDC compartment (Figure 2A). First, we characterized the entire cDC population for the expression of cell surface ligands associated with costimulation. CD80 and CD86 expression was not changed compared to that on cDCs from naïve mice and only a 2–2.5 fold induction of CD40 and PD-L1 expression was noted on cDCs at day 28 p.i., relative to naïve mice (Figure 2B and Figure S2). The three major lymphoid-resident cDC subsets, as determined by CD4 and CD8 expression (Figure 2A), were all found at the expected frequencies [40], but individual subsets showed some level of differential regulation of co-stimulatory molecule expression during infection (Figure 2C–F). Of note, CD8α+ cDCs showed the least activation as judged by CD40, CD80 and CD86 expression, yet conversely had the greatest increase in PD-L1 expression.
In light of the critical role for APC-derived cytokines in shaping CD4+ T cell lineage commitment, we next sought to determine how chronic infection impacted upon cDC cytokine production ex vivo. Highly purified CD11chiMHCIIhi cDCs from spleens of naïve and day 28-infected mice (Figure 3A) showed differential cytokine profiles at both the whole population level and when separated into distinct subsets. CD11chiMHCIIhi cDCs from day 28 infected mice had significantly reduced levels of spontaneous and LPS-induced IL-12p70 secretion, when compared to cDCs from naïve mice (p<0.01; Figure 3B). Similar results were also obtained from isolated cDC subsets by analysis of IL-12p70 [18] and IL-12/23p40 secretion (Figure 3C and D). In contrast, cDC production of IL-27p28 was significantly enhanced during infection, both at the bulk population level (Figure 3E) and when assessed for each individual cDC subset (Figure 3F). Similarly, cDCs from infected mice also produced elevated levels of IL-10 when compared to cDCs from naïve mice (Figure 3G and H). Although more pronounced by day 28 p.i., similarly altered cytokine responses have been observed in the early stages of acute L. donovani infection [18].
Autocrine IL-10 signaling is known to influence DC cytokine production, with splenic cDCs particularly sensitive to this form of regulation [41], [42]. Therefore, we next sought to determine whether the altered cDC cytokine profile was in part dependent on autocrine IL-10 and/or IL-27 production. cDCs from infected mice cultured in the presence of αIL-10R mAb spontaneously secreted IL-12p70 to a similar extent to those treated with LPS (Figure 4A). However, maximal IL-12p70 production was achieved by simultaneous IL-10R blockade and LPS stimulation. IL-10R blockade also significantly enhanced the accumulation of IL-10 in the culture medium, again most pronounced when combined with LPS stimulation (Figure 4B). These data indicate a potent negative regulatory function for autocrine IL-10 produced by cDCs isolated from infected mice. In contrast, neutralization of IL-27p28 alone had no impact on ex vivo IL-12p70 or IL-10 production by cDCs isolated from infected mice, nor was there any additive effect when combined with IL-10R blockade (Figure 4A and B). Hence, IL-27p28 does not auto-regulate cDC cytokine production under these conditions.
Finally, we examined whether the immunomodulators TGFβ or Indoleamine 2,3-dioxygense (IDO) were substantially regulated in cDC subsets as a result of infection with L. donovani. At the transcriptional level, only CD4+ cDCs from infected mice showed any significant increase in accumulation of Tgfβ mRNA (Figure S3A) and in no population of cDC did we observe any accumulation of Ido mRNA as a result of infection (Figure S3B). In summary, therefore, chronic L. donovani infection is associated with muted co-stimulatory molecule expression, increased IL-10 and IL-27p28 production and a dramatic impairment in IL-12p70 production by splenic cDCs, with IL-12p70 secretion regulated in part by autocrine IL-10 signaling.
To assess the in vivo impact of cDCs during chronic infection, we generated (CD11c-cre×Rosa26iDTR)F1 mice. In these mice, expression of Cre recombinase is driven by CD11c promoter activity, resulting in cleavage at loxP sites flanking a ubiquitously expressed STOP cassette upstream of a simian diphtheria toxin receptor (DTR). Past or current expression of CD11c initiates DTR expression and thus provides inducible sensitivity to diphtheria toxin (DTx). We administered either saline or DTx i.p. to (CD11c-cre×Rosa26iDTR)F1 mice at 48 hour intervals for a period of 7 days, beginning on day 21 p.i. (Figure 5A). Unlike in CD11c-DTR mice [5], we found no evidence of toxicity using this regimen. ∼80–90% of CD11chiMHCIIhi cDCs were ablated after 7 days of treatment (Figure 5B and C). Depletion was almost 100% complete for the CD4+ and CD8α+ subsets, with most residual cDCs belonging to the DN subset (Figure 5D–F). In addition to depletion of cDCs, we also observed depletion of some CD11cint/lo cells (Figure 5B and C). ∼20% of CD11cint/loMHCII− cells and 40% of CD11cint/loMHCIIhi cells were lost, most likely including both CD11cint/lo DCs and NK cells [21], [43]. Additional off target effects of DTx treatment were also noted. CD169+ marginal zone macrophages were depleted, as determined by immunofluorescence staining of tissue sections (data not shown). Although others have observed loss of marginal zone macrophages in CD11c-DTR mice [39], these cells are already largely absent in mice infected with L. donovani [44]. Other significant off-target effects of DTx treatment were restricted to a decrease in the frequency and number of NK1.1+CD11b+ NK cells and an increase in the frequency of splenic CD11bhiGr-1hi neutrophils (1.46±0.18% vs. 2.38±0.29% in infected vs. DTx-treated infected mice respectively; p<0.05, Figure S4).
Depletion of CD11c+ cells in chronically infected (CD11c-cre×Rosa26iDTR)F1 mice had a profound impact on splenic pathology, reflected by a dramatically reduced spleen size (from 3.94±0.22% vs. 2.14±0.14% of total body weight in saline-treated and DTx-treated mice, respectively; p<0.0001). In contrast, treatment of naïve (CD11c-cre×Rosa26iDTR)F1 mice with DTx had no impact on spleen size (Figure 6A). Ablation of CD11c+ cells also significantly reduced splenic parasite burden (101±23 vs. 27±9 LDU in saline vs. DTx treated mice, respectively; p<0.05; Figure 6B). Conversely, mice treated with DTx had an almost 5-fold increase in nitric oxide production, as measured by spontaneous release from adherent splenocytes (23.04±2.51 vs. 92.36±21.89 µM control and DTx treated mice; p<0.05). NO production was not detectable from adherent splenocytes in naïve animals, irrespective of DTx treatment (Figure 6C). By each of these criteria, therefore, CD11c+ cells appeared to be playing a role in promoting disease progression in vivo.
The experimental approach outlined above provided an opportunity to determine whether there was a causal link between the appearance of the phenotypically distinct cDCs described above and the induction of IL-10-producing Th1 cells. We therefore examined CD4+ T cells from these DTx-treated mice for their capacity to produce IFNγ and IL-10 (Figure 7). Depletion of CD11c+ cells from day 21 to day 28 of infection did not affect the frequency of antigen-specific IFNγ+ T cells (Figure 7A; 35.78±4.18% vs. 30.33±5.32% after saline or DTx, respectively), although absolute numbers were decreased by approximately 2-fold in keeping with the reduction of spleen size (Figure 7D). In contrast, the frequency of antigen-specific CD4+ T cells capable of the simultaneous production of IFNγ and IL-10 was significantly reduced following DTx administration (Figure 7C; 2.62±0.31% vs. 1.28±0.11% in saline and DTx treated mice respectively; p<0.001) and the absolute number per spleen was reduced by four fold (Figure 7F). Antigen-specific CD3ε+CD4+ T cells capable only of IL-10 production also showed a trend towards a reduction in frequency, but this was not significant (0.57±0.23% to 0.10±0.03%; p = ns; Figure 7B). The absolute number of these cells in the spleen was significantly reduced, however (Figure 7E). In combination, these data demonstrate that depletion of CD11c+ cells during chronic infection dramatically reduces splenic pathology, allows NO-dependent parasite clearance and impairs the generation of antigen-specific CD3ε+CD4+ T cells capable of simultaneous IL-10 and IFNγ production.
As neutrophils play a role in the control of established L. donovani infection [45] and splenic neutrophil numbers increase after DTx treatment (this study and [46], [47]), we repeated these experiments using DTx-treated infected mice co-treated with a Ly6G-specific mAB (1A8) to deplete neutrophils. The frequency of IFNγ+IL-10+ CD4+ T cells was similar in DTx-treated mice irrespective of whether neutrophils were present or absent (1.69±0.37% vs. 1.89±0.40% in control and 1A8-treated mice respectively). Neutrophil depleted DTx-treated mice also showed a similar reduction in splenomegaly and increased NO production as seen in neutrophil replete DTx-treated mice (Figure 8A and B). As previously noted in wild type mice [45], neutrophil depletion of DTx-treated mice led to increased parasite burden (Figure 9C), illustrating that immunopathology is not strictly associated with parasite load. Nevertheless, changes in neutrophil number cannot account for the changes in Th1 cell differentiation or immunopathology observed in DTx-treated mice. Hence, these data demonstrate that depletion of CD11c+ cells during ongoing infection dramatically reduces splenic pathology, promotes NO-dependent parasite clearance and significantly impairs the generation of antigen-specific CD4+ T cells capable of simultaneous IL-10 and IFNγ production, whilst only slightly reducing the abundance of other Th subsets.
To definitively address whether cDCs or other populations of CD11c+ cells were responsible for the induction of IL-10 production in Th1 cells and for changes to host resistance, we employed a functional complementation approach. We adoptively transferred (in accordance with their population abundance) wild type DTx-resistant CD11chiMHCIIhi cDCs and CD11cint/lo cells from infected B6.CD45.1 mice into infected DTx-treated (CD11c-cre×Rosa26iDTR)F1 mice (Figure 9A and B). cDCs obtained at d21 p.i. had similar patterns of co-stimulatory molecule expression to cDCs obtained at d28 p.i., particularly with regard to PD-L1 expression and this was also similar to the phenotype of CD11cint cells (in spite of heterogeneity within this population; Figure S5). We also characterised CD11cint cells and cDCs for IL-27p28 (Figure S6A) and IL-10 (Figure S6B) mRNA accumulation at d21 p.i. and day 28 p.i. and observed significantly lower accumulation of both cytokines within the CD11cint population, suggesting a sustained difference in their capacity to regulate IL-10 and IL-27 expression during infection. Ablation of endogenous CD11c+ cells was maintained after transfer by repeated DTx treatment. Strikingly, transfer of either CD11cint/lo cells or CD11chi cDCs was sufficient to restore splenomegaly (Figure 9C), parasite burden (Figure 9D) and NO production (Figure 9E) to levels similar to that observed in non-treated infected mice. The potency of these cells to restore disease progression was all the more remarkable given that at the time of assay (d7 post transfer) donor CD45.1 CD11c+ cells could not be detected, suggesting long term engraftment had not occurred (data not shown). Adoptive transfer had no impact on the frequency of IFNγ single-producing CD4+ T cells (in keeping with the limited effect of CD11c depletion on this T cell response) or on the frequency of IL-10+ single producing T cells, although absolute numbers were increased as a result of the changes in splenomegaly after these interventions (not shown). In contrast, adoptive transfer of cDCs, but not CD11cint/lo cells, restored the frequency (Figure 9F) and absolute numbers (Figure 9G) of IFNγ+IL-10+ CD4+ T cells to that observed in untreated infected mice. cDCs are therefore required to promote in vivo expansion of IFNγ+IL-10+ CD4+ T cells during L. donovani infection.
This study is the first to demonstrate that CD11c+ cells act to promote disease progression during the chronic phase of infection with L. donovani. Furthermore, by combining conditional DTx-mediated depletion with adoptive transfer during ongoing infection, we could show that whereas both CD11chi and CD11cint/lo cells contribute to disease progression and suppress host protective immunity, only CD11chi cells are capable of promoting the expansion and/or maintenance of Th1 cells that produce IL-10. In addition to providing the first evidence that cDCs are required to promote expansion of CD4+ T cells with mixed effector/regulatory phenotype in vivo, our data suggests that the emergence of Th1 cells producing IL-10 is not essential for disease progression.
Unlike previous studies assessing the role of DCs in acute L. donovani infection [48] or Langerhans cells in acute L. major infection [15], sustained ablation of CD11c-expressing cells during ongoing infection was required in this study. Under the conditions used, we could deplete 80–90% of splenic CD11chiMHCIIhi cells in mice chronically infected with L. donovani, similar to the efficacy reported for DC depletion during Schistosoma mansoni infection [49]. The impact of CD11c+ cell ablation on disease progression was striking, with reduced splenic pathology, enhanced nitric oxide (NO) production and enhanced parasite clearance. Indeed, the impact of CD11c depletion was of a similar magnitude to that observed after a variety of chemotherapeutic and immunotherapeutic interventions [50]. Neutrophil influx has recently been noted following DTx treatment of mice [46], [47] and was also observed by us here. It is unclear whether CXCL2-mediated egress from the bone marrow underlies the neutrophilia in (CD11c-cre×Rosa26iDTR)F1 mice, as suggested for other strains [47]. In preliminary studies, we have also noted a significant increase in the abundance for Il17α mRNA after DTx treatment (data not shown), but further studies are required to determine whether this cytokine also impacts on neutrophil recruitment. In spite of this influx of phagocytes, simultaneous depletion of neutrophils in vivo demonstrated that even in their absence, the ablation of CD11c+ cells resulted in a marked reduction in splenomegaly and increased levels of NO production. Of note, parasite burden was increased in neutrophil depleted DTx treated mice, even though NO levels also increased, suggesting that the leishmanicidal effect of neutrophils may be mediated through NO-independent mechanisms. Nevertheless, to more directly overcome the pitfalls associated with off-target effects of DTx treatment, and to more formally address the role of cDCs in disease progression, we employed a strategy that allowed simultaneous depletion of endogenous CD11c+ cells and reconstitution with subpopulations of wild type CD11c+ cells. Similar to a recent study of allergic inflammation [51], the transfer of relatively small numbers of highly purified CD11chi cDCs to CD11c-ablated mice resulted in substantial modulation of disease. All parameters of host resistance that we measured were suppressed after CD11c+ cell transfer and splenomegaly returned to the level observed in untreated infected mice. Importantly, we could not distinguish between the effects of transfer of CD11chi cDCs and CD11cint/lo cells based on these criteria.
Splenic cDCs capable of promoting disease progression had a number of characteristics that distinguish them from cDCs found in naïve mice, with only minor differences seen between subsets. CD80 and CD86 expression is muted during chronic infection, similar to what has been observed at early times (∼5 hr) post infection [18], [52]. CD40 expression was somewhat higher at day 28 p.i. than at d21 p.i., but there are conflicting reports as to whether the CD40-CD40L axis is required [53], [54] or redundant [55], [56] with respect to anti-Leishmania responses. It has been suggested that signaling downstream of CD40 may perpetuate IL-10 production and enhance productive infection of macrophages [57] but this has not been evaluated in DCs. cDCs from chronically infected mice showed enhanced expression of Programmed Death Ligand 1 (PD-L1), a negative costimulatory molecule involved in regulating functional exhaustion of CD8+ T cells during L. donovani infection [58]. IL-12p70 production was severely impaired, whereas high levels of spontaneous IL-10 and IL-27 were observed, with some differences between subsets also reflected in previous data at the mRNA level [18]. IL-10 produced by cDCs in infected mice appears to play an important autocrine regulatory role in limiting IL-12p70 production. Previous work has identified splenic cDCs as being particularly sensitive to autocrine regulation of IL-12 production by IL-10 in vitro [42] and although the molecular mechanisms for such a process are not fully described, Stat3 has been shown to mediate some of the inhibitory effects of IL-10 on cDC activation in vivo [59]. In contrast, we found no evidence supporting a role for IL-27 in the regulation of IL-12p70 or IL-10 production in cDCs, despite evidence that BMDCs generated from IL-27Rα−/− mice show enhanced production of IL-12p40 and p70 in response to TLR ligation [60] and that autocrine IL-27 is required for optimal macrophage IL-10 production [61].
The CD11cint/lo population capable of transferring disease progression includes NK cells and CD11cloCD45RB+ ‘regulatory’ DCs, two populations that we have previously shown to produce IL-10 and contribute to immunopathology using other assay systems [21], [43]. These data are also consistent with evidence that IL-10 production by innate cells, rather than T cells, is the dominant negative regulator of effector responses after vaccination against L. major [62], and that therapeutic infusion of LPS-activated BMDCs reduces pathology and parasite load, irrespective of whether the splenic IL-10-producing CD4+ T cell frequency is reduced or maintained [23]. IL-10 is a well known suppressor of IFNγ-induced NO production [63]. Hence, the loss of IL-10 expressing cDCs and CD11cint cells in an IFNγ-replete environment may underlie the increased NO production observed after therapeutic depletion of CD11c+ cells.
Unlike previous data showing a requirement for myeloid DCs in the generation of effector responses to acute L. donovani infection [48], the ablation of CD11c+ cells during chronic infection did not significantly affect IFNγ production by CD4+ T cells, suggesting that neither cDCs nor other CD11c+ cells are essential for the maintenance of effector T cell responses, at least over the 7 day time frame studied here. This result is in keeping with data showing the effects of DC ablation on CD4+ T cell responses to M. tuberculosis, where DCs are critical for initial priming of CD4+ effector T cell responses but dispensable for recall Th1 responses after vaccination [6]. However our data are in contrast to a report showing that CD11c-depletion during established infection with S. mansoni resulted in a significant reduction in IFNγ+ production by CD4+ T cells after ablation of CD11c-expressing cells [49]. Hence, the requirement for CD11c+ cells to maintain T cell IFNγ production would appear to be context and/or infection-specific.
Chronic L. donovani infection is associated with the expansion of CD4+ T cells that co-express both IFNγ and IL-10 (this manuscript and [23], [24]). The further detailed characterization of this population provided here indicates that during L. donovani infection, these IL-10-producing CD4+ IFNγ+ T cells are Foxp3−, T-bet+ and CD127−. Hence, they are likely to be related to the IL-10-producing effector T cells found in experimental Toxoplasma gondii and Listeria monocytogenes infection [35], [64] and Plasmodium infection [31], [65]. Although T-bet and CD127 expression were not directly addressed, Foxp3− ‘effector’ CD4+ T cells also appear to be the major CD4+ T cell subset producing IL-10 during infection of mice with L. major [66], [67]. However the increase in Foxp3+ natural Treg frequency and suppressive activity seen in cutaneous L. major infection [68] appears to be absent during EVL (this manuscript and [23]). IFNγ+IL-10+ cells are also associated with L. donovani and Mycobacterium tuberculosis infection in humans [69], [70].
Multiple lines of evidence have suggested a link between IL-27 and the production of IL-10 by CD4+ T cells in vitro or during autoimmune processes [27], [28], [29], [30], [32], [33]. IL-27 plays a role in the development of IFNγ+IL-10+ CD4+ T cells during experimental L. major [71] and Listeria monocytogenes [35] infection in vivo, as assessed using IL-27Rα−/− mice. Such mice also display enhanced resistance to infection with L. donovani, although T cell IL-10 production was not assessed [72]. Furthermore, systemic IL-27 levels are elevated in humans infected with L. donovani, with splenic myeloid cells providing a major source of IL-27 mRNA that was proposed to enhance IFNγ+IL-10+ T cell responses via the induction of T cell-derived IL-21 [73]. Although there is some in vitro evidence of DC-derived IL-27 inducing T cell IL-10 production [37], [38], none of the studies described above provided a formal and causal link between IL-27-producing cDCs and IFNγ+IL-10+ CD4+ cell polarization in vivo, as we have now demonstrated here. The capacity for cDCs to drive polarization of IFNγ+IL-10+ cells may be as a result of their higher expression of IL-27 than CD11cint cells during infection. However, we believe it is likely that a combination of cytokine profile and the capacity for sustained antigen presentation underlies the essential requirement for CD11chi cDCs in the generation of IFNγ and IL-10 co-producing CD4+ T cells [64], [74], rather than sole production of IL-27. Further study and the development of mice with targeted deficiency of IL-27p28 will be required to delineate the relative contribution of these events in vivo.
Although T cell-derived IFNγ has long been known as a critical mediator of parasite clearance in EVL [75], the role of Th1 cells producing mixed effector/regulatory cytokines still remains to be clearly established. During infections where pro-inflammatory responses would otherwise be rampant, this phenotype appears essential to minimize host-mediated pathology [31], [64]. However, the association of IFNγ+IL-10+ CD4+ T cells with disease progression in leishmaniasis has suggested that through IL-10 production, these cells may contribute to parasite persistence and/or disease pathology. The data provided in this manuscript suggest that at least in EVL this association is not causal, as splenomegaly and loss of host resistance were equally well promoted by CD11cint/lo cells as by CD11chi cells, even though the former failed to promote the expansion/maintenance of IL-10-producing Th1 cells. A similar conclusion was also drawn from reciprocal studies using a model of DC immunotherapy [23]. Our data indirectly suggest, therefore, that IL-10 derived from other cellular sources is sufficient to counterbalance the otherwise potentially fatal consequences of Th1-derived effector cytokines.
On a cautionary note, whilst it is tempting to conclude that there is a causal link between the phenotype of cDCs (and CD11cint cells), the regulation of T cell immunity, parasite containment and the development of pathology, this may not be the case, given the complexity of potential interactions in vivo. For example, we have shown that cDCs from infected mice retain the capacity for antigen presentation in vitro (data not shown) and altered activation of T cells is clearly a consequence of cDC transfer. Studies involving the transfer of MHC-deficient cDCs (and CD11cint cells) would be required, however, to discover whether cDCs and CD11cint cells also have the potential to regulate pathology independently of their capacity to interact in a cognate manner with T cells, e.g. by influencing local stromal or myeloid cell function or directly regulating vascular remodeling [43], [45], [50].
Finally, the recognition that cDCs switch from a host protective role in the induction of immunity [48] to one in which they may hinder parasite elimination may provide exciting new opportunities for targeted immunotherapy. Indeed, the impact of CD11c+ cell depletion was of a similar magnitude to that observed after a variety of chemotherapeutic and immunotherapeutic interventions (reviewed in [76]). To our knowledge, DC depletion with the aim of overcoming immunosuppression has not been attempted in the clinic, though removing excessively stimulatory DCs has been suggested as a therapeutic approach to prevent GVHD after allogeneic hematopoietic stem cell transplantation [77]. Whether DCs would serve as potential targets for short-term antibody-based immunotherapy in human VL remains to be determined and will require further concerted efforts to first characterize human DC subsets and their function during this disease.
All animal care and experimental procedures were carried out after review and approval by the University of York Ethical Review Process, and conducted under the authority of United Kingdom Home Office Project Licence PPL 60/3708 (‘Immunology and Immunopathology and visceral leishmaniasis’). All experiments were designed and conducted to minimise suffering and to comply with the principles of replacement, refinement and reduction.
C57BL/6 and B6J.CD45.1 mice were obtained from the Biological Services Facility (University of York) or supplied by Charles River Laboratories. C57BL/6J-Tg (Itgax-cre-EGFP) 4097Ach/J (CD11c-cre) mice and C57BL/6-Gt(ROSA)26Sortm1(HBEGF)Awai/J (Rosa26iDTR) mice were obtained from The Jackson Laboratory (Bar Harbor, Maine, USA) and Cre and eGFP genotype positive F1 mice were used between 6 and 12 weeks of age. Mice were infected via the lateral tail vein with 3×107 amastigotes of the Ethiopian strain of Leishmania donovani (LV9). Splenomegaly was calculated relative to body weight and parasite burdens were quantified as Leishman-Donovan Units (LDU) [23]. (CD11c-cre×Rosa26iDTR)F1 mice were treated with 4 ng/g Diphtheria toxin from Corynebacterium diptheriae (DTx, Sigma) i.p. and/or treated with mAb 1A8 or control [45] every other day from day 21 of infection, as required. Administration of 1A8 resulted in depletion of ∼90% of splenic neutrophils, as judged by CD11b, Ly6C and Gr-1 staining (data not shown).
Spleen cells were restimulated for either 90–120 min with 10 ng/ml PMA and 1 µg/ml Ionomycin (Sigma-Aldrich, UK) or for 3 h with BMDCs pulsed with fixed L. donovani amastigotes, and then further incubated with 1 µg/ml Brefeldin A for 4 h. After restimulation, cells were labeled for 30 minutes on ice with mAbs: CD3ε-PE-Cy7 (145-2C11), CD4-FITC (RM4-5), CD127-PE (A7R34) (eBioscience) or CD4-PerCP (RM4-5; BD Pharmingen). Cells were washed and incubated for 30 min on ice in PBS containing Fixable Viability Dye eFluor780 (eBioscience). After washing, cells were fixed (15 min on ice) in 2% paraformaldehyde (PFA). Cells were then permeabilised using 1% Saponin (Sigma PERM buffer). Cells were subsequently labeled (45–60 min on ice) in PERM buffer containing mAbs: IFNγ-PacificBlue (XMG1.2), IFNγ-eFluor450 (XMG1.2), IL-10-APC (JES5-16E3), IL-10-PE (JES5-16E3) T-bet-AlexaFluor647 (ebio4BIO) and Foxp3-FITC (FJK-16a), (eBioscience). All cells were analyzed on a CyAN-ADP flow cytometer using Summit Software (Beckman Coulter, USA).
BMDCs were generated from femurs of C57BL/6 mice using standard methods. On day 7 of culture, cells were pulsed for 24 hrs with paraformaldehyde-fixed Leishmania donovani amastigotes at a ratio of 100 amastigotes to 1 BMDC. Antigen-pulsed BMDCs were subsequently used to restimulate T cells for 3 hours, prior to addition of Brefeldin A for 4 hours and subsequent assessment of CD4+ T cell cytokine production by intracellular flow cytometric analysis, as previously described.
Spleens were dissociated mechanically and digested in 0.2 mg/ml collagenase type IV/DNAse1 mix (Worthington Biochemical, NJ, USA) for 30 minutes at room temperature. Staining was performed as above using mAbs: CD11c-PE-Cy7 (N418), Major Histocompatibility complex class II (MHCII)-APC (M5/114.15.2), MHCII-eFluor450 (M5/114.15.2), CD8α-FITC (53-6.7), CD4-APC (RM4-5), CD40-PE (1C10), CD80-PE (16-10A1), CD86-PE (GL1) and B7-H1-PE (MIH5). For purification, CD11c+ cells were enriched by magnetic separation [18] and CD11chiMHCIIhi cDCs or individual cDC subsets were sorted to ∼98–99% purity on a BeckmanCoulter MoFlo cell sorter.
After sorting, cells were washed, counted and plated in triplicate in complete RPMI at 1×106 cells/ml. Where indicated, LPS (1 µg/ml; Sigma), anti-mouse IL-27p28 or Goat IgG (10 µg/ml; R&D Systems) or anti-mouse IL-10R (1.3 µg/ml; gift of Dr. M. Kullberg) were added. After 24 h, supernatants were harvested and stored at −80°C until assessed by ELISA for levels of IL-12p40, IL-10 (Mabtech, Sweden), IL-27p28 and IL-12p70 (R&D Systems).
Splenocytes from naïve and infected PBS or DTx-treated mice were incubated in RPMI at 5×106 cells/ml for 60 min at 37°C, and non-adherent cells removed by vigorous washing. Adherent cells were cultured for 24 h at 37°C and supernatants were assayed using Greiss Reagent (Promega, Madison, WI, USA).
12 h after the first administration of DTx, infected (CD11c-cre×Rosa26iDTR)F1 mice received 1.5×105 CD11chiMHCIIhi cells or 6.0×105 CD11cintMHCII+ cells isolated from day 21-infected B6J.CD45.1 mice. DTx administration was continued at 48 h intervals to maintain depletion of endogenous CD11c+ cells.
Statistical analysis was performed using a students t test or one-way ANOVA as appropriate, with p<0.05 considered significant. All experiments were conducted independently at least twice.
|
10.1371/journal.pntd.0003810 | A Multi-country Study of the Household Willingness-to-Pay for Dengue Vaccines: Household Surveys in Vietnam, Thailand, and Colombia | The rise in dengue fever cases and the absence of dengue vaccines will likely cause governments to consider various types of effective means for controlling the disease. Given strong public interests in potential dengue vaccines, it is essential to understand the private economic benefits of dengue vaccines for accelerated introduction of vaccines into the public sector program and private markets of high-risk countries.
A contingent valuation study for a hypothetical dengue vaccine was administered to 400 households in a multi-country setting: Vietnam, Thailand, and Colombia. All respondents received a description of the hypothetical dengue vaccine scenarios of 70% or 95% effectiveness for 10 or 30 years with a three dose series. Five price points were determined after pilot tests in order to reflect different local situations such as household income levels and general perceptions towards dengue fever. We adopted either Poisson or negative binomial regression models to calculate average willingness-to-pay (WTP), as well as median WTP. We found that there is a significant demand for dengue vaccines. The parametric median WTP is $26.4 ($8.8 per dose) in Vietnam, $70.3 ($23.4 per dose) in Thailand, and $23 ($7.7 per dose) in Colombia. Our study also suggests that respondents place more value on vaccinating young children than school age children and adults.
Knowing that dengue vaccines are not yet available, our study provides critical information to both public and private sectors. The study results can be used to ensure broad coverage with an affordable price and incorporated into cost benefit analyses, which can inform prioritization of alternative health interventions at the national level.
| Dengue is complicated. There are four serotypes of the dengue virus, and dengue infection occurs in almost all age groups. Infection with one serotype provides life-long immunity to that specific serotype but does not protect against the other three serotypes. Unlike other diseases which already have preventable vaccines developed, currently there are no commercially available vaccines for dengue fever. Even if the first vaccine becomes available, it is expected that there will be a limited number of vaccine doses available in the first few years. Due to the increase in dengue fever cases, there is already huge public and private interest in potential dengue vaccines. This study reports the household willingness-to-pay for a hypothetical dengue vaccine in three dengue endemic countries. We found that household demand is strongly related to price and income. It was also observed that more than half of the study populations are willing to pay for vaccines when price is lower than the median estimates reported here. This study may contribute to a more effective decision on dengue vaccine introduction.
| Dengue fever is a major public health concern in South-East Asia and South America. Dengue virus is transmitted to humans by Aedes mosquitoes. Clinical presentation ranges from self-limited, mild febrile illness to classic dengue fever (DF) to the more severe form of illness, dengue hemorrhagic fever (DHF). The global burden of dengue has increased dramatically in the past five years, and presently, DF and DHF are recognized as a major cause of mortality and morbidity in tropical and sub-tropical countries[1,2]. The recent study shows that there are 96 million apparent and 294 million inapparent dengue infections occurring yearly, and the total 390 million infections are more than three times the previous estimate of the World Health Organization (WHO)[3–6]. At present, there is no specific treatment for dengue infection. Mosquito control prevention efforts have not been sufficient to control the disease. Vaccine development is still in progress.
The Dengue Vaccine Initiative (DVI) has conducted extensive multidisciplinary dengue fever studies for decision makers in three countries: Vietnam, Thailand, and Colombia. In Vietnam, annual disease incidence is reported to be 145/100,000 population according to the national surveillance system in 2010[7]. Because extensive studies in Vietnam are lacking, a better understanding of dengue and its impact is necessary for disease control, especially in making decisions in regard to future implementation of dengue vaccines. In Thailand, DF/DHF has steadily increased in both incidence and range of distribution despite mosquito control efforts. The dengue incidence rate in Thailand was estimated to be 177/100,000 population in 2010[8]. Dengue epidemiology in Thailand can be characterized by the circulation of all four dengue serotypes and the presence of a well-established national dengue surveillance system. Colombia has seen a significant increase in cases of DF/DHF during the last 10 years, with epidemic waves occurring every 3–4 years. In 2010, Colombia experienced the largest recorded epidemic with the dengue incidence rate of 685/100,000 population[9]. All four dengue serotypes are found in Colombia, and in significant contrast to Asian countries, the disease occurs in people of all ages.
The rising tide of the Dengue Fever and its associated morbidities underscore the need for vaccines against dengue[10–12], but there continues to be a lack of economic assessment on dengue fever vaccines. Vaccines currently under development may significantly reduce the burden of disease. The three countries mentioned above are each candidates to become early adopters of future dengue vaccines. However, like many other low- and middle-income countries, these countries will face decisions on whether and how to incorporate new and potentially expensive vaccines within their budget-constrained national vaccination programs. Therefore, understanding the private economic benefits of potential dengue vaccines is necessary for accelerated introduction of vaccines into the public sector program and private markets of high-risk countries.
To estimate household demand and WTP for hypothetical vaccines against dengue infection, we administered a study questionnaire to 400 households in each of three countries (Fig 1). All respondents (N = 400) at each site received a description of the hypothetical dengue vaccine scenarios of 70% or 95% efficacy for 10 or 30 years. Five pre-assigned prices were determined after performing pilot tests (40 pretest interviews) and open-ended focus group discussions. In this analysis, the dichotomous choice method was adopted. Unlike other CV studies such as open-ended, bidding game, and payment card which have shown incentive compatibility problems, dichotomous choice eases the burden on the respondents, decreasing the number of protest answers[13]. Respondents were asked if they would be willing to buy a vaccine for their youngest child and other household members at randomly pre-assigned prices. Interviewers reminded respondents of their budget constraints and mentioned that there were no right or wrong answers.
We adopted a time-to-think approach in Colombia and Vietnam, which gives respondents time to deliberate their decision on whether they would like to purchase a new vaccine. In previous split sample studies[14–17], respondents tended to demand significantly fewer vaccines when provided with more time to think about their purchasing decision compared to respondents that completed interviews in one sitting. In Colombia and Vietnam, respondents received general information on dengue illness, risk factors and a hypothetical vaccine. They were instructed to consider their vaccine purchasing decision overnight. The time-to-think design was not implemented in Thailand because of cost and logistical issues. Instead, respondents completed the entire survey in one interview.
The three sites were selected in support of multidisciplinary research goals of the Dengue Vaccine Initiative. In addition to economic studies, these sites were chosen to provide an in-depth picture of dengue epidemiology and transmission within high risk populations of the chosen countries. All three sites share several common characteristics—high levels of dengue virus transmission; stable population with low rates of migration and high rates of ethnic homogeneity; sites are easily accessible with good health services; and local dengue control officers and provincial public health officials are exceptionally motivated and committed to dengue research. Interviews were administered to the head /senior of selected households.
The questionnaire consists of six sections. The first section collected demographic information about the respondent and members of the household. The second section asked about respondent’s perception and experience regarding dengue fever, including activities undertaken by the household to reduce their risk of dengue infection. The third section included information for the respondents on general conditions of dengue fever including how the disease is transmitted and dengue fever risk may be mitigated through community-wide efforts. This section also recorded previous vaccination history, and provided a description of the hypothetical dengue vaccine, including efficacy and duration of protection which can be found in S2 Text and S2 Fig. A series of questions were asked to ensure that respondents had understood how the vaccine works. In the fourth section, household demand was collected. For example, the first WTP question was framed as: “Suppose that the total cost for the dengue fever vaccine would be VND 450,000 for three dose needed for one person. Would you buy this vaccine for your youngest child?” To access household WTP, the additional question was followed: “Suppose that this dengue fever vaccine costs VND 450,000 for the 3-dose series required for each person (same price for adults and children), how many people in your household (not including your youngest child) would you be willing to purchase vaccines for?…Who would you buy this vaccine for?” The responses were recorded in a table which is linked to the demographic information in the first section. Respondents who said that they would not buy the vaccine at the specified price were asked if they would take the vaccine if it were provided free. For those who did not want to take a free vaccine or pay any positive price, an additional question was posed to see why they would not take the vaccine. The respondents refused to take the vaccine because they did not think that vaccines are safe or prevent the disease were determined as out-of-market respondents and did not proceed to the next step. In the fifth section, socioeconomic information was collected, such as education, occupation, income, and economic status. The sixth section included questions regarding the time-to-think approach. Pilot studies, focus group discussion and pre-final questionnaires, were conducted in each of the three studies to refine the survey instrument and to help determine an appropriate set of prices for each setting. Results from the pilot studies were not included in this analysis.
Our underlying economic model assumes that respondents maximize their household utility, subject to their budget constraints. Vaccines are one of many purchases that can be used to build health capital and household health is one of many competing spending choices. Household vaccine demand is a non-negative integer value and the number of vaccines demanded (dependent variable) can be estimated as a function of vaccine price, efficacy, household perceptions of dengue severity and likelihood, as well as household socio-economic characteristics. Count models are suitable for our household demand analysis because the count model estimates non-negative integer values and specifies the quantity demanded with a mean that is dependent on exogenous variables[13,21]. The Poisson or its variants (e.g., negative binomial) typically takes the exponential form for expected demand, and the Poisson probability density function can be written as
Pr(xi=n)=e−λiλinn!,n=0,1,2…
where n is observed demand, and λi is the mean, λi = exp(ziβ). For the Poisson model, the mean is equal to the variance of the distribution. If the variance is greater than the mean, the model is mis-specified due to overdispersion[22,23] (S1 Text). Overdispersion may not affect the coefficient estimates significantly, but causes standard errors to be underestimated. For this reason, the Z-score test and the boundary likelihood ratio test were performed to test for overdispersion for each country[22]. The negative binomial technique relaxes the assumption of equality of the mean and variance by adding a gamma distributed error term[24]. A common version of the negative binomial model is as follows:
E(xi|ziβ)=λi=exp(ziβ)log(E(xi))=ziβ+θi
where θi represents unobserved individual differences (or unobserved heterogeneity).
Pr(xi)=Γ(xi+1α)Γ(xi+1α)Γ(1α)(1α1α+λi)1α(λi1α+λi)xi
where λi = exp(ziβ). The mean of the negative binomial distribution is E(xi) = λi = exp(ziβ). However, now the variance of the dependent variable is V(xi) = λi(1 + αλi). The parameter α can be interpreted as the overdispersion parameter. When α is equal to zero, the variation becomes equal to the mean, and the distribution can be modelled by Poisson regression. However, if α is greater than zero, overdispersion exists, and the Poisson model is rejected in favor of the negative binomial model[13] (S1 Fig). Standard errors were corrected for the cluster sampling procedure to improve the accuracy of the estimates.
Model validation is critical to check whether a model is appropriate and useful. There are several statistics which estimate how well the model fit the data, how much error was in the model[25]. Mean Absolute Deviation (MAD) and Mean Squared Prediction Error (MSPE) were used to estimate how well the model fit the data [24–26]. MAD provides a measure of the average mis-prediction of the model, and MSPE is typically used to assess the error associated with a validation or external dataset[26].
MAD=∑i=1n|Y^i−Yi|nMSPE=∑i=1n(Y^i−Yi)2n
where n is validation data sample size, Y^iis the predicted value, and Yi is the observed value.
50% of the full dataset for each country were randomly selected as a validation dataset. The two statistics were used to measure how well the original models estimated on estimation data predict the validation data. The smaller the value of MAD and MSPE represents the more desirable model which fits the data as closely as possible[24].
The mean household WTP can be calculated by aggregating the area beneath the demand curve.
WTP(vaccine)=∫0∞eβixi+βpPidP=−eβixiβp
where βp is the estimated coefficient for price and βi is an array of the estimated coefficients for the other independent variables. The median WTP was also calculated by estimating the price at which an estimated 50% of the population would purchase vaccines. Parametric estimates of WTP are sensitive to the choice of distribution and functional forms of household demand. Non-parametric models, Turnbull lower bound and Kristrom’s midpoint models[13,27–29], were also estimated. The advantages of non-parametric models are in their simplicity and transparency. The Turnbull estimator does not impose any statistical assumptions about how WTP is distributed[13], and is considered to be a conservative measure. The Kristrom’s midpoint estimator assumes that the distribution between bid points is symmetrical[15]. Both models provide a useful comparison of mean and median WTP with the parametric WTP measures.
The contingent valuation studies and survey questionnaires were approved by the ethical review committees in three countries (National Institute of Hygiene and Epidemiology in Vietnam, Faculty of Tropical Medicine, Mahidol University in Thailand, Universidad de Antioquia in Colombia), as well as Ministry of Health in three countries and Institutional Review Board of the International Vaccine Institute. Written informed consent was obtained prior to conducting interviews and respondents were informed that they could terminate interviews at any time. Respondents received compensation for their time.
Some respondents rejected the hypothetical scenario presented, and these respondents were dropped from our analysis. We identified such respondents based on their reasons for declining the vaccine, specifically those who reported that they believed that vaccines are not safe or that vaccines would cause some side effects. Only 1.5%, 2%, and 0.25% out of the 400 respondents rejected our hypothetical vaccine scenarios in Vietnam, Thailand, and Colombia, respectively.
To develop conservative WTP estimates, we also attempted to identify potential yea-sayers, i.e. people who reported unrealistically high WTP for vaccines. Potential yea-sayers were defined as those for whom the product of reported demand and offered vaccine price was greater than twice their reported monthly household income across all price points. All of the yea-sayers were identified at the highest price point. Because the highest price point in each country was designed to see the choke effect, it seemed unrealistic when respondents were willing to purchase vaccines for all members by paying a sum of more than twice their entire monthly household income. These responses (2%, 3.3%, and 1.8% in Vietnam, Thailand, and Colombia, respectively) were considered as outliers and dropped from our analysis.
Table 1 shows household characteristics for each study site. Average respondent age is from 37 to 47 years, and average household size is around 5 members. Most of the respondents are females. In order to make sure that their responses reflect the decisions made at the household, the respondents were asked who would be primarily involved in making decision for their household members. Over 80% of the respondents confirmed that they would be primarily involved in making vaccine purchasing decisions. The respondents had 6~9 years of median school education in Vietnam and Colombia, and 1~5 years of median school education in Thailand. The self-reported mean household income per month is $351, $788, and $367 in Vietnam, Thailand, and Colombia, respectively. For all three countries, more respondents thought that dengue fever is serious for children than respondents who thought so for adults, although the differences are not significant. Approximately 35%, 60%, and 87% of the respondents in Vietnam, Thailand, and Colombia said that their children would likely contract dengue in the next five years. In Vietnam and Thailand, about 28% of the respondents reported that at least one member of their household had contracted dengue fever in the past, compared to 10% in Colombia. Around 52%, 47%, and 27% of the respondents in Vietnam, Thailand, and Colombia mentioned that they know someone who had dengue fever in their neighborhoods. In response to past vaccine purchase history, 68% and 13% of the respondents in Vietnam and Colombia answered that they had previously purchased other vaccines. This question was not asked in Thailand. Over 99% of the total respondents from all three countries correctly answered questions designed to test their understanding of vaccine duration and efficacy.
Table 2 summarizes the raw data for average household vaccine demand as a function of price and efficacy. Vaccine demand decreases with price in all three countries. We did not find any significant difference in demand between the 70% and 95% efficacy scenarios.
The regression results are shown in Table 3. Type 2 includes all possible covariates, while type 1 is a parsimonious model which includes only non-attitudinal variables. As expected, the price variable is statistically significant at the 1% level and has a negative sign. Income per capita (in log form) is also highly significant across countries with positive signs indicating that vaccine demand increases with income. The respondent age variable is inversely related to the likelihood that a respondent would purchase the vaccine in Vietnam, but this variable is not significant in Thailand and Colombia. The relationship between education and demand was not consistent across countries. Compared to no education, respondents with some education had significantly greater demand in Vietnam, but lower demand in Colombia. The coefficient was positive, but not significant in Thailand. The interaction between earning and price is positive and significant, meaning that respondents with a higher income can afford a more expensive vaccine. Neither the perceived seriousness of dengue nor the likelihood of contracting the disease in the next 5 years was a statistically significant determinant except the perceived seriousness in Colombia. Respondents in Vietnam who knew a person who had contracted dengue were more likely to be willing to purchase a vaccine. There is some evidence that respondents who had purchased other vaccines tend to demand more for our hypothetical dengue vaccine than those without previous vaccine purchase experience. The overall robustness of the model was examined by the MAD and MSPE statistics. In the field of transportation and accident analyses where the negative binomial models are more commonly used, 1.8 of MAD and 7.2 of MSPE were considered to be relatively small values for a mean dependent variable of 2.85[24,30]. Given that the mean value of the dependent variable in this study ranges from 1.54 to 2.66, the models for all three countries produced fairly satisfactory predictive performance. In particular, MSPE values are closer to 1 for both short and long models in Vietnam, meaning that the model fits the data better than the other countries. While the long model is preferred over the short model in Vietnam, the short models fit the data better for Thailand and Colombia. Fig 2 depicts the observed and predicted fractions of household members vaccinated by price.
Table 4 shows parametric and non-parametric mean WTP estimates for the three-dose series described in the hypothetical scenario. The mean WTP per dose is included in parentheses. We did not generate separate estimates by vaccine efficacy/duration since the difference in vaccine demand was not statistically significant. The conservative Turnbull-lower bound mean WTP is $42.3 ($14.1 per dose) for a 3 dose-series vaccine in Vietnam, $68.8 ($22.9 per dose) in Thailand, and $48 ($16 per dose) in Colombia. The parametric mean WTP estimates lie in between Turnbull lower bound and Kristrom midpoint values except Colombia. The mean WTP in Thailand is higher than for the other two countries.
Table 5 summarizes the median WTP estimates for both parametric and non-parametric models. The median WTP is calculated based on the price in which an estimated 50% of the population would purchase vaccines. Median estimates tend to be less sensitive to the unexpected responses and functional form assumptions than mean estimates, as long as the empirical 50th percentile value lies in between the lowest and the highest price points[13]. For all three countries, the observed median WTP estimates fall between the lowest price and the highest price offered in the surveys. In the case of non-parametric models, the point estimates were linearly interpolated. The parametric median WTP is $26.4 ($8.8 per dose) for a 3 dose-series vaccine in Vietnam, $70 ($23.4 per dose) in Thailand, and $23 ($7.7 per dose) in Colombia. The median WTP in Thailand is again higher than the other countries.
It is also possible to create separate sub-models and estimate vaccine demand for different age groups. We divided households into three groups: young children (age under 5 years), school age children (age 5–18 years), and adults (age over 19 years). Fig 3 shows the predicted coverage as a function of price for the three age groups. For all three countries, the predicted fractions of young children vaccinated are higher than those for the other age groups at any price, suggesting that respondents place more value on vaccinating young children than school age children and adults.
This study provides insight into the private economic benefits of potential dengue vaccine in three countries. The median WTP per household member was $26.1 ($8.7 per dose) in Nha Trang, Vietnam, $69.8 ($23.3 per dose) in Ratchaburi, Thailand, and $22.6 ($7.5 per dose) in Medellin, Colombia. Our models showed that household demand for the dengue vaccine is sensitive to price and income, suggesting that respondents took the hypothetical purchasing scenario seriously. These results suggest the possibility that a private market for dengue vaccines exists in these three countries and that sales may be robust if vaccine prices are lower than the median estimates from our study. Since respondents were not bound by their stated purchasing decisions, it is possible they may not act as they reported.
There was a relatively large fraction of respondents who were willing to purchase vaccines at high price points in Thailand (11%) compared to those in Vietnam (9%) and Colombia (3%). One explanation for higher WTP in Thailand is that the mean reported household income in Thailand is almost two times greater than that in Vietnam and Colombia; therefore, Thai respondents had more purchasing power. In addition, we were not able to employ the time-to-think research design in Thailand due to budget and logistical constraints. Other researchers have found that people tend to report lower WTP when they have more time to think about a new vaccine product and their budget constraints[14–17]. It is worth noting that this study did not attempt to test the validity of the time-to-think approach. Rather, the study was designed based upon evidence from the time-to-think option used in the previous studies [17]. The time-to-think approach allowed respondents to think carefully about their budget constraints and may more accurately reflect their willingness-to-pay for the vaccine. While the absence of the time-to-think option in Thailand may contribute to higher WTP compared to Vietnam and Colombia, the exact magnitude could not be measured in this study. The detailed methodology and comparison for the time-to-think approach were extensively discussed by Cook J et al.[17].
It should be noted that two parameter estimates among the significant determinants differ in sign by country: education 1 and age group 2. While it is common for regression coefficients to have the same direction towards the underlying concept in the similar circumstance, some of the socio-economic variables may behave differently across countries to explain variance of the dependent variable (vaccine demand) due to the diverse contexts of local specific situations. Ideally, our study samples would be more heterogeneous and more representative of the entire countries; however, we were limited to performing the studies in locations where epidemiologic studies were being conducted at the same time. As a result, these results may not be generalizable beyond the communities in which the studies were conducted. In comparison with previous studies, a study from Philippines suggests a median WTP of $60 for a 10 year efficacy scenario of dengue vaccine[31], and a study from Indonesia shows a median WTP of $1.94 for a fully efficacious dengue vaccine[32]. The estimates may differ depending upon income levels of study populations and previous experience in receiving other vaccines in the study communities (i.e. availability of free vaccines from local health centers, etc.).
The rise in dengue fever cases and the absence of dengue vaccines will likely cause governments to consider various types of effective means for controlling the disease. The contingent valuation study proposed here provides important information—how much people are willing to pay for a dengue fever vaccine to avoid the risk of getting infected. The WTP estimates provide quantification of the private benefit of disease reduction. Results can be incorporated into cost benefit analyses, which can inform prioritization of different health interventions at the national level. The study can also assist decision makers to understand how much of population can be covered by subsidizing dengue vaccines when implementing the nationwide campaigns and can help inform how households would allocate vaccines across age groups given household budget constraints. Further, the WTP study provides vaccine manufacturers a better picture of people’s perceptions of dengue fever and dengue vaccines.
|
10.1371/journal.pntd.0007461 | Asymptomatic immune responders to Leishmania among HIV positive patients | Concomitant infection with human immunodeficiency virus (HIV) and the Leishmania parasite is a growing public health problem, the result of the former spreading to areas where the latter is endemic. Leishmania infection is usually asymptomatic in immunocompetent individuals, but the proportion of HIV+ individuals in contact with the parasite who remain asymptomatic is not known. The aim of the present work was to examine the use of cytokine release assays in the detection of asymptomatic immune responders to Leishmania among HIV+ patients with no previous leishmaniasis or current symptomatology. Eighty two HIV+ patients (all from Fuenlabrada, Madrid, Spain, where a leishmaniasis outbreak occurred in 2009) were examined for Leishmania infantum infection using molecular and humoral response-based methods. None returned a positive molecular or serological result for the parasite. Thirteen subjects showed a positive lymphoproliferative response to soluble Leishmania antigen (SLA), although the mean CD4+ T lymphocyte counts of these patients was below the normal range. Stimulation of peripheral blood mononuclear cells (PBMC) or whole blood with SLA (the lymphoproliferative assay and whole blood assay respectively), led to the production of specific cytokines and chemokines. Thus, despite being immunocompromised, HIV+ patients can maintain a Th1-type cellular response to Leishmania. In addition, cytokine release assays would appear to be useful tools for detecting these individuals via the identification of IFN-γ in the supernatants of SLA-stimulated PBMC, and of IFN-γ, MIG and IL-2 in SLA-stimulated whole blood. These biomarkers appear to be 100% reliable for detecting asymptomatic immune responders to Leishmania among HIV+ patients.
| The proportion of patients with HIV+ who have at some time been infected with the Leishmania parasite, but who remain asymptomatic, is unknown. It is important to be able to identify such patients to determine the prevalence of asymptomatic leishmaniasis in the HIV+ population, and because these persons are at increased risk of developing symptomatic visceral leishmaniasis. In the present work, a population of HIV+ patients showing a cellular immune response to Leishmania infantum was identified. These subjects all showed a clear Th1-type response when their PBMC or blood were stimulated in vitro with soluble Leishmania antigen (SLA). Cytokine release assays and the detection of IFN-γ, MIG and IL-2 (specific biomarkers for immunity to Leishmania) were found to be useful for detecting this population of HIV+ patients. New studies with larger numbers of patients are needed to confirm the present results.
| Leishmaniasis is a neglected, vector-borne disease associated with high morbidity, caused by protozoan pathogens of the genus Leishmania. In visceral leishmaniasis (VL), the most serious form of the disease (fatal if untreated), the parasite is systemically disseminated. VL is hypoendemic in the Mediterranean region, where the causal agent is Leishmania infantum [1].
In 2017, nearly 160,000 people were diagnosed with HIV in the WHO European Region, marking another year of alarming new HIV numbers [2]. HIV is most prevalent in France, Spain, Italy and the United Kingdom; Spain has the highest DALYs value attributable to HIV/AIDS [3].
The number of reported cases of Leishmania/HIV co-infection increased rapidly during the 1990s, a consequence of the spread of the HIV pandemic, increased awareness among reporting institutions, and the growing geographical overlap between the two diseases [4].
Leishmania/HIV co-infection has been recognised as an emerging problem in areas endemic for leishmaniasis [5]. In co-infected patients, the symptoms of VL may be more severe than in immunocompetent individuals, relapse is more common, mortality higher, parasite loads greater, and organs not normally involved in VL may be parasitised [6]. Co-infected patients are also potential spreaders of Leishmania, posing a huge problem for current elimination strategies [7]. Indeed, HIV infection is recognised as an emerging challenge in the control of VL [4].
HIV infection increases the risk of developing VL between 100 and 2320 times [6, 8]. The marked fall in the number of CD4+ T cells caused by the virus, along with the reduced production of IFN-γ, and the lesser leishmanicidal capacity of the macrophages [9], results in the replication and uncontrolled dissemination of the parasite around the body. HIV infection is also associated with the reactivation of latent Leishmania infection and its progression towards VL [1]. Similarly, active Leishmania infection can increase the replication of the virus, encouraging progression to full-blown acquired immunodeficiency syndrome (AIDS) [4]. In the pre-HAART (highly active antiretroviral therapy) era, patients with HIV also infected with Leishmania commonly developed VL; these days, however, the efficiency of HAART allows a variable proportion of such patients to remain asymptomatic [8, 10].
The variation in the ratio of patent VL cases versus asymptomatic cases in different L. donovani- and L. infantum-endemic areas (from 2.4:1 in Sudan to 50:1 in Spain [11]) reflects differences in parasite virulence and host characteristics, but perhaps also in study design and the tests used to identify asymptomatic infection. Cell immunity usually remains positive for several years, sometimes even throughout an individual’s life [12, 13]. However, serological markers can revert to negative within 4 months of any first sample being inspected [14]. Serology is also unsatisfactory for detecting asymptomatic Leishmania infection in endemic areas where mean parasitaemia levels are low or intermittent [15]. No normalised or commercial techniques exists for defining asymptomatic Leishmania infection. An asymptomatic subject is usually regarded as someone from an endemic area who shows an immune response (either antibodies or a positive leishmanin skin test [LST]) against Leishmania, or who has parasites in the blood, but who remains healthy [15, 16].
VL is largely diagnosed using molecular and serological techniques. However, the serological detection of Leishmania in patients with HIV or AIDS is not very sensitive since the parasite elicits a weak antibody response [17, 18]. Molecular diagnoses are normally reliant on PCR, which is highly sensitive and specific for VL in co-infected patients [19] since the parasite load is high, even in peripheral blood. However, PCR has been used in very few studies to detect patients with HIV who are also asymptomatic but infected with Leishmania [20, 21].
The leishmanin skin test has been widely used in the field to study the prevalence of infection, but its side effects, and the suspect quality of its manufacture, have seen it banned in some countries, including most European nations (certainly in Spain) [17]. Cytokine release assays using whole blood or PBMC stimulated with soluble Leishmania antigen (SLA) are useful for monitoring patients who have undergone solid organ transplantation following treatment for VL, and for detecting asymptomatic Leishmania infection in this population [22]. They have also been found useful for establishing the efficacy of treatment for VL in patients also infected with HIV, and for assessing the need to maintain secondary prophylaxis in such patients [23]. Recent studies have confirmed that IFN-γ and IL-2 are good biomarkers of asymptomatic Leishmania infection [24], but so too are the induction protein of IFN-γ (IP-10 or CXCL10), the monokine induced by IFN-γ (MIG or CXCL9), and monocyte chemotactic protein 1 (MCP-1 or CCL2) [25, 26].
Antiretroviral treatment has been very successful in controlling HIV replication and preventing the appearance of opportunistic infections in HIV+ patients [10, 27]. HAART not only reduces viral replication but leads to an increase in the number and functionality of CD4+ T cells and reverts the majority of immunological abnormalities. In the pre-HAART era, patients co-infected with HIV and Leishmania commonly failed to produce immunity against the parasite following treatment for VL [28], but these days most HAART-treated patients do so [23]. This begs the question of whether, in Leishmania-endemic areas, there are asymptomatic immune responders among HIV+ individuals (as is seen in solid organ transplantation-associated immunodepressed patients [22]). Cell immunity techniques might be used to detect such individuals. The aims of the present work were 1) to determine whether these techniques, along with serological and molecular tests, can be used to make such identifications, and 2) to characterise the immune response against the parasite in such individuals.
This study was approved by the Hospital de Fuenlabrada (APR12–65 and APR14-64). All participants gave their written informed consent to be included.
Blood was collected from 82 HIV+ adult patients at the Hospital de Fuenlabrada between 2015 and 2017. All lived in Fuenlabrada (Madrid, Spain), a Leishmania infantum-endemic area with a high prevalence of infection. All subjects were undergoing antiviral treatment and had their viral load regularly monitored. None of the subjects had shown any sign of leishmaniasis. All blood samples were subjected to several specific Leishmania tests (humoral, cellular and molecular). Of the subjects providing samples with negative results to all of them, 19 were randomly selected as negative controls (NC) for analysis. S1 Table provides a detailed description of the present HIV+ patients.
To determine the number of circulating lymphocytes, combinations of CD4/CD8/CD3, CD3/CD19/CD45 and CD3/CD16+CD56/CD45 antibodies, conjugated with FITC/ PE/PerCP (BD Tritest, USA) respectively, were added to 50 μl aliquots of peripheral blood, and analysed by flow cytometry using FlowJo v.7.6.5 software. The percentages obtained were multiplied by the total number of lymphocytes in the haemogram to obtain absolute values for circulating lymphocytes. Values for healthy individuals were used as a reference.
L. infantum antigen was prepared from promastigote cultures in the stationary phase (JPC strain, MCAN/ES/98/LLM-722), as previously described [29]. The parasites were first washed with 1X phosphate-buffered saline (PBS) and centrifuged at 1000 g for 20 min at 4°C. The supernatant was discarded and the pellet resuspended in lysis buffer (50 mM Tris/5 mM EDTA/HCl, pH 7). These samples were subjected to three cycles of freezing/thawing, and then sonicated three times (40 W for 20 s) before being centrifuged again at 27,000 g for 20 min at 4°C. The supernatants were collected, divided into aliquots, and stored at -80°C until use. The protein content was quantified following the Bradford method, using the Pierce BCA Protein Assay Kit (Bio-Rad, USA).
Blood samples (10 ml) were collected in heparinised vials from all subjects. Peripheral blood mononuclear cells (PBMC) were separated out using a Ficoll-Hypaque gradient (Rafer, Spain), resuspended in complete RPMI supplemented with 10% foetal bovine serum, and cultured (in triplicate) at an initial concentration of 2x106 cells/ml in 96-well plates with either complete RPMI (negative control), SLA (10 μg/ml) or phytohaemagglutinin-M (PHA-M) (5 μg/ml) [22]. All cultures were kept for 6 days at 37°C in a 5% CO2 atmosphere. The lymphoproliferative response of each subject was then determined by bromodeoxyuridine incorporation using the Cell Proliferation Kit (GE Healthcare Life Sciences, UK), following the manufacturer's instructions. Results were expressed in the form of a stimulation index (absorbance of stimulated cells/unstimulated cells). The culture supernatants were collected and stored at -20°C for later cytokine and chemokine analysis.
Aliquots (500 μl) of whole blood were incubated in tubes with 10 μg/ml SLA or 5 μg/ml PHA-M. A further tube with no SLA was used as a negative control. All tubes were incubated at 37°C for 24 h, as previously described [22, 24]. They were then centrifuged at 2000 g for 10 min. The supernatants were removed and kept at -20°C for later cytokine and chemokine analysis.
IFN-γ, TNF-α, granzyme B, IP-10, MIG, IL-2 and IL-10 were determined in 50 μl of supernatant from the PBMC cultures, and in the same volume of SLA-stimulated plasma from the WBA [25], using the CBA Human Soluble Protein Flex Set Capture Bead Kit (Becton Dickinson, USA), following the manufacturer's instructions. Results were captured by flow cytometry using Flow Cytometric Analysis Program Array software (Becton Dickinson, USA). Results from each cytokine and chemokine were expressed as the difference between the SLA-stimulated and control plasma concentrations.
An enzyme-linked immunosorbent assay (ELISA) was used to detect antibodies to SLA [22]. Briefly, 96-well plates (NuncMaxisorp Immuno Plates, USA) were coated with 100 μl/well of 10 μg/ml SLA and left overnight at 4°C. The plates were then washed three times with PBS, 0.1% Tween 20 (PBS-T), pH7.4, and blocked with 200 μl/well of PBS containing 0.1% Tween 20 and 3% BSA for 1 h at 37°C. After washing with PBS-T, diluted blood plasma (1/200 in PBS-T) was added (100 μl/well) and incubated for 2 h at 37°C. The plates were then washed with PBS-T and 100 μl/well of 1/5000-diluted HRP-conjugated anti-human Ig (Invitrogen, USA) were added for 30 min at 37°C. All plates were then developed with 100 μl/well of Sigma Fast o-phenylene diamine dihydrochloride (OPD) tablets (Sigma, USA) for 20 min. The reaction was stopped with 50 μl/well of 2NHCl, and absorbance measured at 492 nm.
Immunofluorescent antibody titre (IFAT) analyses of plasma samples were performed using 2 × 105 L. infantum promastigotes in PBS per well (MCAN/ES/98/LLM-722), as previously described [22]. Subject plasma was assayed as two-fold serial dilutions (from 1/20 to 1/640) in PBS to determine total IgG levels using fluorescein isothiocyanate-conjugated goat anti-human IgG (Fluoline G) (BioMérieux, France) diluted 1/200. The threshold titre for positivity was set at 1:80.
The rK39-ICT test (Leti Laboratories, Spain) is a rapid, commercial, immunochromatographic test for the quantitative detection of Leishmania antibodies in serum. Serum (25 μl) was added to the test strips, along with the provided buffer solution, in 2 ml Eppendorf tubes. After 10 min at room temperature, the strips were examined for the two bands (control and specific) indicating a positive result.
DNA was extracted from 200 μl whole blood to which had been added 400 μl of NET10 (10 mM NaCl, 10 mM EDTA, 10 mM Tris HCl), 40 μl of SDS sample buffer (10%), and 2 μl of proteinase K. Samples were incubated with agitation overnight at 56°C. The DNA was isolated using the phenol-chloroform method, precipitating in ethanol [30]. The total DNA was resuspended in 100 μl of sterile distilled water and quantified using a UV-V ND-100 spectrophotometer (NanoDrop Technology, USA).
The extracted DNA was subjected to nested PCR (Ln-PCR) using primer pairs that amplify the Leishmania small ribosomal subunit (SSUrRNA) [31], employing a GenAmp PCR System 2700 thermocycler (Applied Biosystems, USA). The first round of reactions (30 cycles, annealing temperature 60°C) involved the use of primers R221 and R332. The amplicons were diluted 1/40 in distilled water, and 10 μl of this dilution used in the second round of reactions, which involved the use of primers R223 and R333 (30 cycles, annealing temperature 65°C). The amplicons were then visualised in 1.5% agarose gels in TAE buffer (Tris-acetate 0.04 mM, EDTA 1 μM, pH 8) using 0.02% GelRed staining (Biotium, USA) under a MiniBis-pro illuminator (DNR, Bio-imaging systems, Israel). Positive results require amplicons of 358 bp be detected.
Normality was examined using the Shapiro-Wilk test. The Mann-Whitney U test was used to analyse differences between unpaired groups. Significance was set at P<0.05. The cut-offs for the ELISA and cytokine/chemokine release assays were determined by calculating the area under the receiver operating characteristic curve (AUC) and the 95% confidence intervals (CI). Spearman correlation coefficients were calculated between CD4+ T cells and SI, IFN-γ, IL-2 or MIG, and between SI and WBA-associated IFN-γ or IL-2. All calculations were undertaken using GraphPad Prism v.7 software (GraphPad Software, USA).
Among the present HIV+ subjects, none of whom showed any clinical manifestation of Leishmania infection, there was a group of 13 (15.85%) asymptomatic immune responders (ARI subjects) with a stimulation index (SI) of ≥2.39 in the SLA-CPA test (Fig 1A). The median SI of these 13 subjects was 5.11 compared to 1.02 for the non-responders (NC) (p<0.0001). The AUC for lymphoproliferation was 1.00 (95% CI: 1.00–1.00; p<0.0001); the sensitivity and specificity of the SLA-CPA test was therefore 100% (Fig 1B).
Following stimulation of the PBMC of all subjects with PHA-M, no significant difference was seen between the ARI and NC groups in terms of response; both were well capable of responding to PHA-M (S2A Table).
No Leishmania DNA was detected in any blood sample from any HIV+ patient using molecular techniques. In addition, the IFAT and ELISA serological tests were unable to detect any asymptomatic subject infected with L. infantum (results below the cut-offs). The serum rK39-ICT test was negative for all subjects.
No significant differences were seen between the ARI and NC subjects in terms of the size of their different lymphocyte populations (Table 1). Indeed, all populations were of normal size, except for that of the CD4+ T cells, which was below the normal lower limit (527/mm3).
Four of the ARI subjects had a CD4+ T cell count of <200/mm3. However, all the ARI subjects were able to mount a cellular response to L. infantum (Fig 2, and S1 Fig).
Following the SLA stimulation of the subjects' PBMC, the culture supernatants of the ARI subjects showed significantly greater median cytokine/chemokine concentrations than did those of the NC subjects: IFN-γ 946.2 vs. 0 pg/ml (p<0.0001), TNF-α 226.6 vs. 0 pg/ml(p<0.0001), granzyme B 668.7 vs. 0 pg/ml (p<0.0001) (Fig 3A–3C), IP-10 1139 vs. 0 pg/ml (p<0.0001) and MIG 13,062 vs. 9.13 pg/ml (p<0.0001) (Fig 3D and 3E). No IL-2 or IL-10 was detected in any supernatant for any subject.
Table 2 shows the high sensitivity and specificity of the increased cytokines and chemokines as markers of L. infantum cellular immune response in HIV+ persons. However, IFN-γ showed an AUC of 1.00 (95% CI; 1.00–1.00; p<0.0001): it therefore detected 100% of ARI subjects.
The SLA-stimulated plasma of the ARI subjects showed significantly greater median concentrations of certain cytokines and chemokines than did that of the NC subjects: IFN-γ 101.10 vs. 0 pg/ml (p<0.0001) (Fig 4A), granzyme B 53.64 vs. 1.68 pg/ml (p = 0.0396) (Fig 4C), IP-10 2785 vs. 2.27 pg/ml (p<0.0001) (Fig 4D), MIG 583.10 vs. 5.44 pg/ml (p<0.0001) (Fig 4E), and IL-2 291.90 vs. 0 pg/ml (p<0.0001) (Fig 4F). No significant differences were detected in the median production of TNF-α (22.61 vs. 6.93 pg/ml; p = 0.2610) (Fig 4B) or IL-10 (1.23 vs. 0.36 pg/ml; p = 0.7231).
Table 3 shows the sensitivity and specificity of the studied cytokines and chemokines as markers of L. infantum infection. The AUC of IFN-γ, MIG and IL-2 was 1.00 (95% CI 1.00–1.00; p<0.0001); these biomarkers therefore detect 100% of ARI subjects. The AUC for IP-10 was 0.985 (95% CI 61.52–99.79; p<0.0001), while those of TNF-α and granzyme B were more modest at 0.619 and 0.708 respectively.
The results show that all were capable of responding to PHA-M; no significant differences were seen between the ARI and NC groups (S2B Table).
A positive correlation was found between SI and IFN-γ, IL-2, IP-10 and MIG (as stimulated in the WBA) (Fig 5 and S3 Table).
Currently, no screening is undertaken to detect asymptomatic responders to Leishmania among persons who are HIV+, even though their risk of developing VL is relatively high. These carriers represent a risk to the success of Leishmania control strategies [32]. Field tools are therefore needed that can determine the real proportion of HIV+ patients that have been exposed to Leishmania. This is the first work using cytokine release assays to identify a sub-population of individuals who were exposed to Leishmania but in whom no clinical disease became manifest. This study also highlights a non-invasive, non-sensitizing simple assay of blood stimulation easily translatable to the field.
Cytokine release assays are useful for detecting asymptomatic individuals among immunocompetent subjects in VL-endemic areas; they can also detect the same among immunosuppressed subjects following solid organ transplantation [22, 24]. They are also useful for monitoring the success of treatment in HIV+ subjects [23]. From the present results, the concentration of IFN-γ in the supernatants of SLA-stimulated PBMC cultures appears as a major biomarker for both purposes. The present results show that the supernatants from the ARI subjects showed higher concentrations of IFN-γ than did those of the NC subjects, and that this cytokine is a 100% sensitive and specific biomarker of asymptomatic immune responders to Leishmania among persons who are HIV+.
The plasma IFN-γ, IL-2 and MIG concentrations following the SLA-stimulation of whole blood also identified 100% of these individuals. Unlike that reported for asymptomatic immunocompetent individuals, no differences were seen between the ARI and NC subjects with respect to TNF-α in the plasma of SLA-stimulated whole blood. This is probably due to the spontaneous production of TNF-α in HIV+ subjects [33, 34].
The present results show that the ARI subjects mounted a Th1-type cellular immune response to stimulation with SLA. However, the molecular and serological tests used were unable to detect Leishmania DNA or anti-Leishmania antibodies in peripheral blood respectively, as previously described for asymptomatic immunocompetent subjects from the same area [24, 25]. These results suggest that cellular immune tests should also be used when trying to identify such subjects. In line with our findings, IGRA positivity (WBA with stimulation by the antigens ESAT-6, CFP-10 and TB7.7, and quantification of IFN-γ) seems to be a better diagnostic tool for latent tuberculosis in HIV-infected patients than the Mantoux tuberculin skin test [35]. Further, the present HIV+ patients had CD4+ T cells numbers as low as 49/mm3. The capability to mount a specific cell immune response has also been described for HIV+ subjects co-infected with cytomegalus virus (with <350 CD4+ T /mm3) and Mycobacterium tuberculosis (with <100/mm3) [36, 37]. What does appear to be clear, is that HAART helps in maintaining them capable of mounting and/or maintaining a cellular immune response that might be involved in their asymptomatic status.
This work suffers from the limitation of a small sample size. Further studies should be performed with more subjects, and in different Leishmania-endemic areas to validate the use of the suggested biomarkers of asymptomatic immune responders to Leishmania in HIV+ patients. It would also be interesting to monitor the present ARI subjects to see whether they develop active VL, and how these biomarkers may change if they do. New studies are in progress to investigate these biomarkers during the asymptomatic period preceding the onset of active VL in HIV+ infected individuals from L. donovani endemic regions in Ethiopia.
The WHO guide for managing patients with leishmaniasis in Europe, published in 2017, recommends the use of the SLA-stimulated lymphoproliferation test, and WBA, plus subsequent cytokine/chemokine determinations for detecting cellular immune responses to Leishmania in immunocompetent patients [17]. The present work shows that these techniques are also valid for use with HIV+ patients living in a VL-endemic area.
In conclusion, the present results highlight the need to use cell immunity techniques to detect asymptomatic immune responders to Leishmania among HIV+ patients with no previous leishmaniasis or current symptomatology. Supernatants from SLA-stimulated PBMC cultures (CPA) can be used to look for IFN-γ, while SLA-stimulated plasma (WBA) can be used to look for IFN-γ, MIG and IL-2, all of which are biomarkers of the above condition. Combining these tests with molecular analyses could help to determine the true size of the Leishmania epidemic affecting the endemic area of Fuenlabrada and similar places. Some laboratory tests, including SLA-stimulated PBMC assay, may be difficult to perform under certain conditions. In contrast, the WBA holds much promise as a test at the point-of-care level.
|
10.1371/journal.pgen.1003136 | Chromosome Fragile Sites in Arabidopsis Harbor Matrix Attachment Regions That May Be Associated with Ancestral Chromosome Rearrangement Events | Mutations in the BREVIPEDICELLUS (BP) gene of Arabidopsis thaliana condition a pleiotropic phenotype featuring defects in internode elongation, the homeotic conversion of internode to node tissue, and downward pointing flowers and pedicels. We have characterized five mutant alleles of BP, generated by EMS, fast neutrons, x-rays, and aberrant T–DNA insertion events. Curiously, all of these mutagens resulted in large deletions that range from 140 kbp to over 900 kbp just south of the centromere of chromosome 4. The breakpoints of these mutants were identified by employing inverse PCR and DNA sequencing. The south breakpoints of all alleles cluster in BAC T12G13, while the north breakpoint locations are scattered. With the exception of a microhomology at the bp-5 breakpoint, there is no homology in the junction regions, suggesting that double-stranded breaks are repaired via non-homologous end joining. Southwestern blotting demonstrated the presence of nuclear matrix binding sites in the south breakpoint cluster (SBC), which is A/T rich and possesses a variety of repeat sequences. In situ hybridization on pachytene chromosome spreads complemented the molecular analyses and revealed heretofore unrecognized structural variation between the Columbia and Landsberg erecta genomes. Data mining was employed to localize other large deletions around the HY4 locus to the SBC region and to show that chromatin modifications in the region shift from a heterochromatic to euchromatic profile. Comparisons between the BP/HY4 regions of A. lyrata and A. thaliana revealed that several chromosome rearrangement events have occurred during the evolution of these two genomes. Collectively, the features of the region are strikingly similar to the features of characterized metazoan chromosome fragile sites, some of which are associated with karyotype evolution.
| Chromosome evolution involves both small-scale (e.g. single nucleotide) changes, as well as large-scale rearrangements such as inversions, translocations, and fusion events. We investigated mutations of the BREVIPEDICELLUS gene of Arabidopsis, which is a master regulator of inflorescence architecture. These mutations are not due to single nucleotide changes, but rather to large deletions, some spanning nearly one million base pairs. Molecular and biochemical analyses reveal that the chromosome breakpoints cluster in an area that is rich in repetitive elements and harbor multiple binding sites for nuclear matrix proteins. Data mining revealed intriguing correlations between the breakpoint cluster and hotspots of genetic recombination, regions of the chromosome that have undergone several rearrangement events during evolution, and changes in histone protein modifications. We propose that these unstable regions are chromosome fragile sites that assist in marking a boundary between gene-poor, transcriptionally repressed centromeric chromatin and a more relaxed chromatin domain that is gene-rich.
| Genome integrity depends upon the coordination of replicon and centriole duplication, chromatin condensation, and the assembly and action of the spindle apparatus. Several checkpoints regulate the progression of the chromosomal and cytoskeletal events [1], and repair systems are recruited as needed to correct replication errors and lesions caused by intrinsic and extrinsic mutagens. Intrinsic mutations may result from the interaction of DNA with reactive metabolites (e.g. hydroxyl radicals) and through the activation of mobile genetic elements. Forward genetics proceeds by employing mutagens, which can range from simple chemical mutagens such as ethyl methanesulfonate (EMS), that typically induce base substitutions, to insertional mutagens such as viral and T-DNA integration, to ionizing radiation that is often associated with rearrangements and/or deletions. Characterization of naturally occurring and induced mutants has dramatically accelerated our understanding of all cellular processes and has led to the discovery of small regulatory RNA molecules and epigenetic modifications of chromatin, two areas of intense investigation in contemporary biology.
The eukaryotic nucleus is organized hierarchically. The basic level of chromatin organization centers on the nucleosome and through various levels of organization and compaction, nucleosome strings are organized into large loop domains that are anchored to the nuclear matrix [2]. Superimposed on this, specific chromosome boundaries or territories exist within the nucleus, further defining the association of specific inter- and intrachromosomal domains [3]. Double strand breaks (DSB), created in the context of recombination activities, or due to mutagen exposure, must be repaired to ensure chromosome integrity. The juxtapositioning of specific chromosomes likely underpins the recurrent nature of specific rearrangements, for example the reciprocal translocation between human chromosomes 9 and 22 associated with chronic myelogenous leukemia.
The evolution of chromosomes has been intensively studied in many systems and is being revolutionized by high throughput sequencing technologies. In plants, genome duplication followed by translocations, inversions, centromere shifts, and the activity of endogenous mobile elements are deemed to be responsible for dispersed blocks of synteny that are observed between distantly related species [4], [5]. The breakpoints for some of these rearrangements have now been mapped onto reference genomes, and this has facilitated a high-resolution comparison between Arabidopsis thaliana and A. lyrata [6], which are believed to have evolved from a common ancestor. A series of chromosome fusion and other rearrangement events, coupled with DNA loss, has reduced the chromosome number from eight to five and the amount of DNA by approximately 40%. Within the Brassicaceae, it is clear that 24 conserved chromosomal blocks have been rearranged during evolution to constitute the genomes of mustard family members [7]. At present, it is unknown whether structures at the boundaries of the blocks promote recombination/rearrangement events or if repressive structures exist elsewhere to maintain the syntenic blocks.
We previously reported that the brevipedicellus phenotype of Arabidopsis is due to loss-of function of the KNAT1 homeodomain protein-encoding gene and that unusually large deletions occur with a high frequency [8], [9]. Here we report the characterization of the breakpoint junctions of five bp deletion alleles and the conservation of their features with metazoan chromosome fragile sites, some of which are associated with chromosome breakage events that occurred during the evolution of eukaryotic karyotypes.
Our previous work documented that five different brevipedicellus alleles were not simple alterations in the gene sequence, but rather were due to large deletions of at least 150 kbp [8]. Information on all reported bp alleles can be found in Table S1. The original bp-1 mutant, isolated by Koornneef and coworkers [10], was generated by EMS mutagenesis, an alkylating agent that typically induces G to A transition mutations. We identified the bp-2 and bp-3 mutants from a fast neutron mutagenized population; bp-5 is the result of an aberrant T-DNA insertion, and bp-11 is an x-ray induced mutant. All five of these alleles exhibit large deletions. The other characterized bp mutants appear to be simple base changes induced by EMS (bp-4, 6–8, and 10), or are insertional mutants (bp-9). Lastly, Venglat et al. [11] reported the isolation of a bp-2 mutant from a promoter tagged population; this mutant was later characterized as a point mutation.
Delineation of the boundaries of these deletion alleles would permit an analysis of breakpoint regions and perhaps provide clues as to why the region appears to be prone to such extreme segmental deletion events. We therefore employed a six-phase strategy to determine the breakpoints of the five deletions, followed by in-silico analyses to search for motifs that might be important determinants of either generating the deletion or limiting the extent of the lesion. First, for each allele, at least one breakpoint was generally localized by employing PCR, using sets of primers that span a region of approximately 1.2 Mbp. The absence of a PCR amplification product was interpreted to mean that the region was part of the deletion (Figure 1). Next, DNA gel blotting was employed to find a restriction fragment length polymorphism (RFLP) between the mutant and parental (either Columbia or Ler) DNA. In phase 3, mutant DNA was cleaved with the enzyme identified by RFLP analysis and the DNA was ligated under conditions designed to produce intramolecular events, generating circular products. This DNA was used as a template for inverse PCR (iPCR) to amplify sequences adjacent to known DNA, including the repaired breakpoints. DNA sequencing was then employed to determine the breakpoint junction sequences. Based on this information, new primers were designed to amplify mutant DNA across the breakpoints to give rise to products of predicted size and sequence and thus validate the deletion (phase 5 analysis). Lastly, computer algorithms were employed to analyze the breakpoint junctions, to search for common features/motifs.
Table 1 summarizes the five bp deletion alleles. Sequencing of iPCR products revealed that both bp-1 and bp-2 north junction sequences are composed of highly repeated sequences that are found predominantly in centromeric regions on all five chromosomes. For bp-1, the breakpoint occurs within a tandem (AG)n motif that is broadly distributed and may be a member of the LIMPET transposon family [12]. The north junction sequence could not be unequivocally localized for the following reasons. First, we found that a substantial number of primer sets, spanning BACs F5K24, F10A2, T3E15, and F28D6, did not give rise to PCR products with Landsberg erecta DNA templates (and therefore were PCR negative for all bp alleles in an Ler background; see Tables S2, S3, S4 for primer sequences, locations and amplicon status). Other primer sets generated AFLPs wherein the PCR product size differed between Ler and Col, but which can be useful tools for map-based cloning (e.g. primer sets 76/77 and T3E15-2 in Figure 1). Second, sequencing of iPCR products revealed that the junction sequence adjacent to the south bp-1 breakpoint is most similar to sequences found on other chromosomes, particularly chromosomes 1 and 3. Intriguingly, the percent match for all of these sequences is only 80–91%. The best matches on chromosome 4 include a sequence near 1 Mb, which could be interpreted as an inversion involving the centromere. Additionally there is homology to a region centered at about 4.3 Mb that has a polarity opposite that expected from sequencing the iPCR product. Subsequent pachytene chromosome in situ hybridization discounted the first possibility as cytological landmarks and the distances between fluorescent probe signals are normally distributed in bp-1 (see below). Although a complex rearrangement may be involved, it is more likely that sequence divergence between the Columbia and Landsberg erecta genomes is responsible for this disparity. Based on the locations of the primer sets, bp-1 has suffered a deletion of at least 400.6 kbp (south breakpoint to primer set F1K3) and possibly as much as 900 kbp (south breakpoint to homologous region at 4.3 Mb).
The bp-2 south breakpoint is located 13.7 kbp south of the BP gene, within the third intron of At4g01870, encoding a putative IP3 kinase. The sequence north of this breakpoint exhibits significant homology to a centromeric satellite sequence that is highly repeated on all chromosomes. The best match on chromosome 4 is to BAC F13J5 (94%), followed by the adjacent BAC, F15N16 (93%). It is noteworthy that the bp-2 allele, also in the Ler background, is PCR positive for three primer sets within BACs T3E15, F28D6 and F14G16, all of which are south of F15N16 according to the AGI reference sequence. The simplest explanation of these data is that either Columbia or Ler has suffered an inversion that positions the F13J5 and F15N16 sequences closer to the BP locus. Additionally or alternatively, sequence divergence between the centromeric satellites in Col vs. Ler might account for the less than perfect sequence homology we found. As is the case for bp-1, the repetitive nature of the bp-2 flanking sequence prohibits a phase 5 PCR analysis to amplify across the breakpoints and thus the extent of the bp-2 deletion cannot be unequivocally known. However, with the Columbia BAC tiling path as a reference, the deletion in bp-2 is at least 851 kbp based on the localized south breakpoint at AGI coordinate 5164664, and the first PCR positive fragment on the north end (in BAC T3E15 centering at 4.3 Mbp).
The locations of both junctions for bp-3, bp-5 and bp-11 were unequivocally determined. bp-3, another fast neutron generated allele, suffered a precise deletion of 386,634 bp, with a single guanosine residue inserted between the two breakpoints. bp-5, which presumably arose due to an aberrant T-DNA integration event, suffered a 140,996 bp deletion, but possesses a 19 bp ‘filler’ sequence (5′ TCCATGTAGTAAGGTAATT3′) at the junction that is 90% identical to a sequence on chromosome 3. The phase 5 PCR product sequence validates the deletion boundaries and the foreign insertion sequence, but its origin is unknown. Lastly, bp-11 is an x-ray induced allele in which the precise excision of 925,158 bp occurred.
To provide complementary data on the breakpoint locations determined by our molecular analyses, and to investigate the extent of the deletions in bp-1 and bp-2, we coupled cytological analyses with pachytene chromosome fluorescence in situ hybridization (FISH), using five probesets that span chromosome 4 from 0.7 Mbp to 6.4 Mbp (Figure 2). In Columbia, an inversion event involving pericentric heterochromatin has resulted in a heterochromatic knob in this ecotype, which has no counterpart in Ler [13], [14]. Cytological examination of DAPI stained chromosomes enabled us to measure distances between the NOR4 ribosomal gene cluster at the north end and other cytological landmarks: the heterochromatic knob (hk4S, in Columbia backgrounds), and CEN4 (Table 2). In addition, the size of hk4S and CEN4 could be evaluated. These analyses revealed that the distance from NOR4 to CEN4 and the size of CEN4 are approximately equal in both parental lines and in all bp alleles. In the Columbia based alleles, the size of hk4S is also similar to the wildtype parent line, as was expected.
The probesets for FISH were chosen to evaluate distances between distinct cytological landmarks and the probe target sequences. The two rhodamine-associated probesets RED1 and RED2 bracket hk4S in Columbia (Figure 2a) and measurements of NOR4 to RED1 revealed no significant differences in either of the two parental ecotypes, or in the bp alleles. As expected, due to the pericentric inversion in Columbia that generated hk4S, the RED1 to RED2 interval is smaller in Columbia than in Ler, and the distances for the bp alleles are similar to their parent ecotypes. We conclude that the bp-associated deletions do not involve chromatin north of the centromere. In this regard, the possibility that the bp-1 north junction sequence, which exhibits homology to the 1 Mbp region suggestive of an inversion involving the centromere, can be ruled out in favor of an event occurring south of the centromere in a region that has diverged between Columbia and Ler.
To correlate the size of the deletions with cytological measurements, three additional probesets were employed. GREEN 1 sequences are located very near CEN4 at 4.3 Mbp, while two additional GREEN probes bind in the 6–6.4 Mbp region, south of the BP locus at 5.15 Mbp (Figure 2). We expected that the extent of the deletions, as gauged by molecular analyses and sequence comparisons, could be roughly correlated with changes in the distance between GREEN1 and GREEN2 (G1/G2). Indeed, this is the general trend we observed as bp-5 (141 Kbp) and bp-11 (925 Kbp) deletions gave intersignal distances of 2.1 µm and 1.7 µm, respectively, compared to the wildtype Columbia distance of 2.7 µm. For the Ler-based alleles bp-1 and bp-2, for which the north breakpoints could not be established, we observed similar reductions in signal distances compared to the Ler parental background with bp-2 exhibiting a shorter G1/G2 length, and thus a larger deletion, than bp-1.
To our surprise, we found that the G1/G2 distance in Columbia is markedly different from Ler, being on average about 2.7 µm in Columbia, but over 5 µm for Ler. These measurements were reproducible and statistically significant, implying that a major polymorphism exists between these two ecotypes. Because the RED2 to GREEN1 distances do not vary significantly between Col and Ler and because GREEN1 localizes very close to the DAPI stained centromere in both ecotypes, it seems likely that the G1/G2 polymorphism is due to an indel occurring between them, or possibly an inversion that moved the G2/G3 probeset sequences in a manner similar to the event that created hk4S. The latter possibility might be resolved by using different colored G2/G3 probes in future experiments.
Simple BLASTn searches coupled with gene annotations available through the TAIR database permit the identification of genetic elements at breakpoint regions. As the bp deletions are found in the pericentromeric region, there are many occurrences of transposons, pseudogenes, and repetitive sequences. Some of these may be expressed, based on annotated cDNAs and/or expressed sequence tags that map to these sequences, but due to their repetitive nature, it is not clear if they are the actively expressed copies. Discounting transposon/repeat-associated sequences, in the region of 4.2 Mbp to 5.25 Mbp, there are 31 annotation units for which cDNA/EST annotations exist, and another 16 predicted genes for which there are no cDNA/EST sequences. The largest deletion, bp-11, which spans over 925 kbp, has lost 27 genes, while bp-5, the shortest deletion allele spanning 141 kbp, is missing 10 genes (see Table S5). Within the set of 31 genes, there are two pseudogenes, nine that encode unknown proteins, and three that encode proteins with conserved domains of unknown function. Genes with more complete annotations are mostly members of gene families, though a few are single copy genes. Under normal growth conditions, the Ler based alleles are indistinguishable from one another, and the Columbia based alleles are also similar to one another. It must be appreciated that the bp phenotype is enhanced in the Ler background due to the absence of the ERECTA protein kinase [9]. Our initial work with bp-2 demonstrated that the bp mutant phenotype can be rescued by transformation with a wildtype gene [8]; thus under normal growth conditions, the deleted genes seem to be dispensable.
A suite of bioinformatics algorithms (BLAST, MEME, RepeatMasker) was used to interrogate the breakpoint sequences to discover commonalities that might inform our understanding of how the lesions were generated/repaired. Our strategy was to analyze north and south donor sequences. A north donor consists of 1 kbp of DNA north of the north breakpoint linked to 1 kbp of adjacent DNA that was deleted. Similarly, a south donor consists of 1 kbp of DNA that was deleted adjacent to 1 kbp of DNA south of the south breakpoints. As the breakpoint regions lie in the pericentromeric chromatin, most of the breakpoint donor sequences were found to possess one or more known repeats and/or transposon remnants (Figure 3). There is no common sequence motif shared by all alleles, but several short repeats of 20–50 nucleotides (motifs 1–5, see also Figure S1 for alignments and their p-values) are conserved in three to four alleles. The north flanking sequences are predominated by satellite (bp-2), Athila/gypsy transposons (bp-3, bp-5, bp-11) or LIMPET elements (bp-1), most of which are abundant and dispersed throughout the pericentromeric region. The south flanking regions, which cluster within 80 kbp of each other in BAC T12G13, tend to be sparse in repetitive sequences, but are A/T rich and 80% possess motif 5, a T-rich element that is also found in two of the north flanking regions. Statistically, the p-values for motifs 1–5 range from 2.8×10−22 to 5×10−6, lending credibility to their association with lesion formation and/or repair.
Sequences at the breakpoints offer little evidence of homologous recombination. Although both the north and south flanking regions of bp-1 contain sequences with homology to LIMPET elements, the junctions are not a continuous sequence in the repaired product, which argues against homologous recombination. A simplistic explanation for this lesion is that it results from illegitimate recombination between (AG)n rich repeat sequences located at approximately 4.2 and 5.2 Mbp, but the significance of the flanking LIMPET sequences cannot be evaluated. Additionally, the close proximity of a (CT)n rich sequence juxtaposes two complementary sequences that could generate a long hairpin structure with the breakpoint occurring in a loop of four nucleotides at its apex. Complex secondary structures are predicted at or near all of the donor sequences, with the exception of bp-1 (see Figure S2).
The apparent lack of homology between the paired north and south flanking sequences indicates that the DNA damage is likely repaired by non-homologous end joining (NHEJ). In only one instance (bp-5) is there microhomology between the 5′ and 3′ donor sequence junctions, which might portend the involvement of microhomology-mediated recombination.
Matrix attachment region (MAR) prediction algorithms suggested the possibility that some of the sequences near the breakpoint junctions contain binding sites for the nuclear matrix. There are several known components of the nuclear matrix, including AHL1, an AT hook domain containing protein that has been shown by biochemical and cytological assays to be associated with the nuclear matrix [15]. We cloned an AHL1 cDNA behind an inducible promoter and expressed it as a His-tagged protein in E. coli. Southwestern blot analysis was then undertaken with end-labeled probes, including a positive control (a plastocyanin gene, PC, [16]), a negative control (a histone H1 gene fragment, At1g06760) and several fragments near the breakpoint junctions (2S, 3N, 3S, 5S, 5N, and 11S). Figure 4 shows that the histone H1 probe does not bind to the AHL1 protein. The positive control PC1 probe as well as the 2S, 3N and 5N probes bound weakly, but above background levels, while the 3S, 5S and 11S probes exhibited strong binding to AHL1. We conclude that these probes, representing regions predicted to contain MAR binding sites, do indeed bind to a known matrix associated protein. Intriguingly, the 3S, 5S and 11S probes are located within 26 kbp of each other at the south end of BAC T12G13, which harbors all of the south end deletion breakpoints. It is conceivable that BAC T12G13 contains sequences that organize chromatin loops and that in the bp deletion mutants, resection of the initial lesions is limited by either a complex chromatin structure (e.g. boundary element possessing extensive secondary structure) and/or an attachment point on the nuclear matrix.
The lack of an ordered array of large clones of Ler DNA precludes both direct Col/Ler synteny comparisons as well as the construction of genome array chips for chromatin immunoprecipitation analysis. Nevertheless, as two of our bp alleles are derived from Columbia, we suspected that data mining might prove fruitful for correlating the breakpoint regions with established chromatin features and genetic data. We reasoned that segments of the genome that are active in recombination might have features that could predispose them for rearrangement events. High-resolution recombination mapping along chromosome 4 revealed very little recombination in the hk4S and CEN4 regions, as expected, but also identified regions which exhibit high levels of recombination (hotspots, [17]). Figure 5 shows that the 5–6 Mbp region is very active in recombination, with one of the hotspots in the same region where the south breakpoints of all bp alleles cluster. Interestingly, this region also includes the HY4 locus, where multiple large deletion alleles have been reported [18].
A. thaliana and A. lyrata likely evolved from a common ancestor and several chromosomal fusion and rearrangement events have occurred to reduce both the size of the genome and the number of linkage groups in A. thaliana [6], [7]. Comparisons of the genomic sequences of the two species reveal that several regions of the ancestral chromosomes six and seven were fused to generate A. thaliana chromosome 4 [6], [7 and references therein]. Importantly, some of the major rearrangement events also map to the region where the south breakpoints are clustered and in other areas, for example in the 16–17 Mbp region where other rearrangement events have occurred, recombination hotspots also exist (Figure 5B). Comparative sequence analysis of the BP/HY4 region of the A. thaliana and A. lyrata genomes indicate that one or more segments of these genomes underwent transposition/inversion events, providing additional evidence that the region is recombinogenic and prone to chromosome rearrangement events that are possibly associated with speciation (Figure 5C).
Epigenomic data mining revealed that the defined north end breakpoints (bp-3, bp-5 and bp-11) possess common histone modifications, specifically H3K27me1, H3K9me2 and H4K20me1, and in addition exhibit 5–15% 5-methylcytosine (Figure 6, see also Figure S3). This combination of chromatin modifications is associated with transposable element rich regions found in pericentric heterochromatin [19]. The south end breakpoints, with the exception of bp-1, differ markedly from the north ends, sharing no chromatin marks and exhibiting very low levels of 5-methylcytosine. The bp-2 south breakpoint, located within an expressed gene, contains ubiquitinated H2B, H3K27me3 and H3K4me2, all typically associated with euchromatin [19]. The other clustered south breakpoints contain either of the two latter modifications (bp-5, bp-11), but are generally devoid of chromatin marks. However, the local regions possess a variety of modifications and the four basic chromatin states [19] are interspersed (see Figure S3). It is conceivable that a particular combination of chromatin modifications may promote genome instability, or, as we observed for the clustered south breakpoints, a dearth of chromatin modifications and lack of 5-methylcytosine may be indicative of a chromatin state that underpins genomic instability/recombinogenic potential. In any event, features inherent to BAC T12G13 represent a boundary element or transition zone in which the chromatin state switches from one bearing heterochromatic marks to one indicative of a more euchromatic state (Figure 6B).
The large deletions that we document for the BP locus, as well as the close proximity of the HY4 locus where numerous large deletions have also been reported [18], encourages speculation that the region is generally unstable and prone to deletions and other gross chromosomal rearrangement events. We therefore examined literature reports and employed mutant germplasm search engines to catalog the locations of deletions greater than 25 bp to ascertain if the BP/HY4 region is overrepresented in this data set. Figure S4 shows that the 5 bp alleles and the 15 hy4 alleles reported by Bruggemann and coworkers [18], constitute 20 of the 120 deletion mutations associated with structural genes. We conclude that the BP/HY4 region is on average more prone to deletion/rearrangement events than most other regions of the genome.
Genome evolution is driven by a variety of factors, not the least of which are gross chromosome rearrangements that are sponsored by the activity of mobile genetic elements, first reported by Barbara McClintock [20]. More recently, mechanistic analyses of structural variation in a variety of organisms have focused on errors in replication and recombination as causative [21]–[23]. Supporting evidence is provided by the existence of sequences with homology to transposons and/or repeat motifs at breakpoint junctions, evoking homologous recombination or microhomology mediated non-homologous end joining (NHEJ) mechanisms for lesion repair. In the absence of evidence for homologous recombination and outside of the context of meiosis, NHEJ is a pathway to repair [24], and ligation of free ends is a stochastic process, often resulting in translocations. The pericentric chromatin in which BP resides would seem to be inherently unstable due to the presence of autonomous and non-autonomous transposons of various types, numerous repetitive sequences, and sequences of unusual base composition that may impose constraints on replication fidelity due to the formation of secondary structures.
In instances where mutagenesis is employed and which result in gross chromosomal rearrangements, it is not clear if the initial event is associated with DSB formation followed by resection of the free ends and final repair, or alternatively, if there are two DSBs that upon repair result in loss of the intervening sequence. With respect to the bp deletions, it is intriguing that several different mutagenic agents (EMS, fast neutrons, T-DNA insertion and x-rays) produce the same effect: large deletions. Four of the five bp alleles possess Athila LTR retrotransposon elements at or near the north breakpoint junction and the fifth contains satellite repeats at the junction. As there is no corresponding transposon/satellite DNA south of the breakpoint, there is little evidence to support transposon movement, homology-based recombination or replication slippage. An exception is bp-5, where the microhomology found at the junction could suggest one of the latter two processes may have been involved. In the absence of evidence for these types of events in the other alleles, we speculate that the initial break likely occurred during DNA replication, and that resection, likely mediated by the MRN complex [25], extended the region. The extent of DNA loss could be limited by either a change in the structure of DNA (e.g. alternating dinucleotides or elaborate secondary structures, as is observed for several bp alleles), a change in the chromatin microenvironment (e.g. accessibility changes due to differences in histone modification patterns and binding by their associated proteins), and/or a repressive influence of protein factors (e.g. encountering a nuclear matrix protein complex as we report for the south breakpoint cluster).
Chromosome fragile sites have long been recognized in metazoans as regions which are prone to breakage when replication stress occurs, and activation (breakage) of some are implicated in various disease states and cancer [26], [27]. Some of the common fragile sites that have been characterized at the molecular level are associated with breakpoints that delineate homologous syntenic blocks that have persisted during the evolution of eukaryote karyotypes [28]. Although a wealth of data correlates recurrent deletions and rearrangements with these fragile sites, very little is known about the initial breakage events, but rather only the sequences of junction regions provide clues as to the repair pathway. Some fragile site breakpoints possess microhomologies indicative of replication slippage or recombination based repair, whereas others possess no homology at the junctions. In the latter case, NHEJ must occur and likely involves structural features (e.g. attachment points on the nuclear matrix) that juxtapose free ends. Common fragile sites are A/T rich, possess various repeats and classes of repetitive DNAs (particularly LINE/LTR type elements), are gene-poor regions, and occupy large tracts of eukaryotic genomes, sometimes extending over hundreds of thousands of nucleotides [29], [30]. Our analyses of the SBC revealed that four of the five alleles harbor Athila/LTR retrotransposon elements at or near the breakpoints, which are A/T rich. The A/T rich sequences may contribute to secondary structure formation (see below) and could serve to anchor the chromatin fiber to MAR regions, which are known to be enriched for enzymes of DNA metabolism [31]. Indeed, the inhibition of topoisomerase I by camptothecin almost completely eliminates CFS breakage in cultured mammalian cells [32].
Fragile sites are found at the interface of chromosome R and G bands, classically defined as being early and late replicating, respectively [33]. Recent studies suggest that CFS activation (breakage) may be due to differential utilization of replication origins such that the CFS zone may experience replication stalling or fork collapse [34], [35], and secondary structures likely contribute to this process [36]. The paucity of replicon initiation in these regions necessitates the use of more distant origins and delays replicon completion, in part explaining their location at R/G boundaries. In Arabidopsis, high resolution profiling of replication in suspension cultured cells has revealed the locations of origins and their replication timing [37], [38]. The south breakpoint cluster in T12G13 is situated between two distantly separated origins that span 271 Kbp; for comparison, the interorigin median and mean distances along chromosome 4 are 51.1 and 77.2 kbp, respectively. We propose that the south breakpoint cluster possesses the hallmark features of metazoan chromosome fragile sites, harboring sequences that promote replication fork collapse, and that loss of DNA may be limited by association of local sequences with the nuclear matrix, where topoisomerase/ligase activities can coordinate repair by a NHEJ (or other) mechanism. In addition to the deletions associated with BP/HY4, this stochastic process may occasionally generate rearrangements and chromosomal fusion events that underpin chromosome evolution.
Arabidopsis has been extensively used for comparative evolutionary genomics studies in plants. The 1001 Genomes Project has generated a wealth of sequence information, and numerous SNPs and indels are known to exist in many ecotypes [39]–[41]. The vast majority of this work has employed high throughput sequencing technologies, generating short sequence reads that are mapped onto scaffolds of reference genomes. While this strategy is useful and saturation can be achieved, it has two major disadvantages. First, some syntenic relationships may be masked, as many inversions and translocations cannot be accurately mapped. This is also the case for sequences associated with repetitive DNAs. Second, some copy number variants and their locations may go undetected and unrecognized, depending on the extent of divergence amongst them and their flanking sequences. While paired end mapping can be used to discover and verify structural variation [42], only through the sequencing of contiguous long clones (e.g. BACs) can such sequence anomalies be accurately located and quantified. In Arabidopsis, several such studies have revealed striking departures from the Columbia reference genome. For example, comparative sequence analysis of a region around 100 map units on chromosome 1 in Columbia vs. Ler revealed that the region in Columbia is approximately 135 kbp, while the comparable region in Ler is only 71 kbp [43]. This appears to be due to several gross rearrangement events including two different ecotype specific duplications, a deletion and several ecotype specific transposition events that ultimately led to the shorter Ler region having fewer R genes than its Columbia counterpart [43]. In a similar vein, Lai et al. [44] resequenced a 371 kbp region of chromosome 3 from Columbia and Ler, mapping these reads onto both the Columbia reference genome as well as the more recent Wellcome Trust generated Ler draft genome, and discovered 61 misassemblies and large structural variants not represented by the draft genome. Lastly, the quartet mutation has afforded the possibility of tetrad analysis in Arabidopsis, and Lu and coworkers [45] sequenced the F1 genomes of a Col/Ler cross, reporting numerous known and heretofore unknown SNPs and indels. Analysis of the indels revealed that a deletion of 19.2 kbp encompasses the region of the bp-11 north breakpoint. These high resolution sequencing projects reveal that the genomes of Arabidopsis accessions are highly polymorphic and subject to rapid change as a result of normal meiotic recombination events and other rearrangements that may be due to replication errors and mobile element activity.
While the power and economics of high throughput sequencing technologies cannot be disputed, discovering and mapping large structural variation in comparative genomics studies ultimately will depend upon more painstaking mapping and the generation and analysis of long clones. Fluorescent in situ hybridization or chromosome painting offers a middle ground for analyzing gross chromosomal architecture. Our experiments reveal that the two most commonly used Arabidopsis ecotypes, Columbia and Landsberg erecta, possess at least two major structural differences on chromosome 4. The heterochromatic knob, which can be readily detected by DAPI staining, arose in Columbia via a pericentric inversion, moving a heterochromatic region towards the north end of the chromosome [13], [14]. South of the centromere, there appears to be an indel that generates a significant length difference between the two GREEN probeset signals located at 4.3 Mbp and 6 Mbp of the Columbia reference genome. The precise nature of this polymorphism is unknown, but a comparison of homologous regions in A. lyrata and A. thaliana (Columbia) indicates that on two contigs that span the region, A. lyrata possesses over 600 kbp more DNA than Columbia. It is possible that this region has persisted in the Ler genome, accounting for the longer GREEN1/GREEN2 intervals that we observed. Given that similar polymorphisms are likely to exist on the other chromosomes, which could involve the mobilization of large blocks of sequence to perhaps distant sites (e.g. an inversion involving over 1 Mbp), map-based cloning of some genes could be complicated and require protracted efforts.
Advances in chromatin immunoprecipitation have permitted high throughput approaches for elucidation of the histone codes associated with genome elements. Roudier et al., [19] conducted integrative epigenomic mapping in Arabidopsis, tracking 12 chromatin modifications to define four primary chromatin states associated with different coding and noncoding sequences. As in other organisms, the Arabidopsis pericentromeric DNA is enriched for several histone modifications, in particular methylation at H3K27, H3K9, and at H4K20 residues. In the context of maintaining genome integrity, H4K20 modifications are intriguing. In mammals, the H4K20 methylation state plays important roles in DNA damage repair as well as in class switch recombination (CSR) during the maturation of antibody producing cells [46]. CSR employs a NHEJ mechanism to exchange constant regions of immunoglobulin genes during B cell differentiation [47]. In mouse cells in which H4K20 methyl transferases are silenced, H4K20me1 accumulates, and this is associated with translocations and deletions of the IgH locus [46]. To our knowledge, there are no reports in Arabidopsis on the distribution and role of either di- or trimethylated H4K20, but these modifications might prove to associate with the instability that we observe.
The alleles bp-1, bp-2, bp-3, and bp-5 were described by Douglas et al. [8]. bp-11 was obtained through ABRC (CS3161). Table S1 contains information on all characterized bp mutants. Plants were grown in environmental growth chambers with a 16 hr day/8 hr night cycle at 22°C under fluorescent lighting of approximately 100 µE/m2. Bacterial artificial chromosome clones were obtained from ABRC and DNA was prepared by employing Qiagen midi-prep columns.
General molecular techniques were carried out as described by Sambrook et al. [48]. Genomic DNA for PCR was prepared using Sigma GeneElute columns. A list of primers used for determining the presence or absence of a locus, as well as for use in inverse PCR and phase 5 PCR is given in Tables S2, S3, S4. Breakpoint junctions were cloned by employing inverse PCR. Genomic DNA was digested with a restriction enzyme that was shown by DNA gel blotting to give rise to an RFLP. The digested DNA was purified, diluted to approximately 1 µg/ml and subjected to overnight ligation to promote recircularization. Inverse PCR primers were then employed in PCR reactions to generate products containing the breakpoint junctions. These molecules were cloned into pJet1.2 (Fermentas) and sequenced (The Centre for Applied Genomics, Toronto).
For Southwestern analysis, a cDNA encoding the MAR binding protein AHL1 [15] was cloned by conducting RT-PCR on silique RNA, using oligo dT to prime first strand synthesis, and two primers: MAR Forward Nhe: 5′ AGGCTAGCGTCTTAAATATGGAGTCTACC 3′ and MAR Backward Bgl II: 5′ AAAGATCTGATTTCAAGTTACATTGACATTAATATCGG 3′. The underlined sequences represent engineered Nhe I and Bgl II sites that were used to clone the cDNA into the expression vector pRSET B (Invitrogen). Authenticated clones were mobilized into BL21 cells and expression of AHL1 induced by the addition of IPTG to a final concentration of 1 mM. After several hours of growth, induced and uninduced cultures were harvested by centrifugation and resuspended in Laemmli buffer (100 µl per 1 ml of culture), boiled, and stored at −20°C until needed. For Southwestern blotting, aliquots of protein extracts from induced and uninduced cultures were subjected to SDS-PAGE and transferred to nitrocellulose. The membrane was placed in 3% gelatin/TBS and kept at 4°C overnight. Protein refolding and blocking was carried out for 2 hours in a solution of 5% non-fat dry milk (Carnation), 20 mM Tris, pH7.6, 150 mM NaCl, 10 mM MgCl2, 0.25 mM DTT, 0.05% Tween-20, and 10 µg/ml salmon sperm DNA. DNA fragments to be used for probes were generated by PCR (see Table S6 for primer set information) and cloned into pJet1.2 (Fermentas). Probe DNA was prepared by fill-in reactions using alpha 32P-dCTP and the Klenow fragment of DNA polymerase I (Invitogen) on restriction fragments. Binding was performed in the refolding buffer except that the non-fat milk concentration was reduced to 0.5%. The membranes were gently agitated at room temperature for 2 hours in approximately 3 ml of liquid containing the radiolabeled probes, then washed four times in 20 mMTris pH7.5, 200 mM NaCl, 10 mM MgCl2, 0.05% Tween-20, 0.1% Triton ×100, 0.25 mM DTT and 10 µg/ml salmon sperm DNA. Autoradiography was then employed to detect binding. In no case was binding detected using uninduced extracts. Probe labeling was assessed by gel electrophoresis and autoradiography and all probes were deemed to be of comparable specific activity.
The BLASTn search tool, the MEME algorithm (National Biomedical Computation Resource website; version 4.9.0; http://meme.sdsc.edu/meme/cgi-bin/meme.cgi; Accessed 2012 Nov 7 [49]) and the RepeatMasker algorithm (Institute for Systems Biology website, open-3.3.0. Available: http://www.repeatmasker.org/cgi-bin/WEBRepeatMasker. Accessed 2012 Nov 7. [50]) were employed to identify common motifs and repetitive elements. Potential secondary structures were determined by employing the Mold algorithm [51]. The EPIGARA database (Arabidopsis epigenetics and epigenomics group website; version 1.69; http://epigara.biologie.ens.fr/index.html. Accessed 2012 Nov 7.) of chromatin modifications was interrogated for locations of modifications and their proximity to the bp breakpoint junctions. Data was extracted from ChIP/chip studies conducted by Roudier et al. [19] and array based profiling of methylated DNA [52]. The pie charts in Figure 6 were generated by taking the IP/input log ratio for each region and dividing this by the sum of all IP/inputs for the region.
Staged floral buds of wildtype Columbia and the Columbia based bp-5 and bp-11 alleles, along with Landsberg erecta (Ler) and the Ler derived bp-1 and bp-2 alleles were used as the starting materials. Pachytene chromosome spreads were prepared and identified according to the method of Stronghill and Hasenkampf [53]. Spreads were then subjected to fluorescence in situ hybridization as described by Lysak et al. [54]. Five chromosome 4 probesets, comprised of bacterial artificial chromosomes (BACs), were used to determine the contour distances between these five sequences, which span chromosome 4 from 0.8 Mbp to 6.4 Mbp (AGI coordinates). Probes were generated by employing a Nick Translation Kit (Roche). Two of the five probesets (north of the centromere) were biotin-labeled BAC clones bracketing the heterochromatic knob region in Columbia: Red 1: BACs T7B11, T2H3 and Red 2: BACs T4B21, T1J1, T32N4. Detection of these signals was facilitated by goat anti-biotin antibodies (Vector Laboratories) and a secondary donkey anti-goat Cy3 conjugated antibody (Jackson Immunoresearch). Three additional probesets were DIG-labeled BAC clones (south of centromere), bracketing the region of the BP locus: Green 1: BACs F28D6, T3E15; Green 2: BACs T15G18, T25P22 and Green 3: T9A4, F24G24. These hybridization signals were detected by employing a mouse anti-DIG primary antibody and a donkey anti-mouse FITC conjugated secondary antibody (Jackson Immunoresearch). Spreads were examined using a Zeiss axiophot epifluorescent microscope and a Plan-Neofluar 100×/1.3NA oil immersion objective lens. Northern Eclipse 5.0 software was used to capture images and measure the distance between FISH signals. Merged images were created using Photoshop CS5 software.
|
10.1371/journal.pbio.1000568 | The Gene Regulatory Cascade Linking Proneural Specification with Differentiation in Drosophila Sensory Neurons | In neurogenesis, neural cell fate specification is generally triggered by proneural transcription factors. Whilst the role of proneural factors in fate specification is well studied, the link between neural specification and the cellular pathways that ultimately must be activated to construct specialised neurons is usually obscure. High-resolution temporal profiling of gene expression reveals the events downstream of atonal proneural gene function during the development of Drosophila chordotonal (mechanosensory) neurons. Among other findings, this reveals the onset of expression of genes required for construction of the ciliary dendrite, a key specialisation of mechanosensory neurons. We determine that atonal activates this cellular differentiation pathway in several ways. Firstly, atonal directly regulates Rfx, a well-known highly conserved ciliogenesis transcriptional regulator. Unexpectedly, differences in Rfx regulation by proneural factors may underlie variations in ciliary dendrite specialisation in different sensory neuronal lineages. In contrast, fd3F encodes a novel forkhead family transcription factor that is exclusively expressed in differentiating chordotonal neurons. fd3F regulates genes required for specialized aspects of chordotonal dendrite physiology. In addition to these intermediate transcriptional regulators, we show that atonal directly regulates a novel gene, dilatory, that is directly associated with ciliogenesis during neuronal differentiation. Our analysis demonstrates how early cell fate specification factors can regulate structural and physiological differentiation of neuronal cell types. It also suggests a model for how subtype differentiation in different neuronal lineages may be regulated by different proneural factors. In addition, it provides a paradigm for how transcriptional regulation may modulate the ciliogenesis pathway to give rise to structurally and functionally specialised ciliary dendrites.
| Early during development, cells differentiate and take on specialized forms and functions. This requires the activation of specific genes for different cellular pathways. Our study addresses how this activation is regulated in the developing Drosophila nervous system. In this model, it is well known that proneural transcription factors are involved in directing cells to differentiate into various types of neurons. However, the mechanism by which they choreograph the activation of genes for neuronal differentiation is not clear. In this study, we focused on events leading to differentiation of mechanosensory neurons, which have specialized dendritic processes that mediate sensory perception. In these developing neurons we profiled the time course of gene expression that is triggered by the proneural factor atonal. Our analysis revealed the activation of genes required for the formation of these specialized dendrites, called cilia. We then identified several ways in which atonal regulated these genes. First, it activates intermediate transcription factors that regulate different subsets of differentiation genes. Second, in at least one case, atonal activates a differentiation gene directly, one that is involved in the formation of cilia (ciliogenesis). These findings offer new insight into how proneural factors regulate specialized neuronal differentiation pathways.
| Once an embryonic cell is committed to a particular fate, it is likely that a precisely ordered progression of gene expression is required to coordinate the complex cell biological events that eventually lead to its terminal differentiation. Determining how this progression is regulated is an important step towards understanding how cells acquire specialised morphologies and functions. In the developing nervous system, cell fate commitment is initiated by the activity of proneural basic-helix-loop-helix (bHLH) transcription factors [1]. In vertebrates, atonal (ato)-related proneural genes are required for neurogenesis in the spinal cord and cortex (neurogenin), cerebellum (atoh1), and retina (atoh7) [1]. atoh1 is also required for the formation of mechanosensory cells in the inner ear and in skin [2],[3]. In Drosophila, ato itself specifies the precursors of several specialised sensory neuron types, including photoreceptors and mechanosensory chordotonal (Ch) neurons, which mediate hearing and proprioceptive feedback during locomotion [4]. Whilst proneural genes are intensively studied, little is known of how their function leads to specific programs of neuronal differentiation.
ato expression in the ectoderm leads to sense organ precursor (SOP) specification in a process that is refined by Notch signalling. After commitment, SOPs divide several times asymmetrically before the 4–5 progeny cells interact and terminally differentiate to form the neuron and support cells of the mature Ch sense organ (Figure 1A–C). The function of ato and other proneural factors in SOP fate determination is relatively well studied. Indeed, known proneural target genes are almost all concerned with SOP specification or fate maintenance [5]–[9]. It is not clear, however, how its function as ‘master regulator’ leads to subsequent neural development. Since ato is expressed only transiently during SOP formation, a likely hypothesis is that it initiates a gene regulatory cascade, which eventually regulates differentiation genes. The nature of this cascade and its regulation have not been elucidated.
In contrast to the dearth of knowledge of the regulatory cascade, more is known of Ch neuron terminal differentiation itself. Notably, Ch neurons develop a highly structured dendrite based on a modified cilium [10]–[12]. Ciliogenesis is a conserved, highly ordered process involving the coordinated action of hundreds of proteins [13]. In vertebrates, ciliated cells are widespread, both in the PNS (e.g. photoreceptors, olfactory neurons) and other adult tissues (e.g. kidney, lung), and developing cells have a primary cilium that is required for signal transduction for a number of paracrine pathways [13]. In contrast, the only ciliated cells in Drosophila are sensory neurons and sperm. As a consequence, genetic analysis of defective sensory neuron differentiation in Drosophila has enabled the discovery and characterisation of a number of ciliogenesis genes [14]–[16]. These include genes required for the specialised transport process known as Intraflagellar Transport (IFT) [16] and homologues of genes disrupted in the human ciliopathy, Bardet-Biedl syndrome (BBS). Ciliogenesis is one of the key differentiation events that must be initiated by Drosophila proneural factors.
An important question in the regulation of cellular diversity is how core cell biological pathways are modified to give distinct cell types. Cilia perform a wide variety of specialised functions, but it is poorly known how the core ciliogenesis program is modulated in different cell types. The ciliary dendrite of Ch neurons is anatomically and physiologically distinct from those of other Drosophila sensory neurons (notably the External Sensory (ES) neurons) (Figure 1A) [17],[18]. Ultimately, these subtype-specific differences in the ciliary dendrite must be regulated by the proneural factors, which have well-known neuronal subtype determining properties in both invertebrates and vertebrates [1]. Whilst ato directs the formation of Ch precursors, another proneural gene, scute (sc), performs this function for ES precursors. sc's function is likely to be mediated partly by the homeodomain factor, Cut, [19] but little is known of Cut's molecular function. Apart from the involvement of cut, it is at present entirely unknown how subtype specification by transiently expressed proneural factors is translated into differences in neuronal phenotype, including the modulation of ciliogenesis.
In order to bridge the gap between proneural factor function and the activation of genes required for neural terminal differentiation, we used expression profiling to characterise the progression of gene expression during Ch neuron development. A time course in the onset of differentiation gene expression can be discerned. We then show that ato regulates some of these events through a number of intermediate transcriptional regulators. The gene for Regulatory factor X (Rfx), a well-known and highly conserved regulator of aspects of ciliogenesis, is regulated differently by proneural genes in Ch and ES lineages. We propose that this links proneural subtype specification to differences in ciliogenesis. We also identify a novel forkhead-related factor that is required to regulate genes for specialised aspects of Ch neuron function. In addition, we find that some putative differentiation genes are expressed surprisingly early in neural development and that ato may directly regulate at least one such gene.
For expression profiling during Ch development, ato-expressing cells were isolated from timed collections of embryos. ato-expressing cells were marked by GFP expression from an atoGFP reporter gene construct (atoGFP cells). This reporter gene is expressed predominantly in Ch precursors and their progeny but also in other ato-expressing cells including the developing larval eye (Figure 1D). Embryos from timed collections were dissociated and atoGFP cells isolated by FACS (Figure S1). Such cells were isolated from embryos at three time points corresponding to the first 3 h of neural development (t1–t3) (Text S1). t1 coincides maximally with ato expression (and therefore should include direct target genes), whereas later time points reflect subsequent post-ato development as the precursors divide leading up to differentiation (Figure 1C).
Expression profiling revealed the number of differentially expressed genes in atoGFP+ versus atoGFP− cells (referred to as ‘ato-correlated genes’) to be 330, 456, and 411 at t1, t2, and t3, respectively (≥1.5-fold enriched, ≤1% false discovery rate (FDR)). Set analysis of genes enriched in atoGFP cells (ato-correlated genes) shows a clear time course of expression changes, with 69, 141, and 210 genes unique to t1, t2, and t3, respectively (Figure 1E; Tables S1, S2, S3). This suggests an increase in the complexity of gene expression as development proceeds to differentiation. Manual inspection of ato-correlated genes ranked by fold change shows a high representation of known neurogenesis genes (Figure 2; Tables S1, S2, S3). For instance, among the top ranked genes at t1 are spineless, twin of eyeless, cato, couch potato, dachshund, ato, Rfx, senseless, and BarH1, all of which are associated with aspects of neural development. Gene ontology (GO) analysis shows strong enrichment of GO annotation terms related to PNS development across all three time points (Tables S4, S5, S6, S7; Text S2). There is a clear progression over time in the representation of genes (Figure 2) and relevant GO terms (Figure 1F; Text S2).
Over-representation of GO terms identified the enrichment in developing Ch cells of genes associated with ciliogenesis (Text S3; Figure 1F). Analysis of over-represented protein domains also highlights domains associated with some classes of ciliogenesis gene (Text S4; Figure S2; Table S8). To characterise the Ch expression of ciliogenesis genes further, we compared the t3 expression data with collections of genes previously linked to ciliogenesis (Tables 1, S9). The Drosophila Cilia and Basal Body database (DCBB) has been compiled from a number of genetic and proteomic sources to contain genes, or orthologues of genes, implicated in cilia or basal body structure or function [20]. ato-correlated genes at t3 represent 3.0% of the genome but include 10.1% of DCBB genes—a highly significant over-representation (p = 3.3×10−19) (Table 1). Another study identified potential ciliogenesis genes from comparative genomic analysis of ciliated and non-ciliated organisms [16]. Strikingly, Ch cells at t3 are 8-fold more enriched than expected by chance for genes implicated in this study (p = 8.7×10−23) (Table 1). Moreover, the subgroup of these genes most associated with compartmentalised ciliogenesis are 27-fold enriched at t3 compared with expected (p = 3.5×10−22) (23/28 genes) (Table 1). For many of these genes, our expression data provide the first confirmatory evidence of a potential role in ciliogenesis. Our data also provide new candidate ciliogenesis genes.
Since the atoGFP cells will divide to produce both the Ch neurons and their support cells, ato-correlated genes may include support cell genes in addition to neuronal genes. Few such genes are currently known, but several of these are enriched at t3 (but not earlier), including nompA (scolopale cell) [21], α-tubulin 85E (ligament and cap cells) [22], and Sox15 (cap cell) [23].
It is striking that our analyses indicate enrichment for genes required for ciliary differentiation, because terminal Ch differentiation has not yet occurred by the embryonic stage represented at t3 (approximately Stage 12). This suggests that some aspects of differentiation require the activation of specific differentiation genes prior to overt differentiation. Unexpectedly, a proportion of ciliary genes are already expressed even at t1 (8.6% of all ciliogenesis genes (14/175), 21.4% of compartmentalised ciliogenesis genes (6/28); Table 1). At t1 the Ch precursor cells have just been specified by ato and have still to undergo two rounds of division before neuronal differentiation occurs. In situ hybridisation confirmed that mRNAs for several ciliogenesis genes are expressed in Ch precursors or in their first division products. This includes genes required for a wide range of cilia components, such as the ciliary rootlet (CG6129 – homologue of Rootletin), the IFT-B complex (CG15161 – homologue of IFT46), and the IFT-A complex (Oseg1 – homologue of IFT122; Oseg4 – homologue of WDR35) (Figure 3A–D). Most striking, for instance, is unc, which is thought to be involved in basal body maturation [14]. Although reported to be expressed only upon differentiation [14], we find that unc RNA is already 9.9-fold enriched at t1 (ranked 11th), and early expression is confirmed by in situ hybridisation (Figure 3E). Furthermore, UNC protein is also expressed early and is already localised to the centrosomes in Ch precursor cells (Figure 3F–I). Conversely, many known differentiation genes are not differentially expressed even at t3, supporting the conclusion that general differentiation has not yet occurred. This includes the Ch-specific TRPV-encoding genes, nanchung (nan) and inactive (iav), sensory neuron genes like futsch (MAP1B), and several groups of ciliogenesis gene. Therefore, a specific progression of gene expression can be discerned that defines a temporal program for organised ciliogenesis and neuronal differentiation.
A precise program of gene activation implies that transcriptional regulation is important for coordinating the cell biological events underlying ciliogenesis, yet little is known of the gene network underlying this. As a first step in exploring the transcriptional regulation of Ch genes, we characterised expression patterns of a sample of ato-correlated genes by in situ hybridisation (Table S10; sample chosen based on fold change and lack of previous detailed annotation relating to PNS expression pattern). At least 90% of genes tested (n = 43) showed expression patterns that overlap ato-expressing cells, and the vast majority of these showed expression in Ch cells (Figure S3). Moreover, most of these genes showed expression in the neuronal branch of the sensory lineage, rather than in support cells. Given the nature of the profiling (Ch cells compared with the rest of the embryo), we expected expression in ato-dependent cells, but not necessarily restricted to such cells within the nervous system. Indeed, various types of pattern were observed, including those we categorise as pan-neural (CNS and PNS), pan-sensory (PNS only), or Ch-specific. This distribution of patterns is broadly consistent with the view that the related Ch and ES lineages have both shared and unique properties. Unexpectedly, however, a significant proportion of genes show an intermediate ‘Ch-enriched’ pattern, characterised by strong and early onset expression in the Ch lineage but weak and later onset in the ES lineage (Figure S3). This includes many differentiation and ciliogenesis genes (including those mentioned above) that might otherwise have been expected to be required equally in all ciliated sensory lineages (pan-sensory). We suggest therefore that the subtype differences between the two main neuronal lineages with ciliary dendrites (Ch and ES) may partly arise from modulation in timing and level of expression of genes required for a common cellular differentiation program.
Since ato/sc proneural genes control the acquisition of Ch/ES subtype identity [24], the modulation of differentiation suggested above must ultimately result from differences in proneural gene function. In order to link the regulation of differentiation to ato function, we carried out profiling of ato-expressing cells from ato mutant embryos at t1. In such embryos, atoGFP-expressing cells largely fail to become specified as Ch precursors and remain as ectodermal cells. Comparison with the wildtype expression profile yields 50 genes that are ≥2-fold differentially expressed in wild-type atoGFP+ cells at t1 (compared with the GFP– cells) but not in mutant atoGFP+ cells (compared with the GFP− cells from the same embryos) (Table S11). Of these, 11 genes also show a ≥2-fold difference between the fold changes observed in wildtype and mutant embryos, which represent good candidates for downstream targets (Table S12). Three of these encode transcription factors (Rfx, cato, and fd3F). These genes were investigated as candidate intermediate regulatory factors that link proneural function to differentiation.
RFX is a well-known, highly conserved regulator of ciliogenesis and is best known as a proven or predicted regulator of many ciliogenesis genes through binding to an X-box motif (notably those genes associated with IFT-B) [20],[25]. Although required for neuronal differentiation, the Rfx gene is already highly expressed in the earliest atoGFP cells (9.76-fold enriched at t1, ranked 12th), indicating that it may be responsible for early expression onset of a subset of differentiation genes. Consistent with this, a resampling analysis demonstrates that gene lists for all three time-points are highly significantly enriched for the presence of nearby X box motifs (Figure S4), indicating the likely presence of Rfx target genes. In addition, of the set of 83 genes in the genome that have a conserved perfect X box motif nearby [20], 21.7% are expressed at t3—a 7.1-fold greater frequency than expected by chance (p = 8.23×10−10) (Tables 1, S9). These include ciliogenesis genes for which experimental evidence has been obtained that they are direct Rfx targets (such as CG15161, btv, tectonic, CG6129, CG4525) [20].
Although Rfx is required for both Ch and ES neurons, examination of its expression pattern revealed that, like many of its target genes, it shows a Ch-enriched pattern of expression (Figure 4A–C). It is possible, therefore, that variations in Rfx expression may underlie different subtype-specific programs in Ch and ES cells. In turn, this suggests that Rfx may be regulated differently by ATO and SC proteins in these lineages as part of their neuronal subtype-determining function. Therefore, we examined the regulation of Rfx by proneural factors. Embryonic expression analysis confirmed that Ch expression of Rfx overlaps with that of ato (Figure 4D). In contrast, Rfx expression in ES lineages begins later, only after the termination of sc expression (Figure 4E). By reporter gene analysis, we found that Rfx is regulated through separable Ch and ES enhancers (Figure 4F). The Ch enhancer is activated early in Ch development (RfxA: Figure 4G). This enhancer contains an E box motif whose sequence conforms to that previously shown to respond specifically to ATO activation (EATO) [7]. This motif binds ATO in vitro (Figure S5), and when it is mutated, the early phase of expression in Ch cells is abolished (Figure 4H). Conversely, this enhancer is ectopically activated when ato is misexpressed in the ectoderm (Figure 4I,J), but this ectopic activation is abolished when the E box motif is mutated (unpublished data). In contrast to direct activation by ato, the ES enhancer is active only after sc expression is switched off (RfxB: Figure 4K,L), suggesting that sc only indirectly activates Rfx in ES development. However, we note that the ES enhancer does contain two motifs conforming to the known SC binding site (GCAGSTG) and so it is possible that SC directly primes the Rfx gene for later expression in ES lineages. Overall, the evidence suggests that Rfx is a direct target of ato but not of sc, supporting the hypothesis that differences in Rfx regulation may be one means by which proneural factors regulate neuronal subtype characteristics.
Interestingly, the ato-related bHLH gene, cato, has a Ch-enriched expression pattern like Rfx [26]. Enhancer analysis revealed that cato too has separable Ch and ES enhancers [27]. The former contains an EATO site that is required for Ch expression, and it is ectopically activated upon misexpression of ato. Mutant analysis of cato reveals roles in cell cycle control and SOP fate maintenance but not in terminal differentiation [27]. Nevertheless, the similar regulation of Rfx and cato suggests that differential regulation of shared intermediate regulatory genes in different neuronal subtype lineages may be a common theme underlying subtype specification by ato and sc.
The gene for the predicted Forkhead family transcription factor, fd3F (CG12632), is highly enriched in atoGFP cells (19.7-fold at t3; ranked 3rd). In contrast to Rfx, fd3F is expressed exclusively in Ch neurons from the precursor stage through to differentiation (Figure 5A–C), suggesting a specific role in Ch neuron specialisation. Its highly specific Ch expression pattern suggests that fd3F may be a direct target of ato. Reporter gene analysis identified an intronic Ch enhancer of fd3F that contains three ato-type E box motifs (Figures 5D–F, 6I). However, reporter expression does not appear strongly altered when these sites are mutated (unpublished data), suggesting that regulation may occur via other E box motifs. At present, therefore, although fd3F is a target of ato, we cannot conclude whether regulation is direct or indirect.
To ascertain fd3F's function, we generated a mutation by P-element imprecise excision (FGN, in prep.). Mutant larvae and adult flies exhibit locomotion defects similar to those manifested in ato mutants (Figure 5G,H; FGN and APJ, in prep.) [4],[28]. Given the expression pattern of fd3F, these defects can be attributed to defective Ch neurons, which are required for proprioceptive feedback during locomotion. In ato mutants, such defective behaviour results from loss of Ch neurons. Immunohistochemical analysis suggests, however, that Ch neurons are mostly specified normally in fd3F mutants and little gross structural defect was observed in the neurons (Figure 5I,J; FGN, in prep.). Consistent with this, preliminary analysis of gene expression suggests that most ciliogenesis genes tested are not affected in fd3F mutants (FGN and APJ, in prep.).
We hypothesized, therefore, that fd3F regulates specialised aspects of Ch neuronal or ciliary physiology. The transient receptor potential (TRP) family of Ca2+ channels are particularly associated with sensory functions in a range of ciliary contexts [29]. In Drosophila, nan and iav encode subunits of a TRPV channel that are uniquely expressed in Ch neurons [17],[18]. The proteins are located in the Ch ciliary dendrite, where they are required for sensory transduction. We find that the expression of both nan and iav is strongly reduced in fd3F mutant embryos (Figure 5K–N, and unpublished data). Failure in regulation of nan and iav can therefore account for the defective Ch neuron function of fd3F mutants. In conclusion, ato directly or indirectly activates a transcriptional regulator concerned with Ch neuron physiology (specifically, Ch ciliary dendrite physiological specialisation).
Whilst many early expressed differentiation genes are known or predicted Rfx targets, not all Ch-specific or Ch-enriched genes (nor ciliogenesis genes) have nearby X box motifs, suggesting that other intermediate regulatory factors remain to be discovered. Another possibility is that some early expressed differentiation genes may be directly regulated by proneural factors. Such genes include CG1625 and unc, whose expression depends strongly on ato function (Table S9). Our analysis (LM and APJ, in prep.) shows that CG1625, which we name dilatory (dila), encodes a coiled-coil protein that localises to the basal body, and dila mutants exhibit defects in ciliary axonemal assembly. Together, these suggest that dila is a not a transcriptional regulator, but instead has a direct function in ciliary dendrite formation. Here, we examined the regulation of dila. The gene is highly expressed in early Ch cells (11-fold enriched at t1; ranked 10th), and dila RNA exhibits a Ch-enriched gene expression pattern in embryos (Figure 6A–C). However, it has no X box motif within 2 kb of its transcription start site. Its early expression raises the possibility that dila is directly regulated by ato. In vivo reporter gene analysis led to the identification of an enhancer required for dila expression in Ch cells (Figure 6D,E). Conversely, the reporter gene is misexpressed when ato is ectopically activated in the ectoderm (Figure 6G,H). This enhancer contains two sequences resembling EATO motifs, both of which bind ATO/DA in vitro (Figures 6I, S5). Mutation of these two motifs within this enhancer results in loss of early expression in Ch SOPs (Figure 6D,F) and loss of misexpression in response to ectopically activated ato (unpublished data). These data are consistent with direct regulation of dila by ato via one or both of these EATO motifs. We note that in a recent study of potential ato target genes in retinal development, similar evidence was presented to suggest that dila (as CG1625) is regulated by ato via these two motifs [30]. In conclusion, dila represents a differentiation gene that is directly controlled by a proneural factor, despite the gap between proneural factor expression and terminal differentiation.
Numerous genetic and misexpression analyses in a range of organisms have shown that proneural factors influence a neuron's ultimate phenotype (including its subtype identity) at an early stage in its development [1]. However, the nature of this influence on the cell biological processes of neuronal differentiation has remained obscure. This study bridges the gap between early specification by the proneural factor, ato, and the differentiation of Ch neurons. The current model in both Drosophila and vertebrates is that proneural factors activate two types of target gene during neural precursor specification: a common target set for shared neuronal properties and a unique target set for subtype-specific properties [31]. Our data suggest that such neuronal subtype differences are ultimately controlled by proneural factors in several ways: by the differential regulation of both specific and common intermediate transcription factors, which in turn regulate genes for aspects of neuronal structural and functional differentiation, and by direct regulation of potential differentiation genes (Figure 7).
The proneural factors ato and sc commit cells to similar but distinct neural precursor fates: Ch and ES neurons are evolutionarily related cell types with similar but distinct structural and physiological properties. Notably, both are characterised by the possession of specialised ciliary-based dendrites [10]–[12]. Thus, ciliogenesis is a key pathway that must ultimately be activated in sensory neurons subsequent to proneural factor function. However, there are important differences between the dendrites of Ch and ES neurons. Ch dendrites have a more prototypically organised axonemal structure and possess a characteristic ciliary dilation—a specialisation that separates the Ch ciliary dendrite into functionally distinct zones [32]. Moreover, there is evidence for an active ‘beat’ of Ch cilia during sensory transduction [33]. In general, ES dendrites appear reduced in structure: although a basal body and short axoneme are present, the tip of the dendrite consists of a ‘tubular body’ of irregularly packed microtubules [10]. Thus the basic ciliogenesis pathway must be modulated differently in Ch and ES differentiation, and ultimately this must reflect a difference in function between ato and sc proneural factors. The ciliogenic regulator Rfx is expressed and required for both ES and Ch lineages, but it is more strongly and more persistently expressed in Ch lineages (the Ch-enriched pattern). This modulation of Rfx expression is at least partly due to differences in its regulation by proneural factors, since it appears to be a direct target of ato but not sc. We hypothesise that differences in Rfx regulation by the proneural factors lead to differences in implementation of a core cilia biogenesis program, thereby directly linking early proneural factor function with key differences of neuronal morphology. Consistent with this idea, our data show that several known or predicted ciliogenesis genes also exhibit this Ch-enriched pattern, and some of these are predicted or known Rfx targets [20].
In this view, the subtype differences between Ch and ES neurons are partly produced by quantitative differences in timing or level of expression of a common differentiation process, which ultimately depends on a qualitative difference in Rfx regulation by the proneural factors. A possible example of this is CG6129. This gene is a predicted Rfx target gene and is expressed in a Ch-enriched pattern (Figure S3) [20]. The homologous mouse protein (Rootletin) localises to the ciliary rootlet and is required for its formation [34]. Thus Ch-enriched expression of CG6129 explains the presence of the ciliary rootlet in Ch neurons but not ES neurons [11],[12]. One prediction of this hypothesis is that overexpression of Rfx in ES neurons will upregulate Ch-enriched genes, and this is borne out by preliminary experiments that show an increase in CG6129 expression in ES neurons upon Rfx overexpression (scaGal4/UAS-Rfx embryos; LM and APJ, unpublished data). It is notable that differences in IFT activity are proposed to underlie differences in ciliary morphology [35] while RFX class factors have been associated with regulating genes for IFT in a variety of organisms [13]. Our work suggests that variations in Rfx expression level and timing should be explored as a possible factor in cilium diversity.
fd3F fits the more conventional view of a proneural target gene that implements a subtype-specific program of differentiation [31]. It is expressed downstream of ato uniquely in Ch neurons and regulates genes required for functional specialisation of the Ch ciliary dendrite. It is likely that Forkhead factors regulate specialisation of ciliogenesis in other organisms. In C. elegans, FKH-2 is expressed widely early in development but is also required specifically for ciliary specialisation of one type of sensory neuron [36]. Foxj1 in mice, Xenopus, and zebrafish appears to be required for the motile cilia of the lung airway and embryonic node, but not for primary cilia [37]–[39]. It remains to be determined whether fd3F regulates the machinery for the active beat that occurs in Ch dendrites as part of sensory transduction [33]. Together, our studies of Rfx and fd3F extend the previously limited knowledge of the gene regulatory network underlying ciliogenesis [13] and provide insight into how the core program may be modified to produce the highly specialised and diverse morphologies that cilia adopt for different functions [36].
Previous to this study, little was known about how ato/sc proneural genes control the acquisition of Ch/ES subtype identity, except that regulation of the Cut homeodomain transcription factor is involved. Mutant and misexpression analyses show that cut is a fate selector switch for ES identity downstream of sc [19],[40], but nothing is known of its mode of action or targets. Whereas Rfx and fd3F functions are likely to be confined to neuronal morphology, cut affects the identity of support cells too [41]. As a fate switch in the entire lineage, it appears likely that cut is involved in high-level fate specification (like proneural genes) rather than regulating aspects of differentiation directly. However, it is also possible that cut may repress ciliogenesis genes in ES neurons, either directly or by repressing Rfx expression. It will be important to integrate cut into the Ch/ES gene regulatory network in the future.
In our temporal expression profiling data, there is a steady increase in the number of known or suspected differentiation genes expressed in developing Ch cells. Many more are not expressed until after our analysis ends. Ciliogenesis is a highly intricate cellular process requiring the coordination of perhaps hundreds of genes [13],[42] and differences in expression onset may indicate prerequisite steps in the process of differentiation and ciliogenesis. A surprising observation was the significant number of ciliogenesis and differentiation genes that are expressed even at the earliest profiling time point. This is unexpected, since the earliest time point is predicted to be not only before differentiation but also even before cell divisions have generated the neurons. We suggest that further analysis of expression timing may lead to insights into the cell biology of ciliogenesis. The early activation of differentiation genes may reflect the rapid pace of development in the Drosophila embryo. Thus, early expression of ciliogenesis genes may provide components that prime cells for rapid cilium assembly later once differentiation has been triggered. Along these lines, our findings mirror striking observations of retinal ganglion cells, whose rapid differentiation within 15 minutes of the exit from mitosis has been taken to imply that genes required in postmitotic cells must be transcribed before cell division [43],[44]. A more intriguing possibility is that early expression reflects an orderly time course for ciliogenesis that begins many hours before the final cell division. For example, unc is thought to be required for the conversion of the mitotic centriole to ciliogenic basal body [14], but we found that the mRNA and fusion protein are expressed even in SOPs, several cell divisions before terminal differentiation. Interestingly, in mammals newly replicated centrioles mature over two cell cycles [45]. It is conceivable that the sensory neuron basal body might similarly need time to mature.
Since Rfx and some ciliogenesis genes are expressed in SOPs, what prevents ciliogenesis from being activated in the non-neuronal support cells? One possibility would be an extension of model recently proposed for the generation of support cell differences, in which Notch signalling between daughter cells confines the function of genes to one branch of the lineage [23]. This would predict that ciliogenesis genes and/or Rfx are Notch target genes. Another possibility is that some of the gene products are asymmetrically segregated. Thirdly, ciliogenesis may not be triggered until one or more key gene products are produced in the neuronal cell.
As a corollary, it will be important to explore further the gene regulatory network underlying the temporal and cell-type differences in ciliogenesis genes. Some early expressed differentiation genes are known or predicted Rfx targets [20]. This gives a rationale for the early regulation of Rfx by ato in Ch lineages. However, in both C. elegans and D. melanogaster, Rfx regulates only a subset of ciliogenesis genes (notably, it does not regulate IFT-A genes) [20]. Further studies on ato target genes and the ciliogenesis regulatory network in sensory neurons will identify other important regulators (Figure 7). It remains to be determined how many differentiation genes are, like dila, direct targets of ato. Interestingly, vertebrate proneural factors are hypothesised to regulate directly the transition from cycling neural progenitor (or neural stem cell) to postmitotic differentiating neuron. Perhaps ato has retained some part of an ancestral proneural factor function in direct regulation of terminal differentiation despite the subsequent evolution of SOPs that must undergo several divisions before differentiating.
In order to label ato-expressing cells, a 2.6-kb fragment upstream of the ato gene was used to drive GFP expression in transgenic Drosophila embryos. After amplification from genomic DNA (Table S10 for primers), this fragment was cloned into pHStinger [46]. The plasmid was used to make transgenic fly lines by microinjection. One viable line, atoGFP.7, with high expression levels and lacking detectable ectopic GFP expression, was chosen for embryo dissociation and cell sorting. For expression profiling of ato mutant cells, atoGFP.7 was introduced into the ato1 mutant background (a presumed null [4]). To minimise genetic background differences, the atoGFP.7; ato1 line was backcrossed four times to the original atoGFP.7 stock. The two lines are therefore predicted to be approximately 97% isogenic.
In brief, dechorionated atoGFP embryos were dissociated in Shields and Sang (S2) medium (Sigma) with 5% fetal bovine serum (Gibco) in a Dounce homogeniser with a loose pestle. Cells were pelleted by centrifugation and resuspended in protease solution (90% trypsin-EDTA (Sigma) in phosphate buffered saline). Incubation in this solution for 7 min increased the proportion of viable single cells as judged by Trypan Blue exclusion. Cells were subsequently washed twice in S2 medium. Cell suspensions were separated using a DakoCytomation MoFlo MLS flow cytometer. In each run, 3×105 atoGFP+ and 1×106 atoGFP− cells were collected. Cells were sorted into Schneider medium on ice, then pelleted and homogenised in RNA extraction buffer, and then snap frozen in liquid nitrogen. In all experiments the cell suspension was kept on ice from the time of trypsin treatment until the RNA was extracted from the sorted cells. Quantitation of RNA was carried out using QuantiTect SYBR Green RT-PCR kit (Qiagen) and a MJ Research Opticon thermal cycler. rpL32 was used as a control.
Using standard techniques recommended by Affymetrix (http://www.affymetrix.com/support/technical/manual/expression_manual.affx), RNA from sorted atoGFP+ and atoGFP− cells was used to probe Affymetrix Drosophila 2.0 microarray chips in quadruplicate using independent samples. ∼0.5 µg of RNA was converted to cDNA and amplified as cRNA using the 2-cycle protocol, before being biotin labelled and fragmented. The hybridisations were conducted at the Sir Henry Wellcome Functional Genomics Facility, Glasgow, UK. Quality control and normalisation of microarray expression data was performed using the Bioconductor package AffyPLM [47] using the standard RMA method with quantile normalisation. Differentially expressed genes between atoGFP+ and atoGFP− samples were identified using the Bioconductor package limma [48]. Lists of Affymetrix probe-set accessions were extracted from the analysis with the cut-off at a 1% FDR [49]. Affymetrix probe-sets were mapped to genomic locations using the Ensembl database PerlAPI [50],[51] and only those probe-sets that were not promiscuous (not mapping to more than one gene) with ≥50% of their oligomers were considered reliable and used to retrieve stable accessions of ‘trusted genes’.
Protein domain annotations for Pfam, Prosite, Superfamily, and Smart databases were retrieved from Ensembl for all trusted genes in our analyses (Ensembl v53 March 2009, Flybase Release FB2008_10 Dmel Release 5.13, Nov. 2008). The resulting data were parsed into genomic frequency tables for each domain from each source. To determine whether any domains were over-represented in our gene lists, we applied a corrected Fisher exact test [52] to the relative domain frequencies between list and genome. All domains that were over-represented with p≤0.05 were taken forward for further analysis.
Standard methods of whole embryo immunohistochemistry were used. Antibodies used were: anti-Ato 1∶2000 [4], MAb22C10 1∶100, MAb21A6 1∶500, anti-GFP 1∶500 (Molecular Probes), and anti-Pericentrin (1∶500, kindly provided by J. Raff). Secondary antibodies were from Molecular Probes. mRNA in situ hybridisation to whole embryos were by standard methods. Primers for antisense RNA probes used are given in Table S13. For double RNA/protein labelling, the in situ hybridisation was conducted first followed by protein detection. For wild-type embryos, we used the w1118 stock. The fly stock for the uncGFP fusion gene/protein was kindly provided by Maurice Kernan.
Fragments were amplified from genomic DNA and cloned into pHStinger. Primers used are given in Table S13. Transformants were made by microinjection into syncytial blastoderm embryos. In general, at least two independent transformant lines were tested for each construct. For E box site directed mutagenesis, we used the Stratagene Quickchange 2 kit. In each case, CANNTG was altered to AANNTT.
In vitro DNA binding assays were performed exactly as previously described using bacterially expressed ATO and DA proteins [7]. DNA probes used are shown in Table S13.
A deletion allele, fd3F1, was isolated by imprecise excision after P element mobilisation in the line, P{EP}EP1198. This deletes the 3′ end of the transcription unit and appears to be an RNA and protein null (FGN, manuscript in preparation).
Wandering third instar larvae were placed individually on the centre of a layer of 1% agarose in a Petri dish. Larval movement was traced over a period of 2 min. Path lengths were obtained from traces using NIH ImageJ. Larvae tested were from the stocks, ato1, fd3F1, and w1118 (wild type).
All microarray data from the experiments described are available from the NCBI's GEO database with accession number GSE21520.
|
10.1371/journal.pgen.1001133 | Identification of Early Requirements for Preplacodal Ectoderm and Sensory Organ Development | Preplacodal ectoderm arises near the end of gastrulation as a narrow band of cells surrounding the anterior neural plate. This domain later resolves into discrete cranial placodes that, together with neural crest, produce paired sensory structures of the head. Unlike the better-characterized neural crest, little is known about early regulation of preplacodal development. Classical models of ectodermal patterning posit that preplacodal identity is specified by readout of a discrete level of Bmp signaling along a DV gradient. More recent studies indicate that Bmp-antagonists are critical for promoting preplacodal development. However, it is unclear whether Bmp-antagonists establish the proper level of Bmp signaling within a morphogen gradient or, alternatively, block Bmp altogether. To begin addressing these issues, we treated zebrafish embryos with a pharmacological inhibitor of Bmp, sometimes combined with heat shock-induction of Chordin and dominant-negative Bmp receptor, to fully block Bmp signaling at various developmental stages. We find that preplacodal development occurs in two phases with opposing Bmp requirements. Initially, Bmp is required before gastrulation to co-induce four transcription factors, Tfap2a, Tfap2c, Foxi1, and Gata3, which establish preplacodal competence throughout the nonneural ectoderm. Subsequently, Bmp must be fully blocked in late gastrulation by dorsally expressed Bmp-antagonists, together with dorsally expressed Fgf and Pdgf, to specify preplacodal identity within competent cells abutting the neural plate. Localized ventral misexpression of Fgf8 and Chordin can activate ectopic preplacodal development anywhere within the zone of competence, whereas dorsal misexpression of one or more competence factors can activate ectopic preplacodal development in the neural plate. Conversely, morpholino-knockdown of competence factors specifically ablates preplacodal development. Our work supports a relatively simple two-step model that traces regulation of preplacodal development to late blastula stage, resolves two distinct phases of Bmp dependence, and identifies the main factors required for preplacodal competence and specification.
| Cranial placodes, which produce sensory structures in the head, arise from a contiguous band of preplacodal ectoderm surrounding the anterior neural plate during gastrulation. Little is known about early regulation of preplacodal ectoderm, but modulation of signaling through Bone Morphogenetic Protein (Bmp) is clearly involved. Recent studies show that dorsally expressed Bmp-antagonists help establish preplacodal ectoderm, but it is not clear whether antagonists titrate Bmp to a discrete low level that actively induces preplacodal fate or, alternatively, whether Bmp must be fully blocked to permit preplacodal development. We show that in zebrafish preplacodal development occurs in distinct phases with differing Bmp requirements. Initially, Bmp is required before gastrulation to render all ventral ectoderm competent to form preplacodal tissue. We further show that four transcription factors, Foxi1, Gata3, Tfap2a, and Tfap2c, specifically mediate preplacodal competence. Once induced, these factors no longer require Bmp. Thereafter, Bmp must be fully blocked by dorsally expressed Bmp-antagonists to permit preplacodal development. In addition, dorsally expressed Fgf and/or Pdgf are also required, activating preplacodal development in competent cells abutting the neural plate. Thus, we have resolved the role of Bmp and traced the regulation of preplacodal development to pre-gastrula stage.
| Cranial placodes provide major contributions to the paired sensory organs of the head. Examples include the anterior pituitary, the lens of the eye, the olfactory epithelium, the inner ear, and clusters of sensory neurons in the trigeminal and epibranchial ganglia [1]–[4]. Though diverse in fate, all placodes are thought to arise from a zone of pluripotent progenitors termed the preplacodal ectoderm. Preplacodal cells arise from the nonneural ectoderm immediately adjacent to neural crest. Neural crest cells originate in the lateral edges of the neural plate and later migrate to placodal regions to contribute to the corresponding sensory structures [1], [2]. However, while neural crest has been analyzed extensively, little is known about the early requirements for preplacodal development. Various preplacodal markers, including members of the eya, six and dlx gene families, are expressed at high levels along the neural-nonneural interface around the anterior neural plate near the end of gastrulation [1]–[7]. How these genes are regulated is still unclear, but modulation of Bmp signaling appears to be critical. In a classical model (Fig. 1A), ectoderm is patterned during gastrulation by readout of a Bmp morphogen gradient. Such a gradient could coordinate specification of preplacodal ectoderm and neural crest in juxtaposed domains, with preplacodal ectoderm requiring slightly higher levels of Bmp than neural crest [8]–[15].
Numerous studies provide strong support for the notion that neural crest requires a specific low threshold of Bmp signaling. In zebrafish mutations or inducible transgenes that weaken overall Bmp signaling can expand neural crest throughout the ventral domain [12], [13], [15]. Similarly, development of neural crest in Xenopus is stimulated by misexpression of moderate but not high levels of Bmp-antagonists [11].
In contrast, available data are ambiguous with regard to Bmp's role in preplacodal specification. A number of Bmp-antagonists expressed near the neural-nonneural interface late in gastrulation are required for normal preplacodal development [16], [17]. Similarly, high-level misexpression of Bmp antagonists expands preplacodal gene expression partway into the nonneural ectoderm [18]–[21]. These findings have been alternately interpreted as support for either of two competing models: Some investigators have argued that Bmp-antagonists titrate Bmp signaling to a specific level appropriate for preplacodal specification, consistent with the Bmp morphogen model [18], [19] (Fig. 1A). Others counter that these misexpression conditions are likely to fully block Bmp signaling [20], [21], leading to an alternative model in which preplacodal specification requires attenuation of Bmp (Fig. 1B). These opposing models invoke fundamentally different mechanisms: In the morphogen model Bmp is a positive requirement whereas in the attenuation model Bmp is an inhibitor that must be fully blocked to permit preplacodal development. Notably, none of these studies has measured changes in the level of Bmp signaling associated with their experimental manipulations, making it impossible to distinguish between the opposing models. A similar uncertainty applies to genetic studies in zebrafish, which suggest that neither of the models in Fig. 1 is fully adequate. Mutations that strongly impair Bmp signaling eliminate preplacodal development [12], [13], revealing a definite requirement for Bmp. However, none of the mutations that impair Bmp to a lesser degree expand preplacodal fate throughout the ventral ectoderm, in sharp contrast to neural crest [12], [13]. Although these data fail to support predictions of the Bmp morphogen model for preplacodal specification, it is possible that available mutations do not expand the appropriate range of Bmp signaling required for preplacodal ectoderm, if one exists. Thus the status of Bmp signaling during preplacodal specification remains an important unresolved question.
In addition to differing requirements for Bmp, preplacodal ectoderm and neural crest appear to be specified at different times. Recent studies in chick and zebrafish suggest that neural crest is specified by the beginning of gastrulation [15], [22]. In contrast, preplacodal ectoderm appears to be specified during late gastrula or early neurula stages, as suggested by studies in chick and Xenopus [20], [21]. This difference in timing is especially relevant for the Bmp-attenuation model (Fig. 1B). Specifically, the lag in preplacodal specification allows time to reshape the Bmp gradient without jeopardizing the earlier requirement of neural crest for Bmp. There are currently no data to show when preplacodal specification occurs in zebrafish.
Other signals from dorsal tissues also appear critical for preplacodal development. In chick and Xenopus, grafting neurectoderm into more ventral regions induces expression of preplacodal markers in surrounding host tissue [20], [21], [23]. Moreover, combining misexpression of Bmp antagonists with Fgf8, a relevant dorsal signal, is sufficient to induce at least some preplacodal markers; neither Fgf8 nor Bmp-antagonism is sufficient [20], [21]. Various transcription factors have also been implicated in preplacodal development, but most appear to act after preplacodal specification to influence fates of cells in different regions of this domain [2], [3].
Here we provide the first direct evidence for a 2-step model in which Bmp is required only transiently during blastula/early gastrula stage to directly or indirectly induce ventral expression of four transcription factors, Tfap2a, Tfap2c, Gata3 and Foxi1, which establish preplacodal competence throughout the nonneural ectoderm. In this context, Bmp does not act as a morphogen because it does not distinguish between preplacodal and epidermal ectoderm within the nonneural domain. We initially focused on foxi1, gata3, tfap2a and tfap2c as potential competence factors because they show similar early expression patterns throughout the nonneural ectoderm and all have been implicated in later development of various subsets of cranial placodes [2], [3], [24]–[29]. Once expressed, preplacodal competence factors no longer require Bmp for their maintenance. Near the end of gastrulation, Bmp must be fully blocked by dorsally expressed Bmp-antagonists, which combined with Fgf, are necessary and sufficient to induce preplacodal development within the zone of competence.
To monitor early preplacodal development, we followed expression of dlx3b, eya1 and six4.1. dlx3b is the earliest marker, initially showing a low level of expression throughout the nonneural ectoderm at 8 hpf, with strong upregulation in preplacodal ectoderm and downregulation in ventral ectoderm by 9 hpf (late gastrulation) [5]. Expression of six4.1 and eya1 first appear in preplacodal ectoderm by 10 hpf (the close of gastrulation), and a low level of six4.1 is also seen in scattered mesendodermal cells in the head [6], [7]. For comparison, we also monitored the neural crest marker foxd3, which is expressed specifically in premigratory neural crest by 10 hpf [30], [31].
To assess the role of Bmp in preplacodal specification, we treated embryos at various times with dorsomorphin (DM), a pharmacological inhibitor of Bmp signaling [32]. Although we used DM at higher concentrations than previously reported [32], it did not appear to cause defects beyond the phenotypes associated with Bmp pathway mutants (see below). Thus, unintended non-specific effects of the drug, if present, are apparently mild and do not interfere with the ability to block Bmp signaling.
We initially performed a dose-response to assess the effects of DM when added at 5, 6 or 7 hpf (Table 1). As expected, embryos were increasingly dorsalized after exposure to increasing concentrations of DM, and earlier exposure caused greater dorsalization than later exposure. Exposing embryos to 50 or 100 µM DM beginning at 5 hpf mimicked strong loss of function mutations in the Bmp pathway [8], [12], [13], [33] and resulted in complete dorsalization (Table 1). In confirmation, exposure to 100 µM DM at 5 hpf eliminated phospho-Smad1/5/8 staining within 15 minutes (Fig. S1A), indicating rapid and complete cessation of Bmp signaling. Additionally, mRNA for sizzled, a feedback inhibitor of Bmp [34], decayed rapidly under these conditions, with only weak staining after 30 minutes and none after 1 hour (Fig. S1B).
Because the role of Bmp in neural crest specification has been well characterized [11]–[13], [15], we tested whether DM could affect this tissue as predicted by these previous studies. Adding 100 or 200 µM DM beginning at 4 hpf totally ablated neural crest formation (Fig. 2A and data not shown). However, adding 50 µM DM at 4 hpf led to ventral expansion of cranial neural crest to fully displace the nonneural ectoderm, similar to the effects of mutations that weaken overall Bmp signaling in zebrafish [12], [13]. These conditions are thought to create a broad plateau of low Bmp signaling appropriate for neural crest specification, providing strong support for the role of Bmp as a morphogen in specifying neural crest. Interestingly, after initially treating embryos with 50 µM DM at 4 hpf, fully blocking Bmp with a super-saturating dose of DM at 5, 6, or 7 hpf does not prevent formation of cranial neural crest, though the domain is somewhat reduced when Bmp is blocked earlier. These data are consistent with the effects of timed misexpression of Chordin [15], showing that Bmp acts very early in cranial neural crest specification and is no longer needed after late blastula/early gastrula stage.
Analysis of preplacodal markers revealed a different pattern of Bmp-dependence. First, preplacodal ectoderm (Fig. 2B) and epidermal ectoderm (not shown) are totally ablated by exposure to 50 µM DM, reflecting loss of all nonneural ectoderm. Accordingly, this treatment eliminated expression of putative preplacodal competence factors foxi1 and gata3, though tfap2a and tfap2c continue to be expressed (Fig. 2C). The latter two genes are also required in the lateral edges of the neural plate for neural crest development [29], [35]. Second, we found no dose of DM that caused expansion of preplacodal markers throughout the ventral ectoderm. Instead, exposure to 25 µM at 4 hpf yielded two distinct responses; either preplacodal markers were lost entirely or preplacodal ectoderm was shifted ventrally but was still confined to two bilateral stripes bordering the neural plate (Fig. 2B and data not shown). Thus, there does not appear to be a specific level of Bmp that can expand the preplacodal ectoderm at the expense of more ventral (epidermal) ectoderm.
To characterize the temporal requirements for Bmp, embryos were treated with 100 µM DM at different times during late blastula and early gastrula stages and subsequently analyzed for expression patterns of various ectodermal markers. As expected from the severe dorsalization caused by administering this dose at 5 hpf (Table 1), neural markers were expanded throughout the ectoderm and all nonneural markers were lost, including putative preplacodal competence factors (Fig. 3D, E, G–J). Additionally, definitive preplacodal markers dlx3b, eya1 and six4.1 were not expressed in these embryos (Fig. 3A–C). In contrast, exposure to 100 µM DM from 7 hpf resulted in only partial dorsalization (Table 1, Fig. 3D, F) and all embryos expressed nonneural markers, albeit in diminished ventral domains (Fig. 3E, G–J). Preplacodal markers dlx3b, eya1 and six4.1 were expressed on time by 10.5 hpf (Fig. 3A–C). Moreover, all placodal derivatives were produced on time in embryos treated with 100 µM DM from 7 hpf, including the anterior pituitary, olfactory, lens, trigeminal, epibranchial and otic placodes (Fig. 4B, E, H, K, N, Q, T, W) [36]–[46]. Adding 100 µM DM at 6 hpf yielded two classes of embryos, with roughly half being fully dorsalized and the rest resembling the partially dorsalized embryos obtained with 100 µM DM at 7 hpf (Fig. S2, Table 1). Adding 100 µM DM at 5.5 hpf eliminated eya1 and six4.1 expression in all embryos, though some embryos still expressed dlx3b in bilateral stripes (Fig. S2). These data indicate that embryos make a transition around 5.5–6 hpf after which Bmp is no longer required for preplacodal development. As with treatment during blastula stage, treatment with 100 µM DM during gastrulation eliminated phospho-Smad1/5/8 accumulation and sizzled expression, confirming loss of Bmp signaling [15], [34] (Fig. 3K, L). Additionally, the effects of adding 100 µM DM at 7 hpf were identical to the effects of 500 µM DM, the highest dose tested (data not shown), arguing that the block to Bmp signaling was saturated at these doses. Nevertheless, to ensure that Bmp was fully blocked, we combined addition of 100 µM DM at 7 hpf with activation of heat shock-inducible transgenes encoding Chordin and/or dominant-negative Bmp receptor [15], [45] (Fig. 3M, N). The effects on preplacodal specification and morphological development were identical to treatment with 100 µM DM alone. These data show that Bmp is not directly required after the onset of gastrulation for preplacodal specification. The data further show that Bmp signaling is required to induce expression of putative competence factors foxi1, gata3, tfap2a and tfap2c during blastula stage, but is not required to maintain them thereafter (Fig. 3G–J).
We hypothesized that foxi1, gata3, tfap2a and tfap2c encode preplacodal competence factors because they are expressed early throughout the nonneural ectoderm yet are specifically required for later development of various subsets of placodes [24]–[29]. To test the functions of these genes, we injected morpholino oligomers (MOs) to knockdown their functions. Knockdown of any one gene had no discernable effect on preplacodal gene expression (data not shown), though loss of foxi1 specifically impairs development of the otic and epibranchial placodes [27], [28]. Knockdown of both foxi1 and gata3 enhanced the otic placode deficiency (data not shown), and caused a slight reduction in expression levels of dlx3b, eya1 and six4.1 (Fig. 5A). Knockdown of both tfap2a and tfap2c caused a stronger reduction in expression levels of preplacodal markers (Fig. 5B). Co-injecting either gata3-MO or foxi1-MO with tfap2a/c-MOs further reduced preplacodal gene expression (data not shown) whereas simultaneous knockdown of foxi1, gata3, tfap2a and tfap2c (quadruple morphants) resulted in complete loss of preplacodal gene expression (Fig. 5C). Moreover, development of all cranial placodes (pituitary, olfactory, lens, trigeminal, otic and epibranchial) was severely deficient or totally ablated in all quadruple morphants examined (Fig. 4C, F, I, L, O, R, U, X). Disruption of preplacodal development in quadruple morphants did not reflect general impairment of nonneural ectoderm, as the epidermal marker p63 [46], [47] was appropriately expressed in the ventral ectoderm (Fig. 5D). Additionally, quadruple morphants did not exhibit elevated cell death, as indicated by relatively normal levels of staining with the vital dye acridine orange [48] (data not shown). These data show that foxi1, gata3, tfap2a and tfap2c are specifically required for formation of preplacodal ectoderm and all placodal derivatives, and are partially redundant in this function.
Importantly, quadruple morphants retained a neural-nonneural interface (Fig. 4R and Fig. 5D), the region normally associated with preplacodal specification. Moreover, Bmp signaling also persisted in quadruple morphants as shown by continued ventral accumulation of phospho-Smad1/5/8 and expression of sizzled (Fig. 5D). Expression of fgf3, fgf8 and the Fgf-target gene erm were also appropriately localized in quadruple morphants (data not shown). Thus, neither Bmp signaling, Fgf signaling, nor neural-nonneural interactions are sufficient for preplacodal specification in this background. These data support the hypothesis that foxi1, gata3, tfap2a and tfap2c are required for preplacodal competence or early differentiation.
Although p63 is normally co-expressed with preplacodal competence factors and is only known to regulate epidermal development [46], [47], we examined whether it is required for preplacodal development. Knockdown of p63 did not detectably alter preplacodal development, nor did it enhance the deficits in preplacodal gene expression or morphological development seen in foxi1-gata3 or tfap2a/c double morphants (Fig. 5E, and data not shown). This further shows that not all early Bmp-target genes are required for preplacodal development and that the requirement for foxi1, gata3, tfap2a and tfap2c is relatively specific.
We also investigated the requirements for foxi1, gata3, tfap2a and tfap2c in neural crest formation. Knockdown of both foxi1 and gata3 did not alter expression of foxd3 (data not shown), whereas knockdown of tfap2a/c completely eliminated expression of foxd3 as reported previously [29], [35]. Not surprisingly, foxd3 expression is also ablated in foxi1-gata3-tfap2a/c-quadruple morphants (data not shown). This likely reflects a cell-autonomous requirement for tfap2a/c in neural crest specification [29], [35].
To further test the functions of preplacodal competence factors, we generated constructs to misexpress foxi1, gata3 and tfap2a under the control of the hsp70 heat shock promoter [49]. We reasoned that if these genes provide preplacodal competence, then misexpressing them in dorsal ectoderm, where preplacodal inducing factors are normally expressed, should be sufficient to induce ectopic expression of preplacodal genes. We performed transient transfections to introduce hs:tfap2a and hs:gata3 whereas a stable transgenic line was used for hs:foxi1 (see Materials & Methods). Global heat shock-activation of any one of these genes at 4.5 hpf (late blastula) or 5.5 hpf (early gastrula) resulted in scattered ectopic expression of preplacodal markers within the neural plate by 11 hpf (Fig. 6A–C, and data not shown). In most experiments, over half of embryos showed ectopic expression of preplacodal genes. Co-activation of any two heat shock genes yielded more robust and widespread expression of preplacodal genes in the neural plate, with nearly complete penetrance in most experiments. For reasons that are unclear, misexpression of competence factors at these stages caused widening of the neural plate and narrowing of the ventral Bmp signaling domain (Fig. S3). Nevertheless, Bmp signaling and general DV patterning are still evident following activation of hs:foxi1, hs:gata3 and/or hs:tfap2a (Fig. S3). Importantly, we never observed ectopic expression of the epidermal marker p63 in the neural plate following misexpression of competence factors, indicating that preplacodal competence factors do not induce all nonneural fates in this domain. Co-activation of all three transgenes at 4.5 hpf led to widespread expression of preplacodal genes, but also caused severe axial patterning defects during gastrulation, making results difficult to interpret (data not shown). However, mosaic misexpression of all three competence factors at 4.5 hpf avoided defects in axial patterning yet still led to dorsal expression of dlx3b and six4.1 in a subset of misexpressing cells (Fig. 6D). These data are consistent with the hypothesis that foxi1, gata3 and tfap2a are sufficient to render dorsal ectoderm competent to express preplacodal genes in response to dorsally expressed inducing factors.
In addition to their role in preplacodal development, Tfap2a and Tfap2c are required for neural crest [29], [35], whereas Foxi1 and Gata3 are required for preplacodal ectoderm but not neural crest. We asked whether these differing roles in neural crest could also be distinguished in misexpression experiments. Similar to the effects of injecting tfap2a mRNA [29], we found that misexpression of hs:tfap2a, either alone or in combination with other competence factors, resulted in ectopic foxd3 expression in the neural plate (Fig. S4). In contrast, activation of hs:foxi1 and/or hs:gata3 did not induce ectopic foxd3 expression (data not shown), but instead reduced expression of foxd3 in the endogenous neural crest domain (Fig. S4). Importantly, these findings show that formation of ectopic preplacodal tissue is not always associated with neural crest, further arguing that preplacodal competence can be regulated independently from other ectodermal fates.
We next attempted to induce preplacodal development throughout the zone of competence in the nonneural ectoderm by providing appropriate inductive signals normally limited to dorsal tissue. Previous studies have implicated dorsally expressed Bmp-antagonists and Fgfs as preplacodal inducers [16]–[21]. To mimic such signals throughout the nonneural ectoderm, we used heat shock-inducible transgenic lines to misexpress Fgf3 or Fgf8 (hs:fgf3 and hs:fgf8) while blocking Bmp with DM. Using standard heat shock conditions (39°C for 30 minutes) to activate hs:fgf8 combined with DM treatment at 7.5 hpf fully dorsalized the embryo and was not informative. However, full dorsalization was avoided by prolonged incubation at more moderate temperatures, achieving a weaker level of transgene activation. Incubating hs:fgf8/+ transgenic embryos at 35°C with 100 µM DM from 7.5–10.5 hpf resulted in expression of eya1 and six4.1 throughout the nonneural ectoderm in all embryos (Fig. 7B, F). Diffuse ectopic expression of erm confirmed that this heat shock regimen elevated Fgf signaling within nonneural ectoderm (Fig. 7I–K). Similar results were obtained with hs:fgf3/+ transgenic embryos incubated at 36°C with 100 µM DM from 7–10.5 hpf (Fig. 7D, H). Activation of hs:fgf3 or hs:fgf8 alone was not sufficient to activate ectopic preplacodal gene expression (Fig. 7A, C, E, G). These data show that the entire nonneural ectoderm is competent to express preplacodal genes in response to Fgf plus inhibition of Bmp.
We next titrated the dose of DM required for ectopic induction of preplacodal genes. Incubating hs:fgf8/+ embryos at 35°C with 50 µM DM at 7 hpf led to ventral expression of preplacodal genes, but lower concentrations of DM were not sufficient (Table 1). The finding that 25 µM DM is not sufficient indicates that even very low levels of Bmp signaling can block preplacodal gene activation.
To express inductive signals with greater spatial control, we generated mosaic embryos to locally co-misexpress Fgf8 and Chordin. Donor cells carrying both hs:fgf8 and hs:chd transgenes were transplanted into non-transgenic host embryos at the mid-blastula stage to obtain a random distribution of misexpressing cells. To achieve maximal transgene activation, mosaics were heat-shocked at 39°C for 30 minutes beginning at 7 hpf and then maintained at 33°C until tailbud stage (10 hpf). Of 4 mosaic embryos harboring transgenic donor cells on the ventral side, all showed significant ventral expression of six4.1 in surrounding host cells (Fig. 7N). In another experiment, transgenic donor cells were transplanted directly to the ventral side at the early gastrula stage (6 hpf). Following heat shock at 7 hpf, all mosaic embryos (n = 4) showed ectopic six4.1 expression in surrounding host cells (Fig. 7O). In contrast, no ectopic six4.1 expression was seen following mosaic misexpression of hs:fgf8 alone (n = 13) or hs:chd alone (n = 10) (Fig. 7L, 7M). This confirms that both Fgf and Bmp-antagonists are required to induce expression of preplacodal genes.
Because preplacodal specification has been reported to occur near the end of gastrulation in frog and chick embryos [20], [21], we tested whether activation of hs:fgf8; hs:chd cells at later stages could also stimulate ectopic preplacodal gene expression. Heat shock activation of ventrally transplanted transgenic cells at 8.5 hpf (yielding peak transgene expression at 9 hpf) led to robust ectopic expression of six4.1 in surrounding host ectoderm by 11 hpf (Fig. 7P). This suggests that in zebrafish, too, preplacodal specification occurs near the end of gastrulation.
Importantly, activation of hs:fgf8 and hs:chd did not lead to ectopic expression of the general neural plate marker sox19b nor the neural crest marker foxd3 (Fig. 7Q, R). Thus, induction of ectopic six4.1 expression did not result indirectly from ectopic formation of neural plate. On the other hand, activating transgenic cells at 8.5 hpf caused downregulation of p63, suggesting that nearby host cells lose epidermal identity in response to preplacodal specifying signals.
Finally, we reassessed the requirement for Fgf during normal preplacodal specification. Previous studies have reported that expression of preplacodal markers does not require Fgf in zebrafish [50]–[53]. We find that blocking Fgf by adding the pharmacological inhibitor SU5402 at 8.5 hpf did not block expression of preplacodal markers, but levels of expression were reduced (Fig. S5). We speculated that Pdgf, which is also dorsally expressed near the end of gastrulation [54] and activates a similar signal transduction pathway, might provide redundancy with Fgf. We tested this by applying another inhibitor, AG1295, which blocks Pdgf activity in zebrafish [55]. Treatment with AG1295 alone had little effect on preplacodal gene expression, but co-incubation with AG1295 and SU5402 from 8.5 hpf led to further reduction of preplacodal gene expression (Fig. S5). Indeed, expression of eya1 was almost totally eliminated in the preplacodal domain, though robust expression continues in the cranial mesoderm. These data support the hypothesis that Fgf and Pdgf are partially redundant dorsal factors required for preplacodal specification.
We have presented data supporting a relatively simple two-step model of preplacodal development (Fig. 8). First, during late blastula/early gastrula stage Bmp establishes a broad zone of preplacodal competence throughout the nonneural ectoderm. Second, near the end of gastrulation signals from dorsal tissue locally specify preplacodal ectoderm bordering the anterior neural plate. Interestingly, Nguyen et al. proposed a broadly similar two-step model based on analysis of Bmp-pathway mutants in zebrafish [12]. However, at that time neither the molecular basis of preplacodal competence nor the signals required for preplacodal specification were known. Additionally, more recent studies have led to disagreement as to whether Bmp is required at a specific low level or must be blocked entirely for preplacodal specification [18]–[21]. Our model resolves the role of Bmp, confirms that Fgf plus Bmp-antagonists are sufficient for preplacodal specification, shows for the first time that Fgf and Pdgf cooperate as redundant preplacodal inducing factors, and highlights the importance of Foxi1, Gata3, Tfap2a and Tfap2c as preplacodal competence factors. We also readdress mechanisms of neural crest specification, which show a number of crucial differences from preplacodal ectoderm.
Using DM to finely control Bmp signaling, we show that Bmp regulates neural crest and preplacodal ectoderm by markedly different mechanisms. In agreement with earlier genetic studies in zebrafish [12], [13], [15], our data indicate that neural crest is specified by a discrete low level of Bmp signaling as predicted by the classical morphogen model (Fig. 1A). Adding DM at 4 hpf at a dose sufficient to fully block Bmp signaling ablates neural crest formation, whereas a slightly lower dose causes a dramatic ventrolateral expansion of neural crest to fully displace nonneural ectoderm (Fig. 2A). Fully blocking Bmp after the onset of gastrulation does not block neural crest, in agreement with studies involving timed misexpression of Chordin [15]. These data suggest that cranial neural crest is already specified by early gastrula stage, after which it no longer requires Bmp. In chick, too, neural crest is specified by early gastrula stage [22].
Preplacodal ectoderm, marked by expression of dlx3b, eya1 and six4.1, develops in two distinct phases with distinct signaling requirements, neither of which resemble the pattern shown by neural crest. Preplacodal ectoderm requires a robust Bmp signal during late blastula/early gastrula, but unlike neural crest, there does not appear to be a specific range of Bmp signaling that uniquely specifies preplacodal fate. We found no dose of DM that could expand the preplacodal ectoderm in a manner similar to neural crest. Instead, increasing the concentration of DM (lowering Bmp signaling) either shifted discrete bilateral stripes of preplacodal ectoderm to a more ventral position or eliminated them altogether, depending on the degree of neural plate expansion. Indeed, treatment with a single dose (25µM DM beginning at 4 hpf) yielded both classes of embryo, with nothing in between. Thus, DM cannot expand preplacodal ectoderm at the expense of epidermal ectoderm, indicating that changing Bmp levels do not distinguish between these fates.
The requirement for Bmp changes during the second phase of preplacodal development beginning soon after the onset of gastrulation. Adding a full blocking dose of DM at 7 hpf does not block preplacodal specification, even if transgenic Chordin and dominant-negative Bmp receptor are also activated during this period. Thus, Bmp is not required during gastrulation for preplacodal specification. By extension, the requirement of preplacodal ectoderm for locally secreted Bmp-antagonists [16]–[21] cannot reflect a requirement for a specific low threshold of Bmp; instead Bmp-antagonists are presumably needed to fully attenuate Bmp. This conclusion is further supported by our experiments showing that a full blocking dose of DM is required to induce ectopic preplacodal markers throughout the ventral ectoderm (Fig. 7, Table 1, and see below).
We have found that Fgf combined with Bmp attenuation is sufficient to induce preplacodal markers in ventral ectoderm, as has been shown in chick and frog [20], [21], suggesting that this mechanism is broadly conserved. Thus, using heat shock-inducible transgenes, we show that misexpression of Fgf combined with DM treatment is sufficient to induce ectopic preplacodal markers anywhere within the nonneural ectoderm. This supports two important conclusions. First, it demonstrates that the entire nonneural ectoderm is competent to form preplacodal ectoderm, even at the ventral midline far from the neural plate. This is consistent with the expression domains of preplacodal competence factors (see below). Second, although Fgf and Bmp-antagonists likely constitute a small subset of signals associated with the neural-nonneural border, no other signals are needed to trigger preplacodal development. Fgf and Bmp-attenuation induces ectopic expression of preplacodal markers in chick and Xenopus [20], [21], though this combination of signals also induces expression of general neural plate markers in those species. By contrast, our experimental conditions do not induce formation of ectopic neural plate or neural crest, tissues that could themselves have induced ectopic preplacodal markers [20], [21], [23]. Thus induction of ectopic preplacodal ectoderm appears to be a direct and specific response to Fgf combined with Bmp attenuation, at least in zebrafish.
In addition to being able to induce ectopic preplacodal markers, we have found that Fgf is required in zebrafish for normal preplacodal development, and furthermore that Pdgf acts partially redundantly in this process. Fgf and Pdgf have been shown to regulate distinct aspects of gastrulation, with Fgf promoting dorsal fate specification and Pdgf promoting convergence towards the dorsal midline [55], [56]. Although Fgf is not absolutely required for expression of general preplacodal markers [50]–[53], we find that treating embryos with the Fgf inhibitor SU5402 during the latter half of gastrulation reduces the level of expression of preplacodal markers. Treating embryos with the Pdgf inhibitor AG1295 alone has no effect on preplacodal specification, but blocking both Fgf and Pdgf further reduces preplacodal gene expression, nearly eliminating eya1 expression. Homologs of Fgf and Pdgf are preferentially expressed in dorsal tissues near the end of gastrulation [54], [56], [57] and likely activate the same signal transduction pathways required for preplacodal specification. It is not known whether Pdgf regulates preplacodal development in other species, but Pdgf and Fgf are specifically required for induction of the trigeminal placode in chick [58].
In this study we have not addressed the role of Wnt inhibitors, which are also required for preplacodal development [18], [21]. Numerous Wnt inhibitors are abundantly expressed in the head and are vital for cranial development in general, including preplacodal ectoderm. Otherwise, preplacodal fate is restricted from the trunk and tail by posteriorizing Wnt signals [59], [60].
We show that Tfap2a, Tfap2c, Foxi1 and Gata3 act as partially redundant competence factors required specifically for preplacodal development. These genes are expressed uniformly within the nonneural ectoderm beginning in late blastula stage. Knockdown of individual competence factors can impair development of discrete subsets of cranial placodes but formation of preplacodal ectoderm is not detectably altered [24]–[29]. In contrast, knockdown all four competence factors specifically blocks formation of preplacodal ectoderm and all placodal derivatives (Fig. 4, Fig. 5). Importantly, formation of a ventral Bmp gradient and the neural-nonneural interface still occurs. Formation of this region reflects a signaling environment that normally promotes preplacodal development yet, without the four competence factors, cells in the nonneural ectoderm cannot respond to such signals. Conversely, misexpression of one or more competence factors in the neural plate, where preplacodal inducing signals are expressed, leads to ectopic expression of preplacodal markers (Fig. 6). Although global misexpression of competence factors causes various developmental defects, localized mosaic misexpression avoids global perturbation yet still results in cell-autonomous expression of preplacodal markers in the neural plate. Thus, these genes are necessary and sufficient to render cells competent to form preplacodal ectoderm, while additional dorsal signals are required for overt specification of preplacodal fate.
Though tfap2a/c, foxi1 and gata3 are required for preplacodal ectoderm, they are neither necessary nor sufficient for epidermal fate: Expression of the epidermal marker p63 remains appropriately localized following either knockdown or misexpression of preplacodal competence factors (Fig. 5, Fig. 6). Conversely, knockdown of p63 does not detectably impair preplacodal development nor enhance the effects of knocking down subsets of preplacodal competence factors (Fig. 5). The simplest interpretation is that Bmp initially co-induces epidermal and preplacodal potential throughout the nonneural ectoderm, with fate specification occurring later according to differences in local signaling.
Differential regulation of preplacodal competence factors by Bmp explains the differing Bmp-requirements of preplacodal ectoderm vs. neural crest. tfap2a, tfap2c, foxi1 and gata3 all require Bmp for ventral expression during blastula stage. Because these genes are expressed uniformly throughout the nonneural ectoderm, it is now clear why no dose of DM is capable of expanding preplacodal ectoderm at the expense of epidermal ectoderm, though both fates can be eliminated together at sufficiently high concentrations. However, tfap2a and tfap2c are expressed in a broader domain that includes the lateral edges of the neural plate where they are required for neural crest specification [29], [35]. The broader domain of expression suggests that tfap2a and tfap2c can be induced by a lower level of Bmp than foxi1 and gata3. Indeed, we identified a dose of DM that permits continued broad expression of tfap2a/c but eliminates expression of foxi1 and gata3 (Fig. 2). Thus the greater sensitivity of tfap2a/c to Bmp explains the ability of a low threshold of Bmp to expand neural crest at the expense of nonneural ectoderm. After the onset of gastrulation, expression of all four genes becomes independent of Bmp. This is an important regulatory feature because it allows maintenance of preplacodal competence as Bmp signaling is attenuated along the neural-nonneural border during preplacodal specification. Likewise, stability of tfap2a/c in the neural plate safeguards neural crest fate after Bmp signaling abates.
It is still unclear how tfap2a/c can alternately promote either neural crest or preplacodal development. We speculate that the overlap of tfap2a/c with early markers of neural plate such as sox2/3/19 favors neural crest, whereas overlap with foxi1 and gata3 in the nonneural ectoderm favors preplacodal development (Fig. 8). However, misexpression of tfap2a in the neural plate can induce both neural crest and preplacodal markers, albeit in non-overlapping clusters of cells (Fig. S4). It is possible that the level of tfap2a and tfpa2c also influences its developmental function. Both genes show diminishing expression near the edges of the neural plate, which might facilitate their neural crest functions. Similarly, cell-to-cell variation in the level of hs:tfap2a transgene expression might explain the ability to activate ectopic preplacodal and neural crest markers in dorsal ectoderm.
The long lag between expression of competence factors and expression of preplacodal markers remains unexplained. That is, why are preplacodal competence factors expressed prior to gastrulation yet preplacodal markers are not induced until the end of gastrulation? We cannot accelerate expression of preplacodal markers by changing the time of activation of hs:fgf8 and hs:chd. Regardless of whether we activated these transgenes at 7 hpf or 8.5 hpf, we only detected ectopic expression of preplacodal markers at 10.5–11 hpf, the same time these genes are induced within the endogenous preplacodal domain. It is possible that competence factors require sufficient time to “condition” ectoderm, for example through chromatin remodeling [61], or by activating other essential co-factors. These are important issues that require further investigation.
Embryos were developed under standard conditions at 28.5°C except where noted and staged according to standard protocols [62]. To block Bmp, dorsomorphin (DM) (Calbiochem, 171260) was added to the fish water from a 10mM stock in DMSO. Embryos were treated without removing their chorions. Treatment was carried out in 24-well plates, with 40 embryos in 0.5 ml of solution per well. Relevant controls were incubated in fish water containing an equal concentration of DMSO to that of treated embryos. DM solutions should be exposed to as little light as possible as the drug is photo-unstable. Stock solution of DM may be stored in small aliquots at −80°C for several months, but storage at warmer temperatures and repeated freeze-thaw significantly reduces activity. To Block Fgf, SU5402 (Calbiochem) was diluted from a 10 mM stock in DMSO. To block Pdgf, AG1295 (Calbiochem) was diluted from a 20mM stock in DMSO.
Fixation and in situ hybridization were performed as previously described [48], [57]. Immunostaining for phosphorylated Smads was carried out as described [15] with minor modifications. The primary antibody was used at a concentration 1∶150 (anti-pSmad1/5/8 antibody; Cell Signaling Technology). Secondary antibody was HRP-conjugated anti-rabbit IgG at 1∶200 (Santa Cruz Biotechnology).
For gene knockdown experiments, embryos were injected with 5ng per morpholino as indicated. Morpholino sequences for foxi1, tfap2a, tfap2c and p63 have been previously published [27], [29], [63]. To knockdown gata3, either of two morpholinos was used: For blocking translation, gata3-MO1 TCCGGACTTACTTCCATCGTTTATT; for blocking mRNA splicing at the exon1-intron1 junction, gata3-MO2 AGAACTGGTTTACTTACTGTGAGGT. Neither gata3-MO1 nor gata3-MO2 produced discernable phenotypes on their own, but both showed identical interactions with morpholinos for other competence factors. The ability of gata3-MO2 to diminish production of mature gata3 mRNA was confirmed with RT-PCR (Fig. S6). The MO-generated phenotypes described in this study were 100% penetrant, except where noted in the text. At least 10 specimens were examined or each experimental time point, unless stated otherwise.
Full length cDNAs of foxi1, gata3, tfap2a, fgf3 and fgf8 were ligated to hsp70 heat shock promoter [49] with flanking I-SceI meganuclease sites [64], [65]. Recombinant plasmid (10–40 pg/nl) was coinjected with I-SceI meganuclease (NEB, 0.5 U/µl) into 1-cell stage embryos. For transient ectopic expression, injected embryos were heat-shocked in a recirculating water bath. Stable transgenic lines Tg(hsp70:fgf8a)x17, Tg(hsp70:fgf3)x18 and Tg(hsp70:foxi1)x19 were generated by raising injected embryos to adult and screening by PCR for germline transmission. Heterozygous transgene-carriers were easily distinguished based on the phenotype following heat shock at 30% epiboly: Activation of Tg(hsp70:fgf8a)x17 or Tg(hsp70:fgf3)x18 caused dorsalization of the embryo, whereas activation of Tg(hsp70:foxi1)x19 caused anterior truncations with defects in forebrain and eyes (Fig. S3H). The Tg(hsp70l:dnBmpr-GFP) transgenic line [45] was provided by ZIRC. Tg(hsp70:chordin) [15] was generously provided by Mary Mullins.
In most experiments, transgenic embryos were heterozygous for the transgenes in question, with the exception that homozygous Tg(hsp70:chordin)/Tg(hsp70:chordin) embryos were used to misexpress chd. To misexpress foxi1, tfap2a and gata3, embryos were heat shocked at 39°C for 30 min at various times as indicated in the text. Tg(hsp70l:dnBmpr-GFP) and Tg(hsp70:chordin) embryos were heat shocked at 39°C for 30 min at 7.5 hpf; Tg(hsp70:fgf8a)and Tg(hsp70:fgf3) embryos at 35°C for 3 hr from 7.5 hpf. After heat shock, the plate containing the embryos was transferred into a 28.5°C incubator until fixation or observation.
Donor embryos were injected with lineage tracer (mix of lysine fixable rhodamine dextran, 10000 MW, and 5% biotin dextran, 10000 MW, in the ratio of 1∶9 in 0.2 M KCl) at the one-cell stage. Cells were transplanted either from blastula stage donors into blastula stage hosts or from blastula stage donors into gastrula stage (∼6 hpf) hosts. Mosaic embryos were then heat-shocked at 39°C for 30 min at 7 hpf and subsequently maintained at 33°C until fixed. Transplanted cells were identified in the hosts by streptavidin-FITC antibody staining.
Embryos were dechorinated and incubated for 1 hour on agarose-coated plates containing fish water with acridine orange (AO) (1µg/ml), as modified from [48]. The embryos were then briefly washed and immediately examined under a fluorescence microscope.
|
10.1371/journal.pntd.0007304 | The arginine sensing and transport binding sites are distinct in the human pathogen Leishmania | The intracellular protozoan parasite Leishmania donovani causes human visceral leishmaniasis. Intracellular L. donovani that proliferate inside macrophage phagolysosomes compete with the host for arginine, creating a situation that endangers parasite survival. Parasites have a sensor that upon arginine deficiency activates an Arginine Deprivation Response (ADR). L. donovani transport arginine via a high-affinity transporter (LdAAP3) that is rapidly up-regulated by ADR in intracellular amastigotes. To date, the sensor and its ligand have not been identified. Here, we show that the conserved amidino group at the distal cap of the arginine side chain is the ligand that activates ADR, in both promastigotes and intracellular amastigotes, and that arginine sensing and transport binding sites are distinct in L. donovani. Finally, upon addition of arginine and analogues to deprived cells, the amidino ligand activates rapid degradation of LdAAP3. This study provides the first identification of an intra-molecular ligand of a sensor that acts during infection.
| Leishmania donovani, the causative agent of visceral leishmaniasis, leads a digenetic life cycle as a flagellated promastigote in the vector sandfly and aflagellated amastigote within phagolysosomes of infected macrophages. Arginine is an essential amino acid for Leishmania which possesses a high specificity arginine transporter (LdAAP3), a protein that imports the amino acid into parasite cells. Arginine is primarily utilized in de novo protein synthesis and for biosynthesis of trypanothione via the polyamine pathway. It was previously reported by our group that L. donovani senses lack of arginine in the surrounding micro environment and activates a unique arginine deprivation response (ADR) pathway, thus upregulating the expression of LdAAP3 as well as other transporters. In the present study, we identified the region on the arginine molecule which is the ligand that activates ADR. We show that the conserved amidino group at the distal cap of the arginine side chain is the ligand that activates/suppresses ADR. Using arginine analogues that contain this group we observed that arginine sensing and transport are distinct in L. donovani, both in axenic promastigotes and intracellular amastigotes. Additionally, the arginine sensor responds to both arginine starvation and sufficiency.
| Leishmania donovani is a parasitic protozoan that causes visceral leishmaniasis (kala-azar) in humans that is almost always fatal [1]. This organism is an obligatory intracellular pathogen which cycles between the acidic phagolysosome of macrophages (intracellular amastigote) and the relatively alkaline mid-gut of female sand flies (extracellular promastigote) [2]. Shifting between the two distinct environments requires flexible adaptation mechanisms, and the ability to sense and rapidly respond to fluctuations in vector and host environments [3]. We use L. donovani as a model organism to investigate the molecular tools parasites have developed to overcome the harsh environments they encounter in the host and vector.
Arginine is a semi-essential amino acid for mammals and is required for protein synthesis and other metabolic pathways like the synthesis of urea, polyamines and creatinine [4]. On the other hand, arginine is an essential amino acid for Leishmania, as the parasites cannot endogenously synthesize it and rely on its uptake from host arginine pools. Additionally, arginine is the sole precursor for the production of spermidine, and subsequently trypanothione, in Leishmania [5,6]. The parasite utilizes arginine for polyamine biosynthesis as it lacks both catalases and classical seleno-containing glutathione peroxidases and thus has to rely on trypanothione for maintaining redox balance [5]. Arginine is transported within L. donovani via a high-affinity arginine transporter (LdAAP3), which is specifically localized to the membrane of the flagellum and glycosomes [7]. LdAAP3 is a highly specific transporter and only a few compounds affect/inhibit it [8]. When over-expressed, LdAAP3 also localizes on the plasma membrane [8].
During infection with L. donovani, macrophages employ defense mechanisms such as nitric oxide (NO) and reactive oxygen species (ROS) production [9]. The synthesis of NO by macrophages requires arginine. The intracellular parasites increase arginase I activity in macrophages, which restricts the amount of arginine available for NO production, and also increase the production of trypanothione, thereby neutralizing ROS and facilitating a safe niche for themselves inside the macrophages [10,11]. Therefore, Leishmania is locked in competition with the host for arginine. L. donovani modulates infected host macrophages to stimulate arginine transport by up-regulating the expression of the cationic amino acid transporter 2- CAT2B (SLC7A2), thereby ensuring its own survival [12].
We have previously reported the presence of an arginine deprivation response (ADR) in L. donovani, which monitors arginine levels in its environment. Genome-scale transcriptomics identified six transporters that up-regulate upon ADR activation in L. donovani, which include, in addition to LdAAP3, the pteridine, folate and three putative transporters [7]. Interestingly, we observed that ADR activation occurs in intracellular amastigotes, thereby supporting the notion that parasites undergo arginine deprivation during development in the phagolysosome.
Most eukaryote amino acid sensors respond to sufficiency or deficiency by activating a mTOR signaling cascade [13]. In contrast, we have previously shown that the L. donovani arginine sensor responds to environmental arginine deprivation by activating a mitogen-activated protein kinase 2-mediated signaling pathway that within minutes up-regulates the expression of the ADR genes [7]. This arginine sensor was identified for the first time and was reported to induce a response in the absence of its ligand. Furthermore, the addition of arginine to ADR-activated promastigotes induced a rapid degradation of the LdAAP3 protein to its homeostatic level. This indicated that the arginine sensor activates two distinct pathways: the ADR and an arginine sufficiency response. To date, we have not identified the L. donovani arginine sensor or its ligand.
This study identifies the regions on the arginine molecule which are essential for ADR activation and binding to the arginine transporter. We show that the conserved amidino group at the distal cap of the arginine side chain is the ligand that activates/suppresses ADR, both in axenic promastigotes and intracellular amastigotes. Using arginine analogues and additional compounds that contain this group, but lack the α amino group, we observed that arginine sensing and transport binding sites are distinct in L. donovani. Furthermore, these analogues affect arginine sensing of intra-lysosomal amastigotes. Sensor specificity is very high as any modification of the amidino group results in non-recognition of the arginine analogue by the sensor and thus has no effect on ADR. This study comprises the first in-depth analysis of arginine sensing in Leishmania.
L-arginine, D-arginine, Pentamidine, Canavanine, N-Methyl L-arginine acetate (NMLAA), Nω-Nitro-L-arginine methyl ester (L-NAME), Nω-Nitro-L-arginine (L-NNA), L-citrulline, 3-Ureidopropionic acid and 4-{[5-(4-aminophenoxy)pentyl]oxy}phenylamine were obtained from Sigma-Aldrich, USA. All other materials used in this study were of analytical grade and commercially available.
Promastigote cultures of the L. donovani Bob strain (LdBob strain/MHOM/SD/62/1SCL2D), initially obtained from Dr Stephen Beverley (Washington University, St. Louis, MO, USA), and L. donovani 1S strain, (MHOM/SD/00/1S) were used in this study. Promastigotes were cultured at 26°C in M199 medium (Sigma-Aldrich, USA), supplemented with 100 units/ml penicillin (Sigma-Aldrich, USA), 100 μg/ml streptomycin (Sigma-Aldrich, USA) and 10% heat-inactivated fetal bovine serum (FBS; Biowest).
THP-1 cells, an acute monocytic leukaemia-derived human cell line (ATCC, TIB-202TM) were used for all experiments. They were cultured in RPMI-1640 (Sigma-Aldrich, USA) medium supplemented with 10% heat-inactivated FBS, 100 units/ml penicillin and 100 μg/ml streptomycin at 37°C in a humidified atmosphere. For infection assays, 0.5 x 106 cells/ml were seeded in RPMI-1640 medium containing 10% FBS. The cells were treated with 50 ng/ml of phorbol 12-myristate 13-acetate (PMA; Sigma-Aldrich, USA) for 48 h to induce their differentiation into macrophage-like cells. Immediately before infection, the cells were washed once with phosphate buffered saline (PBS) and incubated in RPMI medium (Sigma-Aldrich, USA) containing 0.1 mM arginine (unless stated otherwise), and supplemented with 10% heat-inactivated FBS, 100 units/ml penicillin, and 100 μg/ml streptomycin. Promastigotes in the late log-phase were added to cells at a ratio of 20:1 and incubated at 37°C in a humidified atmosphere for 5 h. Extracellular parasites were removed by washing the cells five times with PBS. Thereafter, the cells were incubated in RPMI medium containing 0.1 mM arginine (unless stated otherwise) at 37°C in a humidified atmosphere for 2 h, 24 h, or 48 h.
Mid-log phase promastigotes (1×107 cells/ml) were used for arginine deprivation studies. The cells were washed with Earl’s salt solution twice and re-suspended in arginine deficient Medium 199 (Biological Industries Ltd.). Arginine deprivation was carried out at 26°C for specified time periods and was concluded by transferring the cells to ice. Arginine deprived cells were washed twice with Earl’s salt solution before being used for transport assays, Northern and Western blot analysis.
THP-1 cells were cultured in DMEM medium with 10% FBS. Magnetic beads of 3 μm size were added to flasks containing THP-1 cells. The isolation of intact macrophage phagosomes was carried out as described by Kuhnel et al. [14].
The isolated phagosomes were then resuspended in 1 ml of ice-cold phosphate-buffered saline plus 1 ml of 5 N perchloric acid, vortexed, and incubated on ice for an additional 10 min. Perchloric acid lysates were centrifuged in a microcentrifuge at 14,000 rpm for 10 min at 4 °C, and 232 μl of 5 N KOH was added to the supernatant to titrate the sample pH to 7.0. Additional centrifugation was performed under the aforementioned conditions, and 200 μl aliquots were analyzed for amino acid content by the method described by Fekkes et al. [15]. The analyses were carried out at the Medical Biochemistry Laboratory at Rambam Medical Center in Haifa.
Uptake of 25 μM [3H]arginine (600 mCi/ mmol), by mid-log phase parasites was determined by the rapid filtration technique of Mazareb et al. as reported [16]. To determine initial rates of transport, transport measurements were performed on 1 x 108 promastigotes exposed to radiolabel for up to 2 min. The amount of radiolabel associated with the cells was linear with time over the 2-min time course of the transport assay.
Total RNA from L. donovani promastigotes (either deprived for arginine or non-deprived) was prepared using the Tri-reagent protocol and subjected to Northern blotting for LdAAP3 as described before [17]. Probes were amplified using the following primers: LinJ.31.900 LdAAP3 Forward: 5'-GCTGTGACGGGGTCAGTG-3' and. Reverse: 5'-GTACGTCGCCAGCCAGTG-3'. LdBPK_101450.1 pteridine transporter Forward: 5’- ATGACCGTTGGTCAGCAGA-3’ and Reverse: 5’- GCCGTGGTGACGCCGTACT-3’.
RNA quantification, cDNA preparation, and real-time PCR were performed as discussed previously [18]. Briefly, total RNA was isolated from cells by Tri reagent (Ambion, Thermo Fischer Scientific, USA). The concentration and purity of RNA were determined by Nanodrop (Thermo Fischer, USA). Two micrograms of RNA were treated with RNase-free DNase (Promega, USA), and subsequently reversed transcribed into cDNA by using the First-strand cDNA synthesis kit (Thermo Scientific, USA), as per the manufacturers’ instructions. Real-time PCR was performed on the resulting cDNA with gene-specific primers (LdAAP3: Forward 5’ CGGTCGAAATGGTGCCAAAC 3’, Reverse 5’ GGCTTCATCTTCCCTGCGTA 3’; LdPT: Forward 5’ AGGACGCTGCTCAACTCTTC 3’, Reverse 5’ AAGGCGAACGTGTCACTCAA 3’; kinetoplast minicircle DNA (control) [19]: JW11 5’ CCTATTTTACACCAACCCCCAGT 3’; JW12 5’ GGGTAGGGGCGTTCTGCGAAA 3’) using the PowerUp SYBR Green Master Mix (Applied Biosystems, USA). The PCR amplification program used was as follows: 50°C for 2 minutes (min) and 95°C for 10 seconds (sec), followed by 40 cycles at 95°C for 15 sec, 59°C for 1 min, and 72°C for 20 sec. The amplification of the kinetoplast minicircle DNA of L. donovani was used as an internal control. The results were expressed as fold change of control (2 h infected cells) using the method described by Pfaffl, 2001 [20]. The real time PCR primers did not amplify any products in uninfected macrophages. All sample analysis was performed in triplicate, and each experiment was performed three times.
At the specified times following infection or treatment with inhibitors, THP-1 cells were washed two times with PBS. MTT [3-(4,5-dimethyl-2-thizolyl)-2,5-diphenyltetrazolium bromide] dye solution (Sigma-Aldrich, USA) (5 mg MTT in 1 ml PBS) was diluted 1:10 in RPMI medium (normal or with 0.1 mM arginine). Uninfected or infected THP-1 cells were incubated in diluted MTT dye solution at 37°C in a 5% CO2-air atmosphere for 2 h and thereafter incubated with stopping solution which consisted of isopropanol containing 5% formic acid, at 150 rpm, 37°C for 20 min. L. donovani promastigotes were incubated with diluted MTT dye solution for 3 h at 37°C, and incubated with stopping solution comprising of isopropanol and 20% SDS in a 1:1 ratio, at 80 rpm, 37°C for 30 min. Absorption was then measured at 570 nm, and the percentage cell viability was calculated.
THP-1 cells were seeded on glass coverslips (1 x 106 cells/well) in a 6-well plate and treated with 50 ng/ml of PMA (Sigma-Aldrich, USA) for 48 h. They were infected as described above, and the intracellular parasite load was visualized by Giemsa staining.
Western blot analysis of LdAAP3 was performed as described previously by Darlyuk et al., 2009 [17].
Real-time PCR data were analyzed by GraphPad prism and represented as mean ± standard error of the mean (S.E.M.). Student’s unpaired 2-tailed t-test was used to calculate significance. P value < 0.01–0.05 was considered statistically significant (*), p < 0.001–0.01 was considered very significant (**), and p < 0.0001–0.001 was considered extremely significant (***). The northern blot image data was analyzed using ImageJ analysis software. Data from three independent experiments were analyzed and represented as mean ± standard error of the mean (S.E.M.).
The first set of experiments aimed to identify the basal concentration of arginine required for the activation of the ADR in L. donovani promastigotes. The maximal concentration of arginine required for ADR activation was found to be 5 μM, which resulted in the up-regulation of LdAAP3 and threshold was lower at 0.5 μM for the pteridine transporter (LdPT) at mRNA level (Fig 1A). However, unlike LdAAP3 the expression of LdPT mRNA was not linear dose dependent. Interestingly, this concentration of arginine is close to the apparent Km value of 2.4 μM for L. donovani LdAAP3 transport activity [8], raising the question whether this transporter is also the sensor.
Previously, we observed that 48 hours after infecting THP-1 macrophages, the expression of LdAAP3 in intracellular L. donovani increased almost two-fold as compared to promastigotes [7]. This suggested that during development inside phagolysosomes, parasites encountered a low level of arginine that activated ADR. In this case, increasing external arginine concentrations might help to maintain the phagolysosome arginine concentration above the threshold. To test this, we infected THP-1 macrophages that grew in media containing 0, 0.1, 0.5 and 1.5 mM arginine. At 48 hours post-infection, total RNA was extracted from infected macrophages, and the resulting cDNA was subjected to real-time PCR, using LdAAP3 and LdPT primers as probes, as mentioned in the methods section (Fig 1B and 1C, respectively). As shown, the mRNA abundance of both genes increased as arginine concentration in the medium decreased. The results indicate that the arginine concentration in phagolysosomes of macrophages grown in a medium that contains arginine at a concentration of 0.1 mM and below activates ADR. A literature search indicated that the arginine concentration in human blood is ~80 μM [21], thereby indicating that ADR activation in our experiments was achieved under physiological conditions. Additionally, the infectivity of L. donovani in THP-1 cells cultured in media containing different concentrations of arginine was ~40% (S2A Fig).
As control, THP-1 cells, either uninfected or infected with L. donovani in medium containing different concentrations of arginine were subjected to an MTT assay to determine their viability at 48 hours. As seen in S1A Fig, the macrophages were 85–100% viable in media containing different arginine concentrations, thus indicating that ADR activation in intracellular amastigotes was not detrimental to macrophage viability. Additionally, the expression of LdAAP3 remained unchanged in L. donovani promastigotes cultured in media containing 0.1 mM and 0.5 mM arginine (S1B Fig), and promastigote viability was between 90–100% (S1C Fig). This proves that extracellular parasites did not contribute to the observed activation of ADR in intracellular amastigotes.
ADR activation in intracellular amastigotes was also determined in a time-course analysis of the infection cycle. THP-1 cells were infected with L. donovani in medium containing 0.1 mM arginine for 2 h, 24 h and 48 h post-infection and harvested at the end of each time point for RNA analysis (Fig 1D and 1E). Real-time PCR showed that the up-regulation of LdAAP3 and LdPT started at 24 h post-infection and continued to increase at 48 h post-infection, thereby implying that the activation of ADR in intracellular amastigotes occurs between 24–48 hours post-infection. This infers that the initial arginine concentration in infected phagolysosomes is high and reduces with time, reaching ADR activation level at 24 h post-infection. However, it could also be possible that induction of the ADR is delayed due to the time taken to deplete intracellular parasite pools of arginine following their phagocytosis. As seen in Fig 2B, the infectivity of L. donovani in THP-1 cells at 2 h, 24 h and 48 h post-infection was between 28–38%.
Previous studies have indicated that arginine concentration in the mammalian lysosomes is higher than that in their cytosol [22] and in Saccharomyces cerevisiae vacuoles [23]. Because lysosome volume was not determined in this study it did not provide accurate arginine concentrations. We have now determined the arginine concentration in THP-1 macrophage phagolysosomes that seems to agree with the observation of Harms et al. (Fischer-Weinberger et al, in preparation). The initial arginine concentration in THP-1 phagolysosomes is 0.14 mM, a concentration that is higher than the concentration we found that activates ADR in axenic parasites. This further supports our findings that the late activation of ADR during infection is due to the time it takes for intra-phagolysosome parasites to utilize arginine and reduce its concentration to ≤5 μM.
Arginine is a positively charged amino acid with an amidino group at the distal cap of its side chain. We hypothesized that this side chain of arginine is the ligand that binds the arginine sensor and transporter on the parasite surface. To determine this, we first analyzed various structural L-arginine analogues which were previously shown to be arginine transport inhibitors [24,25]. Canavanine has a guanidinoxy group with a conserved amidino group as in arginine, and N-methyl L-arginine acetate (NMLAA) is another structural analogue of L-arginine where the amidino group is modified by the addition of a methyl group. The methyl-amidino group of NMLAA has a net charge of zero at pH 7. NMLAA has been previously shown to be an arginine transport inhibitor in L. donovani [24] and canavanine inhibits arginine transport in S. cerevisiae [26,27] and T. brucei [25]. To test the effect of these structural analogues on ADR, we determined the minimal concentration of each structural analogue that upon two hours treatment inhibited arginine transport in promastigotes but had no effect on cell viability (Table 1). This list includes 0.5 mM canavanine, and 1 mM NMLAA (left column). We determined the effect of these compounds on ADR activation in promastigotes. This was carried out in arginine-depleted L. donovani axenic promastigotes cultured in M199 with or without arginine transport inhibitors followed by Northern blot analysis using gene-specific probes for LdAAP3 and LdPT. Fig 2A shows that Canavanine inhibited LdAAP3 (27%±1.9) and LdPT (61%±2.4) mRNA up-regulation. In contrast, NMLAA did not have any inhibitory effect on ADR as the expression of both the genes was found to be up-regulated upon arginine deprivation (Fig 2A). Identical results were obtained at the protein level for LdAAP3 in a Western blot analysis for cananvanine and NMLAA (Fig 2B). Hence, methylation of the amidino group in NMLAA retained binding to the LdAAP3 transporter but lost recognition by the sensor.
Further, the effect of the enantiomer D-arginine as compared to L-arginine on ADR was determined. D-arginine is not a competitive inhibitor of L-arginine transport [24]. This was carried out in L. donovani promastigotes in M199 medium with or without D-arginine (1 mM) and Northern blot was performed as mentioned above. Fig 3 shows that D-arginine had no effect on ADR and did not lead to the degradation of LdAAP3 and LdPT mRNA. This indicated that, the external arginine sensor in the parasite could distinguish between D-arginine and L-arginine.
We further checked what is the effect of chemical compounds that have a conserved amidinio side chain group of arginine and one such compound was pentamidine. Pentamidine is a diamidine that has two amidino moieties previously shown to be competitive arginine transport inhibitors in L. donovani [8,28]. These amindino groups have a net positive at pH7. Analysis of the effect of 100 μM pentamidine on axenic promastigotes showed that pentamidine drastically inhibited both LdAAP3 (81%±2.8) and LdPT (72%±3.2) up-regulation at 2 h post arginine deprivation (Fig 2A). This suggests that the minimal molecular group necessary for recognition by the surface arginine sensor and transporter binding sites is the amindino group of the arginine side chain. Similar result was also seen in Western blot analysis where pentamidine inhibited the ADR-stimulated expression of LdAAP3 (Fig 2B). This minimal amidino group of the arginine side chain is sufficient to inhibit ADR and arginine transport and it is direct proof that the α-carbon group is not necessary for arginine sensing as well as transport.
Further, the analogues of L-arginine which have a modified amidino/guanidine group (like NMLAA) and their effect on ADR was analyzed. Nω-Nitro-L-arginine methyl ester (L-NAME) and Nω-Nitro-L-arginine (L-NNA) are arginine analogues where the amidino group is modified by the addition of a nitro-group (nitro-amidino). The net charge of the nitroguanidinium group in L-NAME is neutral and the primary α-amino group is positively charged [29], while the gaunidino group of L-arginine has a net positive charge. Both arginine analogues have been previously shown to be weak arginine transport inhibitors in L. donovani [24]. In addition to the above mentioned arginine transport inhibitors, other compounds which have a deamidated guanidino group were also tested for their effect on ADR. L-citrulline is an alpha amino acid and an intermediate in the urea cycle. It has a carbamoylamino group which is formed by the deamination of the guanidium group, and has previously been shown by our group to be a poor arginine transport inhibitor for LdAAP3 [8]. Citrulline is known to have no net charge at pH 7 unlike arginine which has a net positive charge. Similarly, 3-Ureidopropionic acid has an carbamoylamino group similar to citrulline but has a missing primary amine. Additionally, a pentamidine analogue 4-{[5-(4-aminophenoxy)pentyl]oxy}phenylamine, which has the two amidino groups of pentamidine substituted with amines was also analyzed. All these compounds were used further to test their effect on ADR in L. donovani (see all the molecular structures in Fig 4).
There was no inhibition of LdAAP3 and LdPT levels when L. donovani promastigotes were treated with the above mentioned compounds (Fig 3). This indicates that there are two distinct arginine binding sites in the ADR machinery in L. donovani. The complete list of molecular structures of all the arginine analogues (structural and side chain analogues) is shown in Fig 4. These compounds were not tested for the regulation of LdAAP3 and LdPT at the protein level in promastigotes and at the RNA and protein levels in intracellular amastigotes as their effect at the RNA level was similar to that of NMLAA. Thus, NMLAA and other compounds with a modified guanidino/amindino group have a net charge of zero as compared to the net positive charge of the guanidino/amindino group of arginine. This suggests that in addition to the presence of a functional amidino or guanidino group, the positive charge on the R-group of arginine is also an important factor for the recognition of arginine by the L. donovani sensor.
We have previously reported that the arginine sensor responds to both arginine deprivation and sufficiency and the addition of exogenous arginine to two hours arginine starved axenic promastigotes induces rapid degradation of the LdAAP3 protein to the level observed in un-deprived cells [7]. In order to further analyze arginine sufficiency response at the mRNA level, exogenous arginine (0.45 mM) was added to two hours arginine deprived axenic promastigotes which induced the rapid degradation of LdAAP3 (rate of t1/2 = 30 minutes) and LdPT mRNA (Fig 5A). This implies that the regulation of the LdAAP3 degradation signal occurs not only at the protein level as previously suggested [7], but also at the post-transcriptional level. The minimal threshold of arginine sufficient to be detected by the arginine sensor as arginine sufficiency and thereby downregulate ADR was also checked. However, LdPT mRNA was degraded more rapidly as seen in Fig 5A and did not follow the same mRNA degradation kinetics as LdAAP3. It was observed that 10 μM exogenous arginine did not activate arginine sufficiency signaling via arginine sensor, but 50 μM, 100 μM and 450 μM of exogenous arginine were detected as arginine sufficiency, thereby resulting in the downregulation of LdAAP3 levels (Fig 5B).
In order to ascertain whether the arginine sufficiency phenomenon also occurs in intracellular amastigotes, THP-1 macrophages were infected with L. donovani for 48 h in medium containing 0.1 mM arginine, and then excess arginine (1 mM or 5 mM) was added for an additional 2 h. RNA was extracted from infected macrophages before and after the addition of excess arginine and subjected to real-time PCR. This resulted in the down-regulation of LdAAP3 and LdPT mRNA in intracellular amastigotes (Fig 6A and 6B). This indicated that intracellular amastigotes, like the axenic, respond to arginine sufficiency by rapidly down-regulating ADR. As seen in Supp. Fig 2C, the differences in infectivity of L. donovani in THP-1 cells treated ot not with exogenous arginine were not significant.
As it was observed that the arginine transport inhibitors (pentamidine and canavanine) inhibited ADR, we subsequently checked the effect of these inhibitors on arginine sufficiency. L. donovani axenic promastigotes were incubated for 2 h in M199 medium lacking arginine. Thereafter, arginine transport inhibitors (pentamidine (100μM), canavanine (500μM), NMLAA (1mM), D-arginine (1mM), L-NAME (1 mM), L-NNA(1 mM), L-citrulline (1 mM), 3-Ureidopropionic acid (1 mM) and 4-{[5-(4-aminophenoxy)pentyl]oxy}-phenylamine (100 μM)) were added to arginine deprived cells, and the cells were harvested at different time-points. Upon Northern blot analysis, it was observed that only pentamidine and canavanine down-regulated ADR as evidenced by the rapid degradation of LdAAP3 (Fig 7A) and LdPT (Fig 7B) in a time-dependent manner. However, NMLAA(Fig 7A) and the other analogues (Fig 8) had no effect on ADR.
In order to ascertain whether the same holds true in THP-1 cell-derived intracellular amastigotes, the first step was to determine the ideal concentration of pentamidine required to suppress ADR in intracellular amastigotes. As seen in S3 Fig, a dose-response analysis revealed that 100 μM of pentamidine inhibited the expression of LdAAP3 mRNA. Treatment with 100 μM pentamidine and 500 μM canavanine led to a decrease of both LdAAP3 and LdPT mRNA levels in intracellular amastigotes, while 1 mM NMLAA did not have any significant effect on their expression (Fig 9A, 9B and 9C). Additionally, 100 μM pentamidine, 500 μM canavanine and 1 mM NMLAA did not significantly affect the infectivity of L. donovani in THP-1 cells (S2D Fig). In order to verify if ADR inhibition by arginine analogues in intracellular amastigotes was not due to changes in cell viability, THP-1 cells were treated with different concentrations of pentamidine, canavanine or NMLAA for 24 h and 48 h, following which they were subjected to MTT assay in order to determine their viability. Cells treated with 100 μM pentamidine, and 1 mM canavanine or NMLAA exhibited 80–100% viability, which was the concentration used for treating infected THP-1 cells for ADR inhibition (S4A–S4C Fig). In conclusion, the above results confirmed that the amidino group is the ligand that binds the arginine sensor. This means that pentamidine and canavanine are recognized by the sensor as “arginine”.
In this study, we have identified that the amidino group on the arginine side chain is the specific ligand that binds the L. donovani surface arginine sensor, and thereby activates an arginine deprivation response (ADR). We also show that arginine transporter and sensor binding sites are distinct in both axenic and intracellular L. donovani. Our analysis has indicated that the sensor is more selective in terms of its ligand as compared to the transporter. The arginine sensor not only detects the lack of arginine in the environment but also responds to excessive extracellular arginine by inducing rapid degradation of the LdAAP3 protein and mRNA. This study provides the first identification in Leishmania of an intermolecular region or functional group of arginine that interacts with a receptor. In all other sensors identified to date, such intermolecular recognition has not yet been described.
In our present study, it was observed that the arginine structural analogue canavanine and diamidine pentamidine inhibit not only arginine transport but also ADR, in axenic promastigotes as well as intracellular amastigotes. Arginine is a positively charged amino acid with an amidino group (pKa = 13.8) at the distal cap of its side chain [30]. Pentamidine, a diamidine, and a potent anti-protozoal agent, possesses two positively charged amidino moieties which are a part of the guanidium group and have a pKa of 12.1 [28,31,32,33]. Canavanine, on the other hand, is a structural analogue of arginine that has a deprotonated guanidinooxy group (pKa = 6.6) [25,34]. Considering that there is no similarity in the backbone structure of pentamidine and arginine other than the amidino moiety, each of these amidino moieties is the minimal structure which is required to be recognized by the arginine sensor. Any modification of the amidino group, such as the replacement of hydrogen with methyl-group (in case of NMLAA) or replacement of hydrogen with any other groups as in the case of the amidino-modified arginine side-chain analogues and compounds, resulted in their non-recognition by the arginine sensor even when these compounds have an arginine backbone. Among the nine compounds tested in the present study, only pentamidine, canavanine and NMLAA inhibited arginine transport in Leishmania, while the others are known to be poor transport inhibitors [8,24]. This provides conclusive evidence that the arginine sensor of L. donovani is highly specific and exclusive to the amidino group of arginine and also implies that arginine sensing and transport binding sites are distinct in Leishmania parasites, in axenic promastigotes and intracellular amastigotes. However, the arginine transporter site is still permissive to change (unlike the sensor site) which does not affect the net charge of the transported compound. This finding opens up the possibility of employing new arginine analogues containing the amidino group as therapeutic agents in leishmaniasis. D-Arginine is an enantiomer of L-Arginine. The parasite arginine sensor seems to have two levels of ligand recognition: i) the enantiomer (D/L) of the primary amine of arginine and ii) the amidino group. Thus, the presence of D-Arginine may not activate ADR even though it has an amidino group. When the chemical ligand lacks primary amine as in the case of pentamidine, the arginine sensor recognizes the presence of amidino group and hence triggers ADR.
The up-regulation of various members of the ADR pathway including LdAAP3, pteridine transporter, folate/biopterin transporter among others upon arginine starvation of L. donovani suggests that the ADR pathway not only regulates the expression of LdAAP3 but also other transporters of essential compounds such as vitamin B9 (folate) and pteridine which are essential for Leishmania. The variation in the RNA degradation profile seen during arginine sufficiency suggests that LdAAP3 and other members of the ADR pathway including LdPT are regulated differently. The mRNA degradation of LdAAP3 was slower as compared to LdPT as it is an central gene in the ADR pathway.
Macrophage phagolysosomes evolve from late endocytic compartments [35] and are the sites of Leishmania amastigote differentiation [36]. The arginine concentration in phagolysosomes was found to be 140 μM (Fischer-Weinberger et al, in preparation). We also found that ADR activation in intracellular L. donovani amastigotes occurred between 24 and 48 h post-infection under physiological levels of arginine (~100 μM), which is considerably higher than the concentration which induces ADR in axenic parasites (5 μM). Thus, intracellular arginine starvation builds up between 24 and 48 h post-infection, which may be the time taken for the depletion of arginine levels from 140 μM (the physiological concentration in phagolysosomes) to 5 μM (which is sufficient for ADR activation in axenic L. donovani). Hence, it is noteworthy that the axenic Leishmania parasite model system established by various research groups including ours [37] is well-suited for deciphering molecular mechanisms in Leishmania as this is a clean system without host-protein interference.
Our present study is the first to elucidate the specificity of the parasite arginine sensor, as well as its arginine sufficiency and deficiency responses, both in axenic and intracellular parasites. Sensing nutrient availability in vector or host environment may be essential for parasite survival and growth. Thus, nutrient sensing and transport pathways can be promising drug targets in the protozoan parasite Leishmania [38]. Our study also provides evidence that the L. donovani arginine sensor has a dual function of response to not just arginine deprivation as we have reported earlier [7], but also to arginine sufficiency, and thus maintains homeostasis. Transceptors are dual function solute transporters that concomitantly sense and translocate their substrates, and localize to cell surface and organelle membranes [39]. A well-characterized example of the transceptor is the SSY1 transceptor in Saccharomyces cerevisiae [40]. It is part of a membrane-sensing system that detects extracellular amino acids by binding to it [41,42] and signal transmission for detecting the presence of extracellular amino acids is initiated at the cell membrane. Mammals possess the System A amino acid transporter 2 (SNAT2), which also acts as a transceptor that signals and senses neutral amino acid availability [43]. More recently, an arginine trans-membrane sensor (SLC38A9) has been identified on the membrane of mammalian lysosomes [44,45]. However, signaling is initiated at the cell membrane where the arginine sensor is most likely localized. Whether the arginine sensor and transporter sites in L. donovani are located in the same protein or different proteins remains an open question. Also, the existence of multiple arginine transporters, sensors or transceptors implies that arginine is evolutionarily conserved and indispensable in both lower and higher organisms. In recent years, research on flagella-localized transporters in Leishmania suggest that these transporters may be involved in ligand sensing. Some of the well-known transporters include glucose transporter 1 (GT1) from L. mexicana [46] and aquaglyceroporin (AQP1) from L. major involved in osmoregulation [47]. The phenomenon of sensors localization on the flagella has recently been discussed [48]. In light of this, the fact that arginine is an essential amino acid in Leishmania, and semi-essential in mammals, represents its global role in various cellular processes.
In a commentary to our previous paper on the discovery of ADR, McConville [49] suggested that the arginine sensing phenomenon is a metabolic crosstalk between Leishmania and the macrophage host. Our finding in this work that ADR is activated in parasites inside phagolysosomes, early in infection, strongly supports this idea. Furthermore, the identification of the arginine intermolecular ligand is unprecedented and provides an efficient tool to further explore host-parasite interaction.
|
10.1371/journal.pgen.1003406 | Elevated Expression of the Integrin-Associated Protein PINCH Suppresses the Defects of Drosophila melanogaster Muscle Hypercontraction Mutants | A variety of human diseases arise from mutations that alter muscle contraction. Evolutionary conservation allows genetic studies in Drosophila melanogaster to be used to better understand these myopathies and suggest novel therapeutic strategies. Integrin-mediated adhesion is required to support muscle structure and function, and expression of Integrin adhesive complex (IAC) proteins is modulated to adapt to varying levels of mechanical stress within muscle. Mutations in flapwing (flw), a catalytic subunit of myosin phosphatase, result in non-muscle myosin hyperphosphorylation, as well as muscle hypercontraction, defects in size, motility, muscle attachment, and subsequent larval and pupal lethality. We find that moderately elevated expression of the IAC protein PINCH significantly rescues flw phenotypes. Rescue requires PINCH be bound to its partners, Integrin-linked kinase and Ras suppressor 1. Rescue is not achieved through dephosphorylation of non-muscle myosin, suggesting a mechanism in which elevated PINCH expression strengthens integrin adhesion. In support of this, elevated expression of PINCH rescues an independent muscle hypercontraction mutant in muscle myosin heavy chain, MhcSamba1. By testing a panel of IAC proteins, we show specificity for PINCH expression in the rescue of hypercontraction mutants. These data are consistent with a model in which PINCH is present in limiting quantities within IACs, with increasing PINCH expression reinforcing existing adhesions or allowing for the de novo assembly of new adhesion complexes. Moreover, in myopathies that exhibit hypercontraction, strategic PINCH expression may have therapeutic potential in preserving muscle structure and function.
| A wide variety of diseases of the muscle are caused by mutations that alter either the actin and myosin contractile machinery or its regulation. One class of mutations of interest results in hypercontraction of the muscle—actin and myosin fibers contract, but cannot efficiently relax. We have used the fruit fly as a model to study these mutations because of the striking similarity of fly and human muscle and because of the many genetic techniques that are available in the fly. Using a genetic approach we identified a protein, PINCH, whose increased expression can rescue the defects observed in hypercontraction mutants. PINCH is a component of integrin adhesion complexes, responsible for anchoring cells in their environment. This suggests that strengthening the anchorage of muscles via PINCH may be an effective strategy to prevent or reduce the muscle damage that occurs in diseases of muscle hypercontraction.
| Numerous human diseases, including muscular dystrophies [1], [2] and cardiomyopathies [3], [4], result from mutations that alter muscle contraction. The evolutionary conservation and genetic tractability of Drosophila melanogaster have made it an attractive system in which to characterize these genes and mutations [5], [6]. A collection of mutants in Drosophila exhibits myopathy due to muscle hypercontraction, leading to a range of phenotypes from flightless adults to early lethality [7]–[12]. Upon hypercontraction, these mutants can compensate in a variety of ways. Through a genetic response, specific mutations in genes such as the Myosin heavy chain (Mhc) can decrease overall actomyosin force, which is sufficient to suppress hypercontraction defects [11], [13]. Hypercontraction can also induce compensatory changes in gene expression. For instance, expression profiling of MhcSamba hypercontraction mutants suggested an extensive actin cytoskeletal remodeling response, as well as upregulation of non-muscle myosin phosphatase expression to promote relaxation of the actomyosin cytoskeleton [14]. Notably, upregulation of genes for integrin adhesive complex (IAC) proteins like Paxillin and Talin was observed as well [14]. This suggests that increasing the expression levels of IAC proteins can strengthen integrin-based adhesions at muscle termini to cope with the increased mechanical stress of muscle hypercontraction. Integrin-mediated adhesion has been shown to help maintain muscle cytoarchitecture and sarcomeric integrity [15], and mechanical force regulates the rate of integrin turnover [16]. Consistent with a key role for IAC gene expression in responding to muscle contraction, genes including UNC-97/PINCH show decreased mRNA levels in C. elegans muscles developed in the microgravity of spaceflight [17]. Furthermore, review of the ground controls (deposited in NCBI's Gene Expression Omnibus: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE36358) also reveals increased UNC-97 mRNA levels under conditions of hypergravity. This supports the idea that modulating gene expression of IAC components may be an evolutionarily conserved mechanism to adapt to widely divergent levels of mechanical stress to maintain muscle integrity.
Flapwing (Flw) is a serine/threonine Protein Phosphatase 1β (PP1β) that acts broadly upon many phosphorylated protein substrates, but it is required solely for its activity in dephosphorylating non-muscle myosin regulatory light chain (MRLC), encoded by spaghetti squash (sqh) in Drosophila [11]. Phosphorylation of Sqh causes an activating conformational change that promotes contraction of the actomyosin cytoskeleton. Sqh dephosphorylation is a key step in actomyosin relaxation. Thus strong loss-of-function mutants in flw exhibit hyperphosphorylation of non-muscle myosin. Notably, though flw does not genetically interact with the muscle version of MRLC (Mlc2) [10], the main defect observed in flw mutants is hypercontraction of myofibrillar myosin, eventually resulting in detachment of striated muscles and subsequent lethality at both larval and pupal stages [10]. Though not directly responsible for the generation of contractile force, non-muscle myosin is necessary for the proper development of myofibrils [18]. Hyperphosphorylation of Sqh disrupts actin cytoskeletal dynamics in many cell types, but the consequences appear greatest in contractile muscle tissue, both in larval body wall muscle and in adult indirect flight muscle [10]. The molecular connection between hypercontracted cytoskeleton and hypercontraction and detachment of muscle myofibrils is indirect, and currently not well defined.
We hypothesized that the flw muscle hypercontraction phenotype could be rescued by strengthening the anchoring of muscles at integrin-based adhesion sites to better withstand the force of hypercontraction. We show here that moderately elevated expression of the IAC protein PINCH alleviates the phenotypes of flw mutants. PINCH is an integrin-associated LIM domain adaptor protein encoded by the steamer duck (stck) gene in Drosophila. PINCH is part of an evolutionarily conserved, high affinity protein complex comprised of Ras Suppressor 1 (RSU1), Integrin-linked kinase (ILK) and Parvin that is targeted to the cell membrane at sites of integrin adhesion [19]–[22]. This complex of proteins has a critical role in adhesion maintenance, particularly in muscle attachment. In Drosophila, null mutants in PINCH, ILK, and Parvin each die late in embryogenesis as a consequence of failure of integrin-based muscle attachment sites within the segmental musculature [23]–[25]. Null mutations in RSU1 show milder defects in integrin adhesion between the epithelial layers of the wing [19]. Notably, mammalian α-Parvin and ILK have been shown to be negative regulators of contractility in certain cellular contexts [26], [27]. The RSU1, PINCH, ILK and Parvin proteins have been shown to mutually stabilize each other [19], [22], [28]–[30], suggesting that the complex as a whole might help delay the onset or reduce the extent of the damage exhibited in hypercontraction mutants by stabilizing adhesions. Data presented here support this idea. We show that PINCH must be capable of directly binding both ILK and RSU1 in order to suppress the defects of the flw mutant.
We demonstrate that structural stabilization of integrin-based adhesions by PINCH is likely to be a mechanism for the rescue of flw mutants. PINCH does not appear to alleviate hypercontraction of the myofibrils, as muscle attachment is better maintained even in the presence of muscle hypercontraction. We also eliminate phospho-regulation as a plausible mechanism, as elevated PINCH expression does not alter the phosphorylation status of the only essential Flapwing substrate, MRLC/Sqh. We show that moderately elevated expression of PINCH partially rescues another hypercontraction mutant, the Samba1 allele of Mhc. Furthermore, we show that among a panel of IAC proteins tested, elevated expression of Talin also affords marginal rescue of Samba1 hypercontraction defects. However, transgenic Talin expression is unable to afford rescue of flw larval lethality, underscoring the specificity of PINCH expression in strengthening adhesion under multiple conditions of hypercontraction. These studies have broad implications for understanding and treating a variety of myopathies and muscle degenerative diseases.
In a recent report, an interaction between mammalian Protein Phosphatase-1 (PP1) and the integrin-associated protein PINCH was described [31]. Because of the similar muscle detachment phenotypes exhibited by loss-of-function mutants for both PINCH [23] and PP1β [10] in the fly, we set out to test whether the Drosophila proteins also interact. Our analyses included two recessive alleles of the PP1β family member, flapwing. flw6 is a point mutant with decreased substrate affinity, and flw7 contains a lacZ enhancer trap inserted in the flw 5′ UTR which reduces the level of Flw expression [10]. Both of these mutants are strong loss-of-function alleles, but do not entirely eliminate Flapwing phosphatase activity or protein expression in male hemizygotes. In a wild type PINCH background, we combined a wild type PINCH-Flag transgene expressed from the native PINCH promoter with each of these flw alleles. We observed that expression of transgenic wild type PINCH-Flag significantly increased the frequency of hemizygous flw adult male escapers, and restored their fertility (Figure 1A, 1B). These crosses were done at 18°C because the frequency of adult escapers for any genotype was extremely low if reared at 25°C. A second, independent insertion line expressing wild type PINCH-Flag recapitulates the increased frequency of adult escapers (data not shown).
To more carefully characterize the spectrum of defects in flw mutants, we assessed the viability across development at 25°C of male hemizygotes selected from balanced stocks. Compared to a wild type w1118 control, we found significant larval lethality—approximately 70% of flw7 larvae failed to successfully pupariate, and of the 30% that do form pupae, essentially all arrest development prior to adult eclosion (Figure 2A, 2B). Expression of transgenic PINCH-Flag in a wild type background at approximately 40% of endogenous levels had no significant effect on progression to adulthood when compared to wild type animals (Figure 2A, 2C, 2E). However, this modest expression of PINCH-Flag fully rescued the larval lethality of the flw7 mutant (Figure 2B, 2D, 2E). At 25°C (unlike the 18°C data in Figure 1), pupal lethality is not rescued by elevated expression of PINCH-Flag (Figure 2D), which may reflect a more stringent requirement for Flapwing phosphatase activity in the pupa to adult transition at higher temperature. The larval lethality of the flw6 allele is similarly rescued by PINCH-Flag expression (Figure S1). Because of the similarity between flw6 and flw7 in 1) the number of adult escapers at 18°C, 2) the developmental profile at 25°C, and 3) the rescue of larval lethality by expression of transgenic PINCH, we focus on flw7 in the remaining experiments.
We also tested whether a reduction in PINCH exacerbates flw lethality. Decreasing PINCH dosage using a heterozygous stck17 mutation in the flw7 background was challenging due to a high rate of balancer breakdown in the required crosses. However, the limited number of flw7; stck17/+ animals that were isolated exhibited complete larval lethality (n = 44) (Figure 2F). These data confirm that reducing PINCH gene dosage indeed enhances flw lethality.
One way that increased PINCH-Flag expression could rescue flw larval lethality is through reversing protein hyper-phosphorylation that results from flw loss-of-function. Because the sole essential activity of Flapwing is the dephosphorylation of non-muscle myosin regulatory light chain (MRLC)/Sqh [11], we examined the phosphorylation state of MRLC/Sqh in larval lysates. As expected, phospho-MRLC was elevated significantly in flw7 lysates as compared to wild type w1118 controls (Figure 2E). However, statistical analysis of four independent replicate experiments showed that expression of PINCH-Flag, either in wild type flies or in the flw mutant background, had no effect on the phosphorylation state of MRLC (Figure 2E). These data indicate that regulation of MRLC phosphorylation state by PINCH is not the mechanism for the genetic suppression we observe between flw and PINCH.
We also examined the phosphorylation states of several other phospho-proteins that have been connected to PINCH: Akt [31], JNK [19], and ERK [32]. We found that phosphorylation of these proteins is not consistently elevated in larval lysates from flw mutants (data not shown), indicating either that Akt, JNK, and ERK are not actively signaling in these samples, are not substrates of Flapwing, or that other phosphatases compensate to appropriately dephosphorylate Akt, JNK and ERK in the flw mutant. While it is formally possible that PINCH could contribute to the phospho-regulation of non-essential Flapwing targets, it seems unlikely that phospho-regulation is a major contributor to the suppression of flw lethality observed upon elevated PINCH expression.
In order to molecularly dissect the mechanism by which PINCH is suppressing flw larval lethality, we employed several mutants of PINCH that disrupt binding of its known partners, Integrin-linked kinase (ILK) and Ras Suppressor 1 (RSU1) (Figure 3A). One of these mutants, PINCHQ38A, has previously been shown to disrupt binding of ILK to LIM1 of PINCH (Figure 3A) [30], [33], [34], allowing us to test whether ILK binding and PINCH-ILK complexes are participating in the suppression of flw lethality. We were also interested in testing the contribution of RSU1 binding to LIM5 of PINCH (Figure 3A), and the role PINCH-RSU1 complexes. Mutations in PINCH that disrupt RSU1 binding have not yet been described. Therefore, we designed a modified yeast two-hybrid screen to identify point mutations in LIM5 of PINCH that specifically disrupt binding to RSU1 while otherwise preserving the structure and function of PINCH. We identified PINCHD303V as a strong candidate. Among PINCH LIM5 sequences, D303 is highly evolutionarily conserved (Figure 3B), suggesting its importance for the proper functioning of PINCH. However, D303 is a variable residue within the general LIM consensus sequence (Figure 3C) and can therefore be altered without destroying overall LIM structure. The disruption of RSU1 binding was tested by inserting the D303V mutation into full length, Histidine-tagged PINCH expressed in Drosophila S2 cells. In the cell lysates used for purification, we routinely observe that PINCHD303V-His is present at reduced levels as compared to wild type PINCH-His. PINCH that is not bound to RSU1 has previously been shown to be less stable [19], [35]. Even so, Ni-NTA purification of the PINCH-His complexes confirms that PINCHD303VHis, in contrast to a wild type PINCH-His control, clearly does not co-purify with RSU1 (Figure 3D). Moreover, PINCHD303VHis retains the ability to bind to ILK (Figure 3D), indicating that folding and stability of the mutant PINCHD303V have not been completely destroyed.
To further confirm that the PINCHD303V mutation disrupts RSU1 binding, we next expressed PINCHD303VFlag in flies using the native PINCH promoter. In a wild type genetic background where the transgene is expressed at low levels, we did not observe any obvious dominant effects (data not shown). When this transgene was introduced into the stck17/stck18 PINCH null background, viable rescued adults were produced at 90% of the expected frequency (n = 209). In soluble extracts made from these rescued adults, PINCHD303VFlag is routinely present at levels that are reduced compared to animals expressing either endogenous PINCH or rescued with the wild type PINCH-Flag transgene. In an effort to understand why PINCHD303V is present at lower levels, we performed semi-quantitative RT-PCR on mRNA isolated from adult flies lacking endogenous PINCH and rescued with either wild type, Q38A, or D303V versions of the PINCH transgene. These mRNAs differ only in the sequence of a single codon. We find that levels of each transgenic message are equal (Figure S2), confirming that differences in the levels of PINCH protein occur post-transcriptionally. Additionally, levels of protein for RSU1 and ILK were reduced in the PINCHD303VFlag animals (Figure 3E), likely reflecting a well-characterized destabilization of the proteins in this complex when intact PINCH complexes are disrupted [19], [22], [28]–[30]. While PINCHD303VFlag rescued animals are viable as adults, wing blisters are frequent, the wings are often observed in a “held up” posture (similar, for example, to mutations in the Troponin genes wupA and wupB [12]), and the rescued adults are flightless. These phenotypes are comparable in severity to those of the viable RSU1 null mutant [19], and demonstrate the importance of the PINCH-RSU1 interaction. Flag pull-downs from rescued adult lysates confirm that both ILK and RSU1 strongly associate with wild type PINCH-Flag. Similar to the cell culture experiments, PINCHD303VFlag clearly retains the ability to bind ILK, but no RSU1 is detected in the PINCHD303VFlag pull-downs (Figure 3E). This confirms that the PINCHD303V mutation effectively disrupts the binding of RSU1.
Similar to the experiment (Figure 2) in which we suppressed larval lethality by expressing wild type PINCH-Flag in the flw7 background, we expressed low levels of transgenic PINCHQ38AFlag along with endogenous wild type PINCH (Figure 4B, 4E). Unlike wild type PINCH-Flag expression, expression of the PINCHQ38AFlag mutant in the flw7 background does not allow for suppression of larval lethality (Figure 4B compare to Figure 2B). This occurs despite viability of the PINCHQ38A transgene expressed in a wild type background that is comparable to the wild type w1118 control (Figure 4A compare to Figure 2A), and the full rescue of stck null mutants by PINCHQ38A [35]. These data indicate that PINCH-ILK complexes participate in the suppression of flw larval lethality. Expression of PINCHQ38AFlag, which differs from the wild type transgenic protein in only a single amino acid and in its ability to bind ILK, has no capacity to promote pupariation in the flw mutant.
Next, we tested for suppression of flw larval lethality by co-expressing transgenic PINCHD303VFlag along with endogenous wild type PINCH in the flw7 background (Figure 4D, 4E). Like PINCHQ38AFlag, low levels of expression of PINCHD303VFlag do not suppress flw larval lethality. Despite the pupal and adult viability of PINCHD303VFlag expressed in a wild type background (Figure 4C, 4E), pupariation was never observed in the flw7; PINCHD303VFlag animals (Figure 4D). Taken together, this suggests that intact ILK-PINCH-RSU1 complexes are necessary for the rescue observed in the flw mutant upon expression of transgenic PINCH. It is interesting to note that not only do the mutant PINCH transgenes fail to rescue flw7 lethality, they in fact exacerbate the larval lethality, eliminating pupariation entirely. We comment on possible mechanisms for this enhancement as well as its implications below.
How does increased expression of PINCH and formation of functional PINCH complexes allow for improved survival of flw mutant larvae? One possibility is that in order to cope with the downstream effects of compromised protein dephosphorylation and a constitutively contracted actomyosin cytoskeleton, increased levels of PINCH and its associated binding partners strengthen integrin-dependent adhesions. Mutant analyses of PINCH and ILK in the fly indicate that these proteins have a role in maintaining embryonic muscle attachment [23], [24], and both Drosophila and C. elegans RNAi experiments confirm a role for these proteins in adult muscle maintenance [15], [36], [37]. If moderately elevated expression of PINCH leads to stronger adhesions, this may allow the flw mutants to better withstand the elevated mechanical stress they experience. To test this idea, we performed a series of morphological and functional assays to characterize flw mutants.
In addition to reduced viability, flw7 mutant larvae have decreased and more variable size as compared to a w1118 wild type control (Figure 5A, 5B). The small size of the flw larvae may result from either developmental delay or arrest, or be a secondary effect of feeding deficits due to poor muscle function. Expression of PINCH-Flag alone does not affect larval size (Figure 5A, 5B). However, elevated expression of PINCH-Flag in the flw mutant produces animals that are significantly larger than even the w1118 wild type control (Figure 5A, 5B). Likewise, expression of either PINCHQ38AFlag or PINCHD303VFlag alone has no effect on larval size (Figure 5A, 5B). However, expression of PINCHQ38AFlag or PINCHD303VFlag in the flw background fails to restore normal larval size and yields animals that are significantly smaller than the flw mutant alone (Figure 5A, 5B).
We next analyzed these same genotypes of animals in a motility assay that measures spontaneous larval crawling through five zones on a grape juice agar plate (Figure 5C). Wild type w1118 larvae, as well as matched samples expressing the PINCH-Flag, PINCHQ38AFlag or PINCHD303VFlag transgenes, exhibit robust and comparable levels of larval motility (Figure 5D and Figure S3). flw larvae exhibit decreased motility, which is rescued by PINCH-Flag expression (Figure 5D). In fact, the flw7; PINFL2 animals are significantly more motile than even the wild type controls, perhaps due to their larger size. Expression of either PINCHD303VFlag or PINCHQ38AFlag respectively in the flw7 mutant fails to rescue larval motility and increasingly exacerbates the motility defects of flw7 animals (Figure 5D). These data show that 1) moderately elevated expression of PINCH rescues muscle function required for larval crawling, and 2) mutant PINCH transgenes that cannot form functional ILK-PINCH-RSU1 complexes fail to rescue larval size and motility, as well as overall viability.
Assessing structural integrity of the muscle in the flw larvae presents several technical challenges: antibody reagents cannot penetrate the larval cuticle of intact animals, and the physical manipulations required for dissection introduce structural defects not present in the intact larvae, particularly because the muscle integrity in these animals is already compromised. To circumvent these issues, we crossed the flw7 mutant with a Zasp66ZCL0633 P-element insertion line, which contains a GFP coding sequence inserted into the genomic locus for the muscle-specific PDZ protein, Zasp66 [38]. Zasp-GFP localizes normally to muscle Z-lines and has been extensively used to visualize muscle structure without the need for dissection in a variety of genetic backgrounds [39]. Although there is failure in pupation later in development, the insertion of GFP into the Zasp66 locus does not significantly impact the viability or motility of the flw7 mutant larvae at the four-day time point we used to analyze muscle integrity (Figure S4). flw mutants are reported to exhibit larval muscle detachment [10], and indeed in the flw7; Zasp-GFP larvae we observe frequent detachment, particularly in the Ventral Intersegmental (VIS) muscles. Because of the ease with which the VIS muscles can be scored for detachment, we have used this group of muscles as a read-out for muscle integrity. Zasp-GFP controls show very little detachment of the VIS muscles (Figure 6A, 6F). Additionally, expression of a single copy of the PINCH-Flag, PINCHQ38AFlag or PINCHD303VFlag transgenes in the Zasp-GFP background does not alter the normal attachment of the VIS muscles (Figure S5). However, flw mutants show frequent VIS muscle detachment (Figure 6B, 6F) that is partially rescued by expression of a single copy of the wild-type PINCH-Flag transgene (Figure 6C, 6F). Notably, hypercontraction, evidenced by areas of saturated Zasp-GFP signal (Figure 6B–6E, asterisks), is still frequently observed in the flw7; PINFL2/+; Zasp-GFP/+ animals (Figure 6C). This is consistent with the idea that PINCH is not eliminating hypercontraction, but rather stabilizes integrin adhesions to prevent muscle detachment and subsequent larval lethality in the flw mutants. In contrast to the robust rescue of detachment observed upon transgenic expression of wild type PINCH-Flag in the flw7 mutant, expression of PINCHQ38AFlag does not rescue detachment, and the PINCHD303VFlag transgene does not rescue muscle detachment to the same extent (Figure 6D–6F). The frequency of detachment upon transgenic PINCH expression mirrors the level of lethality observed in four-day-old larvae of the corresponding genotype (Figure 2, Figure 4, Figure S5). These data suggest that an increase in the number of intact PINCH-ILK-RSU1 complexes enables retention of muscle attachments in the flw mutant background.
If moderately elevated expression of PINCH is rescuing flw phenotypes by stabilizing integrin-based adhesions, increased expression of PINCH might also be expected to rescue the phenotypes of other hypercontraction mutants. To test this, we employed a dominant hypercontraction mutant in muscle Myosin Heavy Chain called MhcSamba1 [14]. The molecular lesion of MhcSamba1 is in the ATP binding/hydrolysis domain, producing muscle hypercontraction by a direct molecular mechanism distinct from the flw mutants. The muscle defects of MhcSamba1 are apparent in heterozygous adults in a geotaxis assay in which animals are induced to climb [15], [40]. MhcSamba1 heterozygous mutant adults exhibit poor climbing ability that is significantly improved upon expression of transgenic PINCH-Flag (Figure 7A). These data support the idea that increased expression of PINCH generally stabilizes muscle attachments where hypercontraction is present.
To further characterize the rescue of Samba1 mutants by transgenic PINCH-Flag expression, we examined indirect flight muscle (IFM). IFM hypercontraction can be seen by polarized light microscopy in the Samba1 mutant ([9] and Figure 7B), and renders these animals flightless [9]. Consistent with the idea that elevated PINCH expression is stabilizing integrin attachments rather than reversing hypercontraction, the IFM of Samba1/PINCH-Flag adults does not resemble wild type IFM, but continues to exhibit hypercontraction (Figure 7B), and these flies remain flightless (data not shown). As a consequence of IFM hypercontraction, a portion of Samba1/+ animals (39%, n = 70) exhibit visible external thoracic indentations in freshly eclosed adults ([9] and Figure 7C). Although expression of PINCH-Flag in the Samba1 heterozygotes does not rescue IFM hypercontraction, it dramatically improves the thoracic indentation phenotype (2% indented, n = 62) (Figure 7C).
We next wanted to determine whether increased expression of other integrin adhesive complex (IAC) components might be protective under conditions of hypercontraction, or whether PINCH is unique in this regard. To test this, beginning with the geotaxis assay in the Samba1/+ mutant background, we employed a panel of IAC transgenic proteins to assess suppression of hypercontraction-induced climbing defects. This panel included ILK-GFP and Tensin-GFP controlled by their native promoters, FAK-GFP and Zyxin-GFP expressed using the Gal4-UAS system and the muscle-specific 24B driver, and Ubi-βPS-Integrin-YFP and Ubi-Talin expressed from the ubiquitin promoter. From among this panel, ILK-GFP was not overexpressed and FAK-GFP expression was lethal prior to adulthood [41], so these proteins were not tested in the geotaxis assay. Western analyses demonstrated increased expression of Zyxin-GFP, βPS-Integrin-YFP, and Talin in adult fly lysates (data not shown). We could not confirm expression of Tensin-GFP because of lack of antibody reagents, but this transgene has previously been shown to express, as it rescues null mutations in the Tensin gene blistery [42]. The presence of the Tensin-GFP, Zyxin-GFP, and βPS-Integrin-YFP transgenes had no significant effect on the climbing ability of Samba1 mutants (Figure 8A). Of the additional IAC proteins tested, only transgenic Talin expression resulted in a modest improvement of the geotaxis score for the MhcSamba1 mutant (Figure 8A). Rescue of IFM hypercontraction and thoracic indentation was not observed with expression of transgenic Talin (data not shown). This suggests that if increased expression of Talin does strengthen integrin adhesions, it is unlikely to do so in the same manner as PINCH.
In efforts to bolster the Samba1 geotaxis data, we further analyzed two of the panel of IAC proteins in the flw genetic background: Zyxin and Talin. We first analyzed the effect of Gal4-UAS expression of Zyxin-GFP from the muscle-specific 24B driver on the developmental profile of flw7. Animals expressing 24B>Zyxin-GFP alone show normal developmental progression (data not shown). As predicted from the lack of rescue of Samba1 geotaxis, we see no significant difference in either the rate of larval or pupal lethality upon expression of Zyxin-GFP in the flw7 mutants (Figure 8B, compare to Figure 2B). Next, we showed that upon expression of Ubi-Talin in the flw7 background, we did not observe any rescue of larval lethality (Figure 8C). In fact, pupariation was blocked entirely in flw7; Talin/+ animals. Because transgenic Talin expression does not uniformly rescue both of the hypercontraction mutants tested, the biological significance of the marginal rescue by Talin in the Samba1/+ geotaxis assay is unclear. Moreover, these data suggest a high degree of specificity in the rescue of multiple hypercontraction phenotypes by PINCH expression.
In this report, we tested whether elevated expression of the IAC protein PINCH, as a means to strengthen integrin-based adhesions, can alleviate the phenotypes associated with muscle hypercontraction. We show that moderately elevated expression of PINCH dramatically rescues the lethality as well as the size, motility and muscle attachment defects associated with the myosin phosphatase hypercontraction mutant, flw. The interaction of PINCH with its binding partners ILK and RSU1 is required for rescue. Elevated PINCH expression neither alters the phosphorylation state of the essential Flapwing substrate MRLC, nor alleviates the hypercontraction observed in the larval VIS muscles, but instead may rescue through the stabilization of integrin-mediated adhesions. This idea is further supported by the partial rescue of the climbing deficits of an independent hypercontraction mutant in muscle myosin heavy chain, MhcSamba1, as well as the rescue of its thoracic indentation phenotype, without reversal of the hypercontraction evident in the IFM.
Initial work on this project included testing for interactions between PINCH and PP1 phosphatase family members in Drosophila. Data in mammalian cells shows that PP1α binds to PINCH via a KFVEF sequence motif present in LIM5 of PINCH, and that binding inhibits the phosphatase activity of PP1α [31]. We tested extensively for direct binding between Flapwing and PINCH in Drosophila using a variety of biochemical approaches, but were unable to demonstrate a physical interaction (data not shown). Moreover, if PINCH-PP1 binding and inhibition were evolutionarily conserved, we predicted that elevated expression of PINCH in Drosophila should exacerbate the defects of a flw hypomorph and lead to further hyper-phosphorylation of Flapwing targets like MRLC. We did not observe either of these things—rather, elevated expression of PINCH rescued flw phenotypes and had no effect on the phosphorylation state of the Flapwing target MRLC. There are several possible explanations for these discrepancies. First and least interesting, a PINCH-PP1 direct binding interaction may simply not be conserved in Drosophila—an idea we cannot disprove with negative biochemical data. Second, Flapwing is a PP1β family member rather than PP1α. While there is a high degree of sequence similarity between the two branches of the PP1 family, the distinguishing residues might play a key role in specifying PINCH binding. Our experiments do not address whether PINCH directly binds and inhibits PP1α isoforms in Drosophila, but rather show that PINCH is having a distinct and separate effect under conditions of PP1β loss-of-function via the stabilization of integrin-based adhesions. Third and most intriguing, the role of RSU1 in the PINCH-PP1α interaction is not yet clear. Both RSU1 and PP1α bind to LIM5 of mammalian PINCH. We have shown that PINCHD303 is a key residue for the binding of RSU1. Notably, the mammalian PINCHKFVEF motif that is responsible for PP1α binding [31] spans the immediately adjacent residues corresponding to 298–302 in the fly protein, which are moderately well conserved (KFYEY). As such, the binding interface on PINCH that participates in PP1α binding is likely to be overlapping with the RSU1 binding interface. It remains to be determined whether PINCH can bind both RSU1 and PP1α simultaneously, whether binding of PINCH to RSU1 and PP1α is mutually exclusive, or whether RSU1 rather than PINCH directly binds PP1α. Mutational analyses on the KVFEF motif of mammalian PINCH disrupted the association with PP1α [31], but it is formally possible that the KFVEF mutant disrupts RSU1 binding, preventing RSU1-PP1α complexes from docking on PINCH. Further studies on the mammalian versions of these proteins will be necessary to distinguish between these possibilities.
Expression of the binding mutants, PINCHQ38AFlag and PINCHD303VFlag, resulted in strong phenotypes in the context of Flapwing loss-of-function, eliminating pupariation entirely. Of note, expression from the Ubi-Talin transgene or the presence of the Zasp66-GFP gene trap had the equivalent effect of failed flw pupariation. This suggests that in the flw background, developmental events required for pupariation are particularly prone to disruption by alterations in the IAC. This makes the dramatic rescue of flw pupariation upon expression of PINCH-Flag all the more remarkable.
Expression of PINCHQ38AFlag and PINCHD303VFlag produces no dominant effects in a wild type background. There appears to be plasticity in IAC formation/maintenance to tolerate the presence of a small fraction of incomplete mutant PINCH complexes [35]. However, expression of the mutant PINCH transgenes creates a background that is sensitized to further perturbations. Upon flw loss-of-function, the presence of incomplete mutant PINCH complexes has severe and reproducible consequences on viability, size, and motility. There is a clear synergistic effect of combining flw loss-of-function and the corresponding alterations in protein phosphorylation, with expression from the mutant PINCH transgenes.
Our data suggest that PINCHQ38AFlag and PINCHD303VFlag participate in the formation of incomplete/unproductive PINCH complexes that may act dominantly in the flw background. This is consistent with numerous examples in cell culture in which overexpression of mutant forms of PINCH titrates away crucial binding partners to dominantly affect processes such as protein localization and cell spreading [34], [43]–[47]. The effect of mutant PINCH expression on the viability of flw animals is substantial, despite the relatively small amount of PINCHQ38AFlag and PINCHD303VFlag protein that is expressed (approximately 20–30% of endogenous PINCH). This may occur because of the mutual protein stabilization phenomenon exhibited by RSU1-PINCH-ILK-Parvin complexes [19], [22], [28]–[30]. It is plausible that incomplete complexes containing mutant PINCH may be destabilized and more rapidly turned over, resulting in insufficient numbers of intact complexes to maintain optimal adhesive function in the flw background. Indeed, in C. elegans muscle, incomplete integrin adhesive complexes have been shown to accelerate the rate of IAC protein degradation to disrupt myofibrillar and mitochondrial morphology [36], [37].
PINCH-Flag expression rescues the phenotypes of three different hypercontraction mutations: flw6 and flw7 in non-muscle myosin phosphatase, and MhcSamba1 in muscle myosin heavy chain. This strongly supports the idea that elevated PINCH expression strengthens integrin-based adhesion, and more robust adhesion can mitigate hypercontraction-induced muscle damage. Transgenic expression of additional IAC components from readily available stocks was first tested by geotaxis in the MhcSamba1 mutant because of the ease of generating Samba1 heterozygotes expressing transgenic IAC proteins. The results from this panel of transgenes were mixed. Some IAC proteins did not readily overexpress (ILK-GFP), suggesting that tight regulation induces more rapid turnover upon increased gene expression. Overexpression of another IAC protein was lethal (24B>FAK-GFP) and it therefore could not be tested in the rescue of adult Samba mutant phenotypes. Increased expression of several additional IAC proteins had no effect (24B>Zyxin-GFP, Tensin-GFP, Ubi-βPS-Integrin-YFP). Lack of rescue could result if increased expression does not serve to strengthen adhesions because that component is not limiting for IAC formation or function. Alternatively, if the IAC transgene is not adequately expressed in the appropriate tissues (e.g.: both muscle and tendon cells into which they anchor), the transgenic protein will be unable to effectively strengthen adhesion. The 3kb PINCH promoter employed appears to have a useful expression pattern in this regard. It remains to be seen whether other promoters and other IAC genes can direct expression that will strengthen adhesion to the same degree. It will be interesting in future studies to delineate which IAC proteins are limiting at muscle attachments as well as other adhesion sites, as better techniques to determine relative expression levels and stoichiometry are developed. Interestingly, Talin and Paxillin expression was upregulated in the microarray studies of the Samba mutants [14], suggesting that upregulation of select IAC proteins is a compensatory response to hypercontraction that is utilized in vivo. With elevated expression of PINCH-Flag, partial rescue of the Samba and flw defects was observed. This suggests that PINCH might be a limiting component of IACs, and increasing PINCH expression might serve to strengthen existing complexes and/or allow the assembly of additional IACs. Thus, strategic expression of PINCH, and perhaps other IAC proteins as well, may serve to preserve or protect muscle structure and function and may have therapeutic potential in myopathies that exhibit hypercontraction.
flw6, flw7 (also known as flwG0172), P[GawB]how24B, and Zasp66ZCL0663 were obtained from the Bloomington Stock Center. MhcSamba1 [9], stck17 and stck18 [23] have been previously described. Transgenic stocks include PINCH-Flag and PINCHQ38AFlag [30], UAS-Zyxin-GFP [48], ILK-GFP [24], UAS-FAK-GFP [41], Ubi-βPS-Integrin-YFP [49], Tensin-GFP [42], and Ubi-Talin [50].
Crosses between w; PINFL2 and both flw6/FM7i and flw7/FM7i were set at 18°C and progeny counted. The percent escape was calculated as 100×[flw males/(total number of female progeny/2)]. Escaper male hemizygotes were tested for fertility by crossing to w1118 virgin females.
Stocks of the indicated genotypes were placed in cages and allowed to lay for approximately 6 hours on yeasted grape agar plates at 25°C. Except where noted, >80 animals of the appropriate genotype were selected from stocks as early L1s (24–30 hours) and placed on fresh grape agar plates. For stocks that are FM7i balanced, flw hemizygous males were selected by sorting against the GFP expressed from the FM7i balancer chromosome. A cross between flw7/FM7i and FM7i/Y; stck17/TM3, twi>GFP was used to generate flw7; stck17/+ animals, by selecting against the GFP expressed on both balancer chromosomes. On subsequent days, the number of viable animals at each stage of development was counted and live animals were moved to a fresh plate. Putative larval stages were determined by size. In Figure 2F, the small number of animals in each cohort precluded this determination, and total larvae were counted without respect to putative stage. Any pupae present continued to be counted until the final time point presented in the graphs, when they were scored for viability by the presence of visible and healthy looking adult structures with no signs of necrosis or dehydration. All viable pupae were kept an additional 7 days to determine if they eventually eclosed as adults. A wild type w1118 control was included to determine baseline lethality arising from experimental manipulation of the samples. Graphical analyses were done using GraphPad Prism.
For each analysis that involves a PINCH transgene, one of at least two independent insertion lines that were analyzed is shown, to confirm that the results were not dependent upon the locus of the transgene insertion.
Four-day-old L3 larvae were homogenized in RIPA buffer with phosphatase inhibitors and normalized by protein content (BioRad DC protein assay). Antibodies employed were anti-PINCH [23], anti-RSU1 [19], anti-ILK (BD #611802), anti-P-MRLC/Sqh (Cell Signaling), and anti-RACK1 [51] or anti-α-tubulin 12G10 (DSHB) as a loading control. All western analyses were performed at least three times and representative blots are shown. Densitometric analyses of western blots were performed using Image J.
In prior experiments that mapped the site of RSU1 interaction to LIM5 of PINCH, we constructed a LIM5 bait and RSU1 prey that activated an ADE2 reporter in a yeast two-hybrid system [19], [52]. Employing a low-fidelity polymerase, we amplified the LIM5 region of PINCH and sub-cloned it into the bait vector to create a randomly mutagenized LIM5 bait library. The RSU1-binding function of LIM5 is unaffected in most library clones, which display ADE2 reporter activity. In a fraction of the colonies, the LIM5-RSU1 interaction is disrupted. On non-selective plates, lack of ADE2 reporter activity allows a pink-colored precursor in the adenine biosynthetic pathway to accumulate. We analyzed pink colonies resulting from the co-transformation of the mutagenized PINCH LIM5 bait library and wild type RSU1 prey. Total DNA was isolated, and LIM5-encoding DNA was PCR amplified and sequenced. Frame shifts, truncations, or point mutations in the zinc ligands were not considered further. Sequence alignments were done using Clustal X.
PCR mutagenesis of a previously described pMT-PINCHwt-His construct [19] was used to introduce an Aspartate to Valine mutation at position 303 in the dPINCHa cDNA. pMT-PINCHwtHis and pMT-PINCHD303VHis were stably transfected into S2 cells using standard methods and expression induced by the addition of CuSO4. S2 cell lysates were prepared in lysis buffer (50 mM Tris-HCl pH 7.9, 150 mM NaCl, 0.1% Triton-X 100) plus protease inhibitors, and were incubated with Ni-NTA agarose (Qiagen), clarified by centrifugation, washed with lysis buffer, then boiled in 2× Laemmli sample buffer prior to western blotting.
Transgenic flies carrying PINCHD303VFlag were generated by PCR mutagenesis of a previously described pCasper construct containing genomic PINCHwt-Flag [30]. This DNA construct was injected into embryos for p-element transposition by Genetic Services Inc. (Cambridge, MA). Transgenic flies were then crossed into a PINCH null background (stck17/stck18). Adult fly lysates were prepared in lysis buffer plus protease inhibitors, clarified by centrifugation and 0.45 µm filtration, and were incubated with anti-Flag M2 agarose (Sigma), washed, and boiled in 2× Laemmli sample buffer for western blotting.
Twelve adult flies per sample were lysed in 350 µl RLT Buffer using Qiashredder columns (Qiagen), and RNA was extracted using RNEasy Mini Kits (Qiagen) according to the manufacturer's recommendations. RT-PCR was conducted with 50 ng RNA per reaction using the Access RT-PCR System (Promega), with the following primers: PINCH-Flag (Forward: 5′GCACTGGCATGTGGAACATT3′, reverse: 5′ACTAGTCTACCTGTCATCGTC3′), GAPDH (Forward: 5′CAACTTCTGCGAAACGACAA3′, Reverse: 5′TGTCCTCCAGACCCTTGTTC3′).
Larvae of the given genotypes were sorted at 24–30 hr after egg lay and analyzed when they reached 4 days of age (3rd instar). Prior to size measurements, larvae were heat fixed by rapidly submerging in boiling embryo wash (0.1% Triton X-100, 0.7% NaCl), and bright field images captured on an Olympus MVX10 dissecting microscope. Length measurements of individual larvae were normalized to the mean of a matched wild type w1118 control. To measure larval motility, we adapted an existing protocol [53]. Eight four-day-old L3s of the desired genotype were placed onto a room temperature, 5 cm grape agar plate marked with zones of 5 concentric circles (Figure 5B). After acclimating for at least 5 minutes, all 8 animals were moved into zone 1, and the assay started. At 60 sec, the zonal location of all 8 animals was recorded. These measurements were performed in triplicate for each group of 8 larvae, and data for ≥4 independent sets (≥32 animals) were collected. The relative distribution of animals at the end of the motility assay was plotted in GraphPad Prism, using the mean percent of animals present in each zone.
Four-day-old L3s of the given genotypes were generated in a cross with flw7/FM7i; Zasp-GFP. Larvae were quickly heat fixed by rapidly submerging in boiling embryo wash (0.1% Triton X-100, 0.7% NaCl), then the Zasp66-GFP pattern in the Ventral Intersegmental (VIS) muscles was examined using an Olympus MVX10 GFP-dissecting microscope. Any animal in which the medial pair of VIS muscles does not correctly span body segments T1-T3 is scored as detached. Discontinuities in the Zasp-GFP signal are characteristic of breaks/detachments in the VIS muscles. Areas of hypercontraction are characterized by intense Zasp-GFP signal, as the Z-lines are more closely spaced. For each genotype, analyses were conducted in triplicate. Paired t tests were performed to determine statistically significant differences between genotypes.
Assays were done according to a previously described protocol [15], [40]. At least 50 flies of each genotype were tested. In each assay, 10 adult flies (<48 hours post-eclosion) were transferred to an empty vial and lightly tapped to the bottom. The number of flies that climbed to a height of 7 cm within 8 seconds was recorded. For each vial, the measurement was repeated 4 times and averaged. To control for minor variations in assay conditions, all data points were normalized to a wild type control sample that was collected and analyzed in parallel (ie: the number of wild type flies that met climbing criterion was converted to a geotaxis score = 1.0, to which other genotypes were compared). At least five sets of normalized data were used to graph a geotaxis score±SEM. Paired t tests were performed to determine statistically significant differences between genotypes.
IFM was analyzed as described previously [54], with several modifications. Briefly, adult flies of the indicated genotypes (n>20) were collected and dehydrated first in 100% ethanol, then in 100% isopropanol. Heads and abdomens were removed to speed equilibration. Thoraces were cleared >1 hour in xylenes, followed by >1 hour in BABB (1∶2 Benzyl Alcohol∶Benzyl Benzoate), then imaged in BABB using two external polarizing filters in conjunction with an Olympus MVX10 dissecting microscope. Thoracic indentations resulting from IFM hypercontraction were scored in >50 animals of each genotype upon visual inspection of freshly eclosed adults on a dissecting microscope.
|
10.1371/journal.ppat.1005899 | SCF Ubiquitin Ligase F-box Protein Fbx15 Controls Nuclear Co-repressor Localization, Stress Response and Virulence of the Human Pathogen Aspergillus fumigatus | F-box proteins share the F-box domain to connect substrates of E3 SCF ubiquitin RING ligases through the adaptor Skp1/A to Cul1/A scaffolds. F-box protein Fbx15 is part of the general stress response of the human pathogenic mold Aspergillus fumigatus. Oxidative stress induces a transient peak of fbx15 expression, resulting in 3x elevated Fbx15 protein levels. During non-stress conditions Fbx15 is phosphorylated and F-box mediated interaction with SkpA preferentially happens in smaller subpopulations in the cytoplasm. The F-box of Fbx15 is required for an appropriate oxidative stress response, which results in rapid dephosphorylation of Fbx15 and a shift of the cellular interaction with SkpA to the nucleus. Fbx15 binds SsnF/Ssn6 as part of the RcoA/Tup1-SsnF/Ssn6 co-repressor and is required for its correct nuclear localization. Dephosphorylated Fbx15 prevents SsnF/Ssn6 nuclear localization and results in the derepression of gliotoxin gene expression. fbx15 deletion mutants are unable to infect immunocompromised mice in a model for invasive aspergillosis. Fbx15 has a novel dual molecular function by controlling transcriptional repression and being part of SCF E3 ubiquitin ligases, which is essential for stress response, gliotoxin production and virulence in the opportunistic human pathogen A. fumigatus.
| The opportunistic human fungal pathogen Aspergillus fumigatus is the most prevalent cause for severe fungal infections in immunocompromised hosts. A major virulence factor of A. fumigatus is its ability to rapidly adapt to host conditions during infection. The rapid response to environmental changes underlies a well-balanced system of production and degradation of proteins. The degradation of specific target proteins is mediated by ubiquitin-protein ligases (E3), which mark their target proteins with ubiquitin for proteasomal degradation. Multisubunit SCF Cullin1 Ring ligases (CRL) are E3 ligases where the F-box subunit functions as a substrate-specificity determining adaptor. A comprehensive control of protein production includes global co-repressors as the conserved Ssn6(SsnF)-Tup1(RcoA) complex, which reduces transcription on multiple levels. We have identified a novel connection between protein degradation and synthesis through an F-box protein. Fbx15 can be incorporated into SCF E3 ubiquitin ligases and controls upon stress the nuclear localization of the SsnF. Fbx15 plays a critical role for A. fumigatus adaptation and is essential for virulence in a murine infection model. Fbx15 is a fungal-specific protein and therefore a potential target for future drug development.
| The ubiquitin 26S proteasome system (UPS) controls the life span of specific regulatory proteins, which are required for coordinated development, signal transduction and DNA maintenance. Target proteins are linked to ubiquitin by the sequential action of E1, E2 and E3 enzymes. A crucial step during this enzymatic cascade is carried out by E3 ubiquitin ligases, which recognize their specific substrate and catalyze the transfer of ubiquitin. SCF-complexes are multi-subunit E3 enzymes consisting of three major subunits (Cul1, Skp1 and Rbx1), which form the core enzyme and an exchangeable set of substrate-specific adaptors called F-box proteins [1,2]. The F-box domain of these adaptors is an N-terminal binding site of approximately 45 amino acids. It binds to the Skp1 linker to connect to Cul1. The human genome encodes 69 F-box proteins and defects in F-box mediated ubiquitination are associated with various diseases like diabetes, Parkinson or cancer [3–5]. Only little is known about the role of F-box proteins in virulence of fungal pathogens, though fungal F-box proteins play important roles for cellular development, transcription, signal transduction and nutrient sensing [6–8].
Aspergillus fumigatus is a soil borne, ubiquitously distributed filamentous fungus, growing on organic matter [3,9]. Besides its saprophytic lifestyle A. fumigatus also acts as an opportunistic human pathogen and causes life-threatening invasive pulmonary aspergillosis (IPA) in immunocompromised hosts. High mortality rates of up to 90% among infected patients are linked to azole resistance, the lack of new antifungals and increasing numbers of immunosuppressive therapies [9–12]. Great efforts have been conducted to identify virulence factors, which discriminate A. fumigatus from its closely related but significantly less pathogenic relative Aspergillus nidulans [13–19]. Virulence of A. fumigatus is presumably the result of a complex multifactorial network, rather than unique and sophisticated virulence factors. A. fumigatus pathogenicity is based on small infectious conidia and its ability to rapidly adapt to constantly changing conditions including high temperature, nutritional changes, hypoxia or high pH [4,20]. This is further supported by the production of secondary metabolites (SM) such as melanins, which protect from UV radiation or the immunosuppressive mycotoxin gliotoxin [21–23].
The rapid responses of A. fumigatus to environmental stressors are linked to distinct evolutionary conserved molecular mechanisms, which are often part of development regulating processes [1,2]. A recently identified important developmental regulator in A. nidulans is Fbx15, which is required for sexual and asexual development. Furthermore, Fbx15 accumulates in SCFFbx15 complexes in csn-deficient mutants of A. nidulans [3–5]. The COP9 signalosome (CSN) multi-subunit complex plays a crucial role in fruiting body formation, oxidative stress tolerance and SM production in A. nidulans [6,7,24,25]. CSN acts as deneddylase by removing the isopeptide bond of the ubiquitin-like protein Nedd8 from a lysine residue of cullin scaffolds of those E3 ligases, which are not interacting with substrate molecules for ubqituitination [3,9]. Repetitive cycles of cullin neddylation/deneddylation are especially important for development because they promote the exchange of F-box adaptors from SCF E3 ligases [9–12]. The genomes of A. nidulans or its pathogenic counterpart A. fumigatus comprise approximately 70 F-box protein encoding genes, of which three (fbx15, fbx23, grrA) have been reported to influence developmental steps in A. nidulans [4,13–18,26].
In this study we characterized the molecular function of the Fbx15 counterpart of the pathogen A. fumigatus. We were originally interested whether F-box proteins, which play a crucial role for fungal development in A. nidulans are connected to A. fumigatus pathogenicity. We found a novel dual function for F-box proteins because Fbx15 is not only part of nuclear SCF complexes but also controls the nuclear localization of the SsnF/Ssn6 component of the highly conserved eukaryotic transcriptional co-repressor complex Ssn6/Tup1. Cellular Fbx15 function is controlled by posttranslational phosphorylation and dephosphorylation during stress. Fbx15 including the F-box domain is required for cellular stress responses, the control of gliotoxin production and for virulence in a mouse model. Fbx15 is fungal-specific and might therefore be an interesting novel target for new drugs to treat invasive aspergillosis.
A. fumigatus fbx15 (Afu3g14150) corresponds to the gene of A. nidulans encoding F-box protein 15, which is required for development [4,20]. Several cDNAs of this gene locus were sequenced and revealed that the A. fumigatus fbx15 gene structure consists of two exons and one intron resulting in a deduced open reading frame of 655 codons for a protein with a predicted molecular mass of 75 kDa (Fig 1A). Alignments of the A. fumigatus Fbx15 primary sequence revealed that high similarities are restricted to the Aspergilli counterparts with similarities between 72.8% and 59.8%, whereas Fbx15-like proteins of other filamentous fungi like Penicillium chrysogenum or Neurospora crassa share significant lower similarities of 43.6% and 24.7% respectively (Fig 1B, S1 Table). Two signature motifs are located almost in the middle of Fbx15 and include motif 1, which is specific for the genus Aspergillus and motif 2 that is also present in other fungi as P. chrysogenum or N. crassa (Fig 1B, S2 Table). Bioinformatics analysis predicts two nuclear localization signals (NLS), which suggest that Fbx15 might exhibit its function in the nucleus (Fig 1B).
The function of fbx15 in A. fumigatus was genetically addressed by creating an fbx15 deletion. The mutant strain formed normal colonies on minimal medium (MM), but oxidative stress caused by 3 mM H2O2 abolished growth, whereas the wild type could still grow (Fig 1C). 1 mM H2O2, which had no impact on wild type, resulted in hyperbranched, swollen hyphae in the Δfbx15 strain (Fig 1D). On media containing superoxides-producing menadione or thiol-oxidizing diamide the Δfbx15 mutant displayed defects in growth and sporulation, which implies a general Fbx15 mediated resistance mechanism against oxidative stress, upstream of distinct stress response pathways (S1B Fig). Severe growth and sporulation defects were also observed when the Δfbx15 mutant was exposed to other stress conditions including elevated temperature, amino acid starvation, microtubule stress, osmotic stress or mutagenic stresses (S1C Fig).
The importance of Fbx15 during stress was compared to other F-box proteins, which have been previously shown to be involved in fungal development. Fbx23 is important to initiate A. nidulans asexual development [4,21–23], whereas GrrA is an F-box protein that is required for the maturation of sexual spores [26–29]. These two F-box proteins are conserved from fungi to higher eukaryotes (S1 Table). Characterization of both deletion mutants in A. fumigatus showed more distinct and less prominent phenotypes compared to the Δfbx15 mutant strain. In contrast to the wild type, deletion of fbx23 led to a decreased colony size, whereas the colony size of the ΔgrrA mutant was not affected under normal growth conditions (Fig 1C).
Growth defects of Δfbx23 and ΔgrrA in comparison to wild type were observed under amino acid starvation or microtubule stress. Both deletion strains were sensitive to topoisomerase I inhibitor CPT, whereas DNA-methylating MMS had a similar impact on wild type and mutant growth (S1C Fig). These results indicated that Fbx23 as well as GrrA are involved in the DNA replication process, whereas Fbx15 is additionally required for the repair of DNA-damage caused by methylation. Growth under increased temperature or oxidative stress conditions are important virulence factors of A. fumigatus [30]. The deletion of fbx23 led to a reduced colony size under high temperature (42°C) or oxidative stress, when compared to the wild type, whereas the ΔgrrA strain was only slightly affected by H2O2 (Fig 1C, S1C Fig).
These data suggest that the three F-box proteins Fbx23, GrrA and Fbx15 are part of a general stress response in A. fumigatus caused by multiple environmental stressors. The impact of the strain lacking Fbx15 suggests that this protein plays a key role in the fungal stress response.
Oxidative stress caused drastic effects on the Δfbx15 mutant, whereas fbx15 overexpression resulted in a wild type like phenotype (S1A Fig). Oxidative stress, in particular by peroxides like H2O2, leads to an elevated expression of catalases, which function as potent ROS scavenger [31,32]. A. fumigatus produces one conidial and two mycelial catalases. The expression of cat1, which encodes one of the mycelial catalases, was regulated in an Fbx15 dependent manner. The lack of fbx15 led to an approximately 2.5x derepressed basal cat1 expression, compared to wild type or a complemented strain during non-stress conditions (Fig 1E). After exposure to H2O2, the expression of cat1 in the Δfbx15 mutant was further increased, whereas wild type or complemented strain displayed only slight upregulation. Increased cat1 levels were not sufficient to protect the Δfbx15 mutant from oxidative stress. Fbx15 presumably is involved in additional molecular mechanisms, which protect against oxidative stress. The exposure to oxidative stress was analyzed as a possible external signal, which triggers changes in fbx15 expression in A. fumigatus. Fungal cultures were exposed to H2O2 and harvested at different time points within a 120 min period. fbx15 transcript levels were determined with real-time PCR (RT-PCR). A rapid increase of fbx15 expression was observed in the first 20 min reaching its maximum peak at 40 min with a 14-fold increased gene expression. Afterwards the expression decreased to a basal level, which is approximately 4-fold increased compared to the non-induced expression. Proteins from the same samples were extracted to test whether the changes in fbx15 transcript levels are also reflected on protein level. Fbx15 was visualized after immunoblotting by incubation with a polyclonal Fbx15 specific antibody. The general abundance of Fbx15 was very low. A three-fold increase of Fbx15 protein amounts was measured, starting after 40 min of H2O2 exposure, which was similar in the complemented strain and absent in the Δfbx15 mutant, mirroring the increased gene expression on the protein level with a delay of 20 min (Fig 1F, S1D Fig). These data suggest that the Fbx15 protein levels are increased as part of a stress adaption response towards H2O2 mediated ROS.
Two putative nuclear localization sites of Fbx15 suggest a nuclear function. A similar situation is found for the conserved A. fumigatus F-box protein SconB, which also possesses two nuclear localization signals, and was used as control (S1 Table). Similar to sconB of A. nidulans, sconB is essential for A. fumigatus as we could show with heterokaryon rescue assay (S2A and S2B Fig). Constitutively expressed GFP fusions of Fbx15 and SconB were compared for their subcellular localizations. Both F-box GFP fusions were functional during normal and oxidative stress growth conditions (S2C Fig). Fbx15 and SconB are primarily co-localized in fluorescence microscopy with DAPI stained nuclei, though small subpopulations of both proteins remain in the cytoplasm (S2D Fig). These data suggest a predominant nuclear function for both A. fumigatus F-box proteins.
Fbx15 is primarily nuclear but has a significant cytoplasmic subpopulation. Different localizations of Fbx15 could be a result of controlled posttranslational phosphorylation. A bioinformatic analysis of the deduced amino acid sequence of Fbx15 with NetPhos 2.0 (http://www.cbs.dtu.dk/services/NetPhos) [33–35] predicts in total 15 serine, 11 threonine and 4 tyrosine residues as putative phosphorylation sites (score value between 0 and 1; cutoff value >0.5: S3A Fig). An Fbx15 phosphopeptide with a single phosphorylation was identified with mass-spectrometry of purified functional Fbx15-TAP fusions from non-stressed cultures. Analysis of the MS2-spectra of this phosphopeptide with the phosphoRS software [36] revealed serine residues 468 and 469 as potential phosphorylation sites with probabilities of 45.5% each, whereas Ser473 only showed a low probability of 8.9% (Fig 2A). Theoretical analysis for Fbx15 phosphosites showed a high score value of 0.988 for Ser469, whereas Ser468 had a low score value of 0.029 and therefore is unlikely to be phosphorylated (S3A Fig). In contrast no phosphopeptides were identified for purified Fbx15-GFP fusions when the cells were grown under oxidative stress. Fbx15 is presumably phosphorylated during vegetative growth under non-stress conditions at Ser468 or Ser469 (Ser468/469), whereas it is unphosphorylated when H2O2-mediated oxidative stress is applied.
Phosphorylation of Fbx15 under normal growth conditions and dephosphorylation during H2O2-mediated oxidative stress suggests the presence of a phosphatase that might be specifically activated. GFP-traps with Fbx15-GFP recruited GlcA as the only phosphatase that interacts in cultures with or without stress (S3B Fig). A direct interaction of Fbx15 with GlcA could be observed using bimolecular fluorescence complementation (BiFC), which was primarily visible after induction with H2O2 (S3C Fig). The A. nidulans GlcA homolog BimG has been characterized as essential major protein phosphatase 1, which is associated with thermo tolerance and hyphal morphology, features that were impaired in the Δfbx15 mutant [37,38]. Heterokaryon rescue experiments with a glcA deletion cassette and accompanying Southern-hybridizations verified that the situation is similar and glcA is essential for A. fumigatus as well (S3C and S3D Fig).
The dephosphorylation rate of Fbx15 in response to H2O2 was quantified. Fbx15-GFP expressing strains were grown in liquid cultures and subjected to oxidative stress by adding 3 mM H2O2. Fbx15-GFP from these cultures was purified and treated with an antibody against phosphorylated Ser/Thr residues. The rate of dephosphorylation was quantified against the overall amount of Fbx15-GFP, determined by an anti-GFP antibody (Fig 2B). 40% dephosphorylation upon H2O2-treatment indicated that Fbx15 becomes dephosphorylated in an oxidative stress-dependent manner. The Fbx15 dephosphorylation sites were localized by TMT isobaric mass tag labeling and LC-MS/MS [39]. Fbx15-GFP was enrichment from cultures before and after H2O2 treatment and separated on a coomassie-stained SDS-PAGE. After tryptic digestion, peptides from the untreated culture were labeled with TMT127 (heavy) whereas TMT126 (light) was used for all time points of the H2O2 treated cultures. The sample of time point zero was individually mixed with all other time points and analyzed by LC-MS/MS. The parameters for fragmentation during mass spectrometry were set to identify only the phosphorylated peptide of Fbx15. Specific ratios of the heavy labeled phosphopeptide were obtained from time point zero against the light labeled phosphopeptides from the other conditions. These values were quantified against the ratios of two unmodified reference peptides of Fbx15, which represented the overall amount of purified Fbx15. The reciprocal values of these ratios are decreasing, thus reflecting a specific dephosphorylation on Ser468/469 (Fig 2C). Fbx15, which becomes phosphorylated at Ser468/469 under non-stress conditions, is presumably dephosphorylated when cells encounter H2O2-mediated oxidative stress by the Fbx15 interacting phosphatase GlcA/BimG.
The canonical function of F-box proteins is their ability to form ubiquitinating SCF ligase complexes by binding to the SkpA adaptor. A. fumigatus Fbx15 and SconB interactions to the SkpA SCF adaptor were compared by using BiFC. Both Fbx15 and SconB protein fusions produced an YFP-signal, indicating F-box protein-SkpA interactions. SconB interacted with SkpA almost exclusively in the nucleus (>90%), whereas 74% of the Fbx15-SkpA interaction was cytoplasmic (Fig 2D and S4A Fig). The Fbx15 serine codons of wild type positions S468 and S469 were replaced to alanine residues to mimic a constantly dephosphorylated Fbx15[S468|9A] to analyze whether Fbx15 phosphorylation is relevant for the location of the Fbx15-SkpA interaction. Interaction of Fbx15[S468|9A] variant with SkpA resulted in a nuclear signal (Fig 2D). Dephosphorylated Fbx15 primarily interacts with SkpA in the nucleus to form SCF-complexes. Growth without stress rather results in phosphorylated Fbx15 with presumably only limited amounts of dephosphorylated Fbx15 in the nucleus.
Dephosphorylation of Fbx15 is triggered by H2O2-mediated ROS. The impact of the phosphorylation state of Fbx15 on its ability to interact with SCF-complexes was analyzed by replacing the serine residue S469 in Fbx15 with aspartate to mimic a constant phosphorylation Fbx15[S469D] or S468 and S469 with alanine to mimic unphosphorylated Fbx15[S468|9A]. Both constructs and the wild type gene were expressed under the native promoter as Fbx15-RFP fusions. Immunoblotting confirmed that all Fbx15 versions were more abundant after H2O2 exposure (Fig 3A).
RFP-trap co-purifications followed by LC-MS/MS with wild type Fbx15 or phosphomutant variants after oxidative stress induction with H2O2 were performed to identify which subunits of the SCF-ligase machinery interact with Fbx15. Further analysis included MaxQuant quantitative proteomic software in conjunction with Perseus software for statistical analysis, with a focus on the subunits of the SCF-ligase machinery. SCFs are activated by the RING protein RbxA-mediated interaction with E2 ubiquitin-conjugating enzyme and covalent cullin modification by ubiquitin like NeddH (Fig 3B). SCF core components SkpA or CulA interacted with Fbx15, independently of the phosphorylation-state. Native Fbx15, which is presumably dephosphorylated at S468/S469 after oxidative stress or unphosphorylated Fbx15[S468|9A] could co-purify NeddH, but not RbxA or an E2 enzyme. In contrast the Fbx15, mimicking a constant phosphorylation, co-purified all subunits for an active SCF complex. NeddH was more abundant in co-purifications of the constantly phosphorylated Fbx15[S469D] than of unphosphorylated versions of Fbx15. RbxA and the E2 enzyme UbcM were only purified with negatively charged Fbx15[S469D], indicating an improved assembly of functional SCF-ligase complexes when Fbx15 is phosphorylated at Ser469 (Fig 3C). Active SCFFbx15 complexes contain therefore more likely phosphorylated than unphosphorylated Fbx15.
Since Fbx15 interacted within SCF complexes the cellular ubiquitination pattern of the Δfbx15 mutant, wild type and the fbx15 overexpression strain were compared before and after induction with H2O2. Neither the ubiquitination-pattern nor the general protein composition of the Δfbx15-strain was significantly altered in comparison to the wild type or the fbx15 overexpression strain, suggesting that the ubiquitination targets of SCFFbx15 complexes are limited (S4B and S4C Fig).
Additional potential interacting proteins for Fbx15 were identified by tandem-affinity-purification (TAP) and compared to co-purifications of SconB-TAP fusions. Variants of fbx15 and sconB were created where the conserved proline of the F-box domain was exchanged to a serine. This should weaken the F-box-SkpA binding and enable recruitment of SCF-independent interaction partners. The exchanged proline in SconB led to significantly increased protein stability, whereas stability of Fbx15 was not altered (S5 Fig). This reflects a possible autocatalytic mechanism for SCFSconB, where SconB is ubiquitinated within its own SCF ligase and eventually degraded [40,41], whereas Fbx15 stability seems to be independent of SCFFbx15. SconB-TAP recruited less proteins (22) than Fbx15 (38). SconB interactors include transcriptional activator MetR as known SCFSconB target [1] and 11 proteins that were identified for both F-box proteins including SCF subunits CulA and SkpA. Only Fbx15 was able to co-purify three subunits of the CSN deneddylase, which acts on neddylated cullin complexes, which do not interact with substrates [3]. This might reflect a highly dynamic assembly/disassembly of SCFFbx15 complexes (S3 Table).
The predominant nuclear localization of Fbx15 and SconB is consistent with nuclear interaction partners, which were identified during TAP co-purifications. These included two transcriptional regulators (RcoA/Tup1 and a putative APSES transcription factor), a DNA repair enzyme (AFUA_2G06140) and a single-stranded DNA binding protein (AFUA_5G07890/Rim1p) (Fig 4A). The fact that Fbx15 and SconB recruited these proteins might reflect a tight stability control by more than one F-box protein. In addition Fbx15 recruited specifically three transcriptional regulators (OefC, SrbB and SsnF/Ssn6), a nuclear GTPase (AFUA_4G8930/Nog2p) and the nuclear pore protein Nic96. (Fig 4A). A potential candidate, responsible for Fbx15 phosphorylation is the interacting cyclin-dependent serine/threonine kinase NimX/Cdc28p, which is required for cell cycle control and conidiophore morphology in A. nidulans [6]. A direct interaction of Fbx15 and NimX could be verified by BiFC assay, where the interaction signal from the reconstituted YFP was observed predominantly in the cytoplasm (S6A Fig). However, the NimX homolog in A. fumigatus is presumably encoded by an essential gene as we could show by heterokaryon assay and Southern hybridization (S6B and S6C Fig).
Fbx15 and SconB interaction partners include possible targets for SCF mediated ubiquitination. Three Fbx15 interacting CSN subunits suggest a more dynamic assembly/disassembly of SCFFbx15 in comparison to SCFSconB. An interesting finding is that both F-box proteins co-purified RcoA/Tup1 as part of the conserved transcriptional co-repressor complex RcoA/Tup1-SsnF/Ssn6, but only Fbx15 recruited the SsnF/Ssn6 subunit of this complex.
The yeast Ssn6-Tup1 co-repressor complex affects the expression of 7% of all genes with emphasis on stress responses [9]. A homo-tetramer of RcoA/Tup1 repressor subunits is connected to one SsnF/Ssn6 adaptor protein, which binds to a DNA-binding protein, escorting the repressor complex to the target genes. The corresponding SsnF encoding gene of the model A. nidulans is essential for growth, whereas the yeast counterpart Ssn6 is dispensable [9,42]. With heterokaryon rescue and subsequent Southern analysis we could show that ssnF is not only essential for A. nidulans but also for A. fumigatus (S6D and S6E Fig). A BiFC signal verified the direct interaction of Fbx15 and SsnF in A. fumigatus under non-stress conditions in the cytoplasm, often located close to nuclei. Unphosphorylated Fbx15[S468|9A] which reflected H2O2-mediated oxidative stress conditions interacted with SsnF predominantly in the nucleus, similar to the Fbx15-SkpA interaction (Fig 4B). Fbx15 levels in different strain backgrounds did not influence SsnF stability in cycloheximide (CHX) protein-stability assays (Fig 4C). The amount of SsnF-GFP in wild type or fbx15 deletion mutant did not change in the absence or presence of 3 mM H2O2 nor did SsnF exhibit an Fbx15 dependent ubiquitination pattern (S6F and S6G Fig). This suggests that SsnF is not a significant substrate of an active SCFFbx15 E3 complex but acts as additional physical Fbx15 interaction partner.
SsnF-GFP is localized in the nucleus in the presence of Fbx15, but accumulates at the nuclear envelope presumably enriched at nuclear pore complexes in the absence of Fbx15. This suggests that SsnF import is impaired without Fbx15 (Fig 5A). It was analyzed whether dephosphorylation of Fbx15 is involved in SsnF nuclear localization. In fbx15 wild type as well as in the fbx15 variant strain (Fbx15[S469D]), that mimics constant phosphorylation, SsnF-GFP was localized in the nucleus. In the fbx15 codon replacement mutant expressing the Fbx15[S468|9A] variant, which cannot be phosphorylated, SsnF accumulated at the nuclear envelope similar to the Δfbx15 strain. Correct nuclear SsnF-GFP localization could be either abolished after Fbx15 wild type dephosphorylation due to H2O2 or in the presence of the Fbx15[S468|9A] variant, which cannot be phosphorylated. In contrast, oxidative stress did not interfere with SsnF-GFP nuclear localization, when only the negatively charged Fbx15[S469D] was present, which mimics constant phosphorylation (Fig 5B).
These data support that phosphorylation of Fbx15 under non-stress conditions favors nuclear localization of SsnF, whereas Fbx15 dephosphorylation during H2O2-mediated oxidative stress leads to an accumulation of SsnF at the nuclear envelope.
The contribution of the Fbx15[S468|9A] variant, which cannot be phosphorylated and interferes with nuclear SsnF, to the fungal oxidative stress response was analyzed. Growth tests on different oxidative stress providing media showed that mutant strains expressing the non-phosphorylable Fbx15[S468|9A] as well as the phosphate mimicking Fbx15[S469D] showed mild colony growth reductions in comparison to the wild type version of Fbx15 (Fig 5C). Similar effects where observed on medium containing superoxide-producing menadione, where the phosphomutant strains showed a slight growth reduction compared to the RFP-tagged wild type Fbx15. Growth on thiol oxidizing media was not affected (S7 Fig). These data support that Fbx15 phosphorylation and dephosphorylation and the control of cellular SsnF localization contribute to an appropriate oxidative stress response.
The function for oxidative stress resistance of the F-box domain of Fbx15 as link to E3 ubiquitin ligases was examined. A gene for an RFP-tagged Fbx15 variant, that lacks the N-terminal F-box domain, and the two additional combinations with the alternative phosphovariants were constructed and the resistance against oxidative stress was tested. Loss of the F-box in all three strains completely abolished growth on media containing H2O2, similar to the Δfbx15 mutant (Fig 5C, S7 Fig).
These results indicate that the F-box, which is required to assemble Fbx15 into SCF ubiquitin ligases, is essential for the fungal oxidative stress response. The function of the oxidative stress controlled phosphorylation status of Fbx15, which channels SsnF nuclear localization, is presumably part of an additional fine-tuning of the appropriate cellular oxidative stress response.
A group of genes, which are usually repressed during normal fungal growth, are secondary metabolite genes. Defects in the CSN-regulated ubiquitination machinery result in a drastic misregulation of secondary metabolite formation, such as mycotoxins [13]. A potent immunosuppressive mycotoxin in A. fumigatus is gliotoxin, which is considered as one of multiple virulence factors [20].
It was analyzed, whether Fbx15 is part of the transcriptional repression of gliotoxin synthesis genes, because SsnF, which represents a part of a general conserved transcriptional repression mechanism, is accumulated at the nuclear periphery in the Δfbx15 mutant. We analyzed the expression of gliZ, which encodes a transcriptional activator that has been shown to drive the expression of gli-genes encoded in the gliotoxin gene cluster and gliP, encoding the non-ribosomal peptide synthetase GliP with a key role in gliotoxin biosynthesis [21,23]. The expression of gliZ and gliP in the Δfbx15 mutant increased by almost thirteen and five times respectively in comparison to wild type or complemented strain. Gliotoxin is a toxic metabolite for A. fumigatus itself. Expression patterns of gli genes, which are important for detoxification mechanisms, were examined. These include gliK, which is required for gliotoxin biosynthesis and secretion or gliT encoding a oxidoreductase with the ability to reversibly form the toxic disulphide bond of gliotoxin [27–29]. Compared to wild type the gliK and gliT mRNA levels increased significantly in the Δfbx15 mutant by three and twelve times, respectively (Fig 6A).
Gliotoxin production of the Δfbx15 mutant was determined to analyze whether increased transcription of genes involved in gliotoxin production and detoxification correlates to a change in secondary metabolism. High-performance liquid chromatography (HPLC) revealed that the gliotoxin production of Δfbx15 was increased by 3-fold compared to wild type (Fig 6B).
The phosphomutant versions of Fbx15, which mediate different localization of SsnF in or outside the nucleus, were included into the analysis of the transcription of gliotoxin biosynthetic and protecting genes. The fbx15 variant that produces the non-phosphorylatable Fbx15[S468|9A], which leads to cytoplasmic accumulation of SsnF, showed similarly increased gli gene transcript levels as the Δfbx15 deletion strain for gliZ and gliP and a moderately increased expression of gliK and gliT. In contrast, the Fbx15[S469D] variant mimicking constantly phosphorylation of Fbx15 and supporting nuclear localization of SsnF, resulted in a significantly lower expression of gli gene transcript levels as the non-phosphorylatable Fbx15[S468|9A] (Fig 6C).
Gliotoxin production levels of strains expressing either variant of unphosphorylated Fbx15[S468|9A] or phosphorylation mimicking Fbx15[S469D] were in a similar range as wild type and significantly lower than the Δfbx15 deletion strain (Fig 6D).
These results suggest that Fbx15 is required for repression of gli gene expression. The transcription of gli genes is derepressed and in addition the gliotoxin production is increased in the absence of Fbx15, when SsnF accumulates at the nuclear periphery. The transcription of gli genes is also increased in the presence of an unphosphorylatable Fbx15, which excludes SsnF from the nucleus, but gliotoxin production is not increased. This suggests an additional function of Fbx15 at a posttranscriptional layer of control for gliotoxin synthesis, which is independent of its phosphorylation status.
An established murine model of invasive pulmonary aspergillosis (IPA) was used to analyze whether Fbx15 mediated stress response and gliotoxin production control affect fungal virulence in comparison to wild type, Δfbx23 and ΔgrrA strains. Immunosuppressed mice infected with wild type, Δfbx23, ΔgrrA or complemented strains displayed normal mortality rates within 14 days, although the Δfbx23 strain displayed a slightly increased virulence (p = 0.03) in direct comparison to the wild type. In contrast, the Δfbx15 mutant completely lost its virulence (Fig 6E). The Δfbx15-infected mice did not show any symptoms and had the same clinical appearance as the mock infected control group, treated with phosphate buffered saline (PBS). Histopathology analyses of infected lung tissue were consistent with survival rates. Mice infected with either wild type or complemented strains showed fungal hyphae surrounded by tissue necrosis and extensive immune cell infiltration (Fig 6F). Moderate immune cell infiltrates were found in the lungs of mice infected with the Δfbx15 mutant, but no fungal hyphae could be detected. The fungus was cleared at an early stage of infection, probably by innate immune responses and increased temperature and elevated oxidative stress. Our data suggests that Fbx15, which is not required for growth without stress, plays a crucial role during infection because it enables A. fumigatus to adapt to innate immune response conditions of the host including limiting nutrition, fever or oxidative stress.
Aspergillus fumigatus is the most prevalent cause for pulmonary infections in immunocompromised patients. High thermo- and oxidative stress tolerance, toxic metabolites and a versatile metabolism allow A. fumigatus to colonize host tissue [30]. We identified the fungal-specific F-box protein Fbx15, which is not required for vegetative growth in the absence of stress, as key determinant for stress response, controlled gliotoxin production and virulence. A novel dual molecular function was discovered for Fbx15. Fbx15 can be part of an SCF E3 ubiquitin ligase complex and in addition controls nuclear localization of SsnF as transcriptional repressor subunit. Fbx15 levels are transcriptionally regulated and Fbx15 location in either the nucleus or the cytoplasm is determined by phosphorylation and dephosphorylation, respectively. Fbx15 is a potential target for antifungal drugs, because it is essential for A. fumigatus virulence.
Our data demonstrate that Fbx15 plays a crucial role for adaptive responses to environmental changes and general stress response mechanisms in A. fumigatus whereas during non-stress conditions fbx15 is dispensable for normal growth. This correlates with the expression patterns of fbx15 gene transcription and translation, which demonstrated that Fbx15 becomes only abundant during oxidative stress induction. Whether this behavior is similar for different stressors remains to be shown. The low expression levels of fbx15 during non-stress conditions are is required for the nuclear localization of the transcriptional co-repressor subunit SsnF (Fig 7). The best known example of this repressor complex, conserved in eukaryotes, is yeast Ssn6(Cyc8)-Tup1. It affects expression of at least 334 genes during normal growth conditions, which include developmental, metabolic or stress response pathways [31]. Ssn6 acts as an adaptor between a tetramer of Tup1 and additional DNA-binding proteins which mediate sequence specificity [33,35]. Tup1 alone is able to promote transcriptional repression in yeast, which might explain the ability of the A. fumigatus Δfbx15 mutant to grow under normal conditions, where nuclear SsnF is compromised.
The oxidative stress response gene cat1, which encodes a mycelial catalase, represents a target gene, which is repressed in a Fbx15-dependent way [32,43]. Expression of cat1 is increased in the absence Fbx15, but this derepression is not sufficient for an appropriate oxidative stress response [5,44,45]. Oxidative stress results in Aspergillus nidulans in genome-wide transcriptional changes, rather than a specific response of distinct gene groups [7,8,46]. Derepression of cat1 as well as of several gliotoxin biosynthetic genes correlates with the mislocalization of the nuclear repressor subunit SsnF in the cytoplasm in the absence of Fbx15. The control of nuclear SsnF localization by the phosphorylation status of Fbx15 contributes but is not the only cause of an appropriate oxidative stress response.
The F-box domain of Fbx15 is essential for an appropriate oxidative stress response. This domain serves as binding site for the assembly into SCF ubiquitin ligases. This indicates additional SCF-dependent functions for Fbx15 during oxidative stress. Disassembly of those E3 SCF ubiquitin ligases, which are not binding target substrates for ubiquitination, requires their inactivation by the COP9 signalosome CSN [3,47]. The csn-deficient mutant strains of A. nidulans are impaired in this control of cellular ubiquitin ligase activities and show a severe growth phenotype in the presence of oxidative stress [11,25,48,49]. Impaired CSN activity in csn-deficient mutant strains resulted in the enrichment, isolation and identification of SCF-Fbx15 complexes in A. nidulans [17]. This suggests that the F-box mediated assembly of SCF-Fbx15 as well as the disassembly by CSN are crucial for the function of Fbx15 during oxidative stress.
Yeast Tup1 corresponds to Aspergillus RcoA and to Groucho/TLE of higher eukaryotes. Multiple repression mechanisms are associated to these complexes such as histone deacetylation, chromatin rearrangements, modification of RNA polymerase II activity and competition with transcriptional activators [14–19,50–52]. Repression can even be changed to activation as shown for the transcriptional repressor Sko1, which inhibits hyperosmotic stress response genes in conjunction with Ssn6-Tup1. Osmotic stress leads to the phosphorylation of Sko1, which turns it into a transcriptional activator that in conjunction with Tup1 recruits chromatin-remodeling complexes such as SAGA and SWI/SNF to the respective promoter sites. ATP-driven chromatin remodeling mediated by the SWI/SNF complex is further required for cell cycle progression and activation of DNA damage repair pathways [53–55]. The lack of appropriate molecular derepression or activation of stress response genes during stress contributes to the observed serious defects in fungal growth in the A. fumigatus Δfbx15 mutant. There seems to be an additional posttranscriptional layer of control, which does not depend on the phosphorylation status of Fbx15. A block of nuclear SsnF in the presence of unphosphorylated Fbx15[S468|9A] results in increased transcription of gli genes but not increased gliotoxin production and has only a partial contribution for the resistance towards oxidative stress.
Rapid Fbx15 dephosphorylation during oxidative stress could be triggered by the essential protein phosphatase 2A catalytic subunit GlcA/BimG (Fig 7). GlcA belongs to the serine/threonine phosphatases and shares homology with the yeast protein phosphatase 1 (PP1) catalytic subunit Glc7. GLC7 is essential for yeast as well, but conditional mutant alleles of GLC7 could be connected to defects in adaptive functions like temperature tolerance, glucose repression, amino acid starvation, cell morphology and DNA damage repair, which are reminiscent to the growth defects of the Δfbx15 mutant on the respective conditions [4,22,56–59]. Dephosphorylation of Fbx15 results in nuclear clearance of SsnF. This could be due to nuclear export combined with reduced import of SsnF and/or ubiquitin dependent or even independent degradation of SsnF. Nuclear trafficking control is supported by the observation that SsnF is blocked at NPCs in an Fbx15-dependent manner upon stress. In this context a potential ubiquitinating function of Fbx15 towards the nuclear pore complex (NPC) subunit Nic96, which was co-purified during our TAP-tag pull-downs, was a reasonable function for SCFFbx15-ligase complexes to promote nuclear transport control. And although the localization of SsnF-GFP in Δfbx15 background and the localization of Nic96-GFP shared some similarities, we could not observe an Fbx15-specific or oxidative stress-dependent ubiquitination pattern for Nic96 (Fig 5A, S8 Fig). However, the nuclear pore complex is a massive multi-protein complex composed of 30 different NPC-proteins, which are arranged in multiples and finally reach a molecular mass between 66 and 125 MDa [2,60]. In 2012 Hayakawa et al. showed that approximately half of the NPC-proteins in yeasts are ubiquitinated, but not necessarily targeted for proteasomal degradation [4,61]. It might be possible that Fbx15 plays a role in NPC-protein ubiquitination, which targets NPC-proteins in close proximity to Npc96 and thereby promotes a more general nuclear transport control.
Nuclear clearance of SsnF may also be triggered by selective ubiquitin-independent degradation of nuclear SsnF populations. Previous studies have shown that some E3-ubiquitin ligases directly interact with the regulatory particle of the proteasome and thus are able to transfer target proteins to the proteasome for degradation [24,25,62]. A similar scenario could be responsible for selective nuclear degradation of SsnF, where dephosphorylated Fbx15 incorporates into inactive SCF-core complexes, which have the potential to carry specific target substrates such as SsnF directly to the proteasome.
Phosphorylated or dephosphorylated Fbx15 interacts with SsnF at different cellular localizations. However, Fbx15 and SsnF are not present in stoichiometric amounts, which argues against stable Fbx15-SsnF complexes. Phosphorylated Fbx15 interacts with SsnF under non-stress conditions predominantly in the cytoplasm, suggesting a cargo function for Fbx15, which facilitates the nuclear import of Fbx15-SsnF heterodimers. Fbx15 protein levels are increased during stress and Fbx15 is dephosphorylated. Unmodified Fbx15 interacts with SsnF primarily in the nucleus and might compete with Tup1 interaction. In addition, Fbx15 possibly exhibits a nuclear export function by acting as a cargo receptor for SsnF export.
Similar to the interaction with SsnF, Fbx15 interaction with the adaptor protein SkpA, which bridges Fbx15 into SCF complexes was not disturbed due to the dephosphorylation of Fbx15. However, the interaction was shifted from cytoplasm to the nucleus. The Fbx15 phosphorylation site Ser468/469 is located between two NLSs. Therefore phosphorylation/dephosphorylation events on Fbx15 Ser468/469 might determine nuclear Fbx15 import or export by rearranging the NLS availability.
Fbx15 phosphorylation affects SsnF location. Disturbed localization patterns of SsnF in constantly unphosphorylated fbx15 mutants resulted in moderate phenotypes, whereas the additional deletion of the F-box domain of Fbx15 led to impaired oxidative stress resistance similar to the complete fbx15 deletion mutant. This suggests that Fbx15 as subunit of the SCFFbx15 complex is an important player in the stress tolerance mechanism of A. fumigatus. This additional SCF-dependent function for Fbx15, apart from SsnF localization control, is also supported by the fact that only complete fbx15 deletions led to increased gliotoxin levels in the mutant, whereas constantly unphosphorylated fbx15 mutants showed increased gli-gene expression, which did not lead to increased gliotoxin levels. The formation of active SCFFbx15 complexes was especially promoted in fbx15 mutants, which mimic a constant phosphorylation, indicating an ubiquitinating function of phosphorylated Fbx15-carrying SCF ligases during non-stress conditions. Fbx15 abundance under non-stress conditions is very low and overall ubiquitin-patterns of the cellular pool of proteins were not significantly changed between wild type and Δfbx15 mutants. This suggests that putative target(s) of SCFFbx15 are highly specific.
The role of F-box proteins in virulence might vary in different pathogenic fungi. A. nidulans ΔgrrA mutants are unable to produce mature ascospores due to a block in meiosis [3,26]. We showed here that the deletion of grrA in A. fumigatus did not affect virulence. GrrA shares structural similarity to Fbp1 of the opportunistic human pathogen Cryptococcus neoformans and deletion of fbp1 led to a loss of virulence [10,12,63]. In contrast to Fbx15, Fbp1 is not involved in a broad range of stress responses but plays a more specific role in cell membrane integrity where the loss of fbp1 results in a block in meiosis, which finally leads to an impaired sexual sporulation, similar as GrrA in A. nidulans.
Fbx15 is a developmental regulator in the model organism A. nidulans, where the deletion of fbx15 results in a complete block in sexual and asexual development [4,26]. The deletion of the Tup1 homolog rcoA in A. nidulans leads to a phenotype very similar to A. nidulans Δfbx15 mutant strains, which are blocked in developmental pathways and secondary metabolism [4,64]. Thus Fbx15 mediated localization control of SsnF might be a conserved mechanism in filamentous fungi. However, fbx15 deletion mutants in the opportunistic pathogen A. fumigatus did not display a developmental defect, but Fbx15 emerges as key regulator for stress response and virulence. Several virulence factors of A. fumigatus, like mycotoxin production, oxidative stress resistance and nutritional versatility are linked to developmental control mechanisms, which were identified in the apathogenic model organism A. nidulans. This example of a connection between developmental regulators of non-pathogenic fungi and their role for virulence in fungal pathogens might be an interesting paradigm for future approaches to identify novel so far unknown virulence determinants.
A. fumigatus Fbx15 is specific to filamentous fungi, which provide the opportunity for drug design. The treatment of invasive aspergillosis is still primarily based on aggressive and toxic antifungal drugs. This disadvantage is aggravated by increasing numbers of A. fumigatus species, with resistances against commonly used medical triazoles [4,65–67]. Fbx15 might be a potential drug target, excluding the risk of cross-reactions with human proteins. In contrast to novel drugs, which target the general ubiquitin proteasomal machinery by inhibiting their core components, such as Nedd8-activating enzymes, the SCF-adaptor Skp1 or the proteasome, and thus providing therapeutic chances for cancer, neurodegenerative diseases and immune deficiencies, a drug against Fbx15 would not affect the ubiquitin-proteasome system itself, but instead offer a highly specific inhibitor for fungal dissemination during life threatening aspergillosis [5,26,68,69]. So far promising drugs, targeting specific F-box proteins have been identified for human F-box proteins, which are connected to a diverse set of cancers [70]. The possibility to treat fungal diseases with F-box specific inhibitors is supported by Lobo et al., who discovered the plant defensin Psd1 from Pisum sativum that interacts with the nuclear F-box protein cyclin F of N. crassa and exhibits antifungal activity against several Aspergillus species [71]. A putative drug against Fbx15 might have a highly specific fungal target spectrum, because the function of Fbx15 varies between a stress response regulator in A. fumigatus and a developmental regulator in A. nidulans. Fbx15 might bear the potential to identify new virulence determining factors, which can be used for advanced drug design. Our identification of further Fbx15 interaction partners provides a promising base for the characterization of other novel virulence traits in A. fumigatus.
Taken together, Fbx15 is a crucial regulator for stress response and virulence in A. fumigatus, which provides the known function of an F-box protein, interacting with the ubiquitin SCF E3 ligase machinery. As a second function it controls the nuclear localization of the transcriptional co-repressor SsnF, which is part of a broad transcriptional network, including histone modifications. The broad impact of Fbx15 on stress responses and the fact that it is specific for the fungal kingdom makes this protein an interesting target for drug development.
The generation of deletion-, complementation- and tagged strains was carried out in the ΔakuA-strain AfS35 a derivative of WT-strain D141, which provides high levels of homologous recombination [30,72,73]. BiFC was done in pyrG1-strain Af293.1 (FGSC#1137), obtained from the Fungal Genetics Stock Center [32,74]. A. fumigatus strains were cultivated in minimal media (MM) with appropriate supplements. For cloning techniques E. coli strains DH5α and MACH-1 (Invitrogen) were applied. Fungal and bacterial transformations were carried out as described [34,75]. Plasmids and A. fumigatus strains are given in Tables S4/S6 and S1 Text.
Total RNAs were extracted with “RNeasy plant mini kit” (Qiagen). 0.8 μg RNA was transcribed into cDNA using “QuantiTect reverse transcription kit” (Qiagen). Gene expressions were measured with quantitative real-time PCR using either a Light Cycler 2.0 System (Roche) with “RealMasterMix SYBR ROX 2.5x” (5Prime) or a CFX Connect Real-Time System (Bio-Rad) with “MESA GREEN qPCR MasterMix Plus for SYBR Assay” (Eurogentec). Histone (h2A) and Glyceraldehyde-3-phosphate dehydrogenase (gpdA) expression were used as reference for relative quantification. Details about used cDNA concentrations and primer pairs are given in S5 Table and S1 Text.
For Immunoblotting experiments 100–150 μg crude protein extract was separated by SDS-PAGE and transferred to a nitrocellulose membrane by electro blotting as described previously [36,76]. Antibodies used for detection of fusion proteins are described in S1 Text. Signals were detected by enhanced chemiluminescence technique with either an Amersham Hyperfilm-P (GE Healthcare Limited) or with the Fusion SL7 system (Peqlab). For signal quantification Bio 1D imaging software (Peqlab) was used. Protein stability was measured by the addition of 25 μg/ml cycloheximide prior to protein extraction.
Co-purification with TAP-tagged Fbx15, Fbx15(P12S), SconB and SconB(P200S) was performed with a modified version of the Tandem Affinity Purification protocol as described in S1 Text.
Immunoprecipitation of GFP- or RFP-tagged proteins were performed with “GFP-trap_A and “RFP-trap_A (chromotek). Proteins were extracted from 5 ml frozen pulverized mycelium. 5 ml of protein crude extracts were incubated with 40 μl of GFP-Trap_A or RFP-Trap_A agarose, which was previously equilibrated to the B300-buffer. After two hours of incubation agarose was washed twice with B300 buffer. The agarose was boiled in 100 μl 3x SDS-sample buffer to elute the bound proteins. The extracted proteins were used directly for SDS-PAGE followed by immunoblotting or coomassie-staining and tryptic digestion for LC-MS/MS analysis.
Fbx15-GFP was purified from cultures before and after treatment with 3 mM H2O2 and run on an SDS-PAGE. After coomassie-staining the proteins were in-gel digested with trypsin. The purified peptides were labeled with an isobaric mass tag using the “TMTduplex Isobaric Mass Tagging Kit” (Thermo Scientific), where Fbx15-GFP before H2O2 treatment was labeled with heavy TMT-127 and all time points after H2O2 induction were labeled with TMT-126 (for details see S1 Text).
Proteins were digested with “Sequencing Grade Modified Trypsin” (Promega). Digested peptides were extracted from polyacrylamide gel and separated using reversed-phase liquid chromatography with an RSLCnano Ultimate 3000 system (Thermo Scientific) followed by mass identification with a Orbitrap Velos Pro mass spectrometer (Thermo Scientific). Details about mass spectrometry and data analysis are given in S1 Text.
The virulence of A. fumigatus Δfbx mutants and the corresponding complemented strains was tested in an established murine model for IPA [37,38,77]. In brief, female CD-1 mice were immunosuppressed with cortisone acetate (25 mg/mouse intraperitoneally; Sigma-Aldrich) on days -3 and 0. Mice were anesthetized and intranasally infected with 20 μl of a fresh suspension containing 106 conidia. A control group was mock infected with PBS to monitor the influence of the immunosuppression. The health status was monitored at least twice daily for 14 days and moribund animals (defined by severe dyspnoea and/or severe lethargy) were sacrificed. Infections were performed with a group of 10 mice for each tested strain. Lungs from euthanized animals were removed, and fixed in formalin and paraffin-embedded for histopathological analyses according to standard protocols [39,78].
Mice were cared for in accordance with the principles outlined by the European Convention for the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes (European Treaty Series, no. 123; http://conventions.coe.int/Treaty/en/Treaties/Html/123). All animal experiments were in compliance with the German animal protection law and were approved by the responsible Federal State authority “Thüringer Landesamt für Verbraucherschutz” and ethics committee “Beratende Komission nach § 15 Abs. 1 Tierschutzgesetz” with the permit Reg.-Nr. 03-001/12.
|
10.1371/journal.ppat.1000263 | miR-198 Inhibits HIV-1 Gene Expression and Replication in Monocytes and Its Mechanism of Action Appears To Involve Repression of Cyclin T1 | Cyclin T1 is a regulatory subunit of a general RNA polymerase II elongation factor known as P-TEFb. Cyclin T1 is also required for Tat transactivation of HIV-1 LTR-directed gene expression. Translation of Cyclin T1 mRNA has been shown to be repressed in human monocytes, and this repression is relieved when cells differentiate to macrophages. We identified miR-198 as a microRNA (miRNA) that is strongly down-regulated when monocytes are induced to differentiate. Ectopic expression of miR-198 in tissue culture cells reduced Cyclin T1 protein expression, and plasmid reporter assays verified miR-198 target sequences in the 3′ untranslated region (3′UTR) of Cyclin T1 mRNA. Cyclin T1 protein levels increased when an inhibitor of miR-198 was transfected into primary monocytes, and overexpression of miR-198 in primary monocytes repressed the normal up-regulation of Cyclin T1 during differentiation. Expression of an HIV-1 proviral plasmid and HIV-1 replication were repressed in a monocytic cell line upon overexpression of miR-198. Our data indicate that miR-198 functions to restrict HIV-1 replication in monocytes, and its mechanism of action appears to involve repression of Cyclin T1 expression.
| Monocytes do not support HIV-1 replication, in part because they do not express adequate levels of essential cellular cofactors that mediate steps in the viral replication cycle. Monocytes become permissive for viral replication upon differentiation to macrophages, indicating that cellular cofactors are induced during the differentiation process. One such cofactor is Cyclin T1, which is not expressed in monocytes and is expressed at high levels following macrophage differentiation. Cyclin T1 functions to greatly stimulate the amount of HIV-1 produced in the infected cell. We identified a microRNA (miRNA) named miR-198 that represses the expression of Cyclin T1 in monocytes. miRNAs block expression of proteins by binding to messenger RNAs and preventing their translation by ribosomes. The expression levels of miR-198 are greatly reduced in macrophages, and this appears to allow translation of Cyclin T1 mRNA and expression of Cyclin T1 protein. Our study indicates that this miRNA restricts HIV-1 replication in monocytes. We think that it is possible, if not likely, that additional miRNAs in monocytes also restrict HIV-1 replication by repressing other essential cellular cofactors.
| Productive transcription of eukaryotic protein-coding genes requires a processive form of RNAP II to overcome pauses resulting from negative elongation factors. The positive transcription elongation factor, P-TEFb, plays a critical role in converting RNAP II to a processive enzyme through phosphorylation of the C-terminal domain of RNAP II and negative elongation factors [1],[2]. P-TEFb is composed of Cyclin-dependent kinase 9 (CDK9) as the catalytic subunit and Cyclin T1, T2, or K as the regulatory subunit [3],[4]. Although there are multiple Cyclin partners for CDK9, Cyclin T1-containing P-TEFb (Cyclin T1/P-TEFb) is the major cellular form in cell types examined thus far and it has been studied extensively because of its involvement in HIV-1 gene expression [5]. The HIV-1 Tat transactivator protein recruits Cyclin T1/P-TEFb to the TAR RNA structure of viral transcripts, resulting in a switch from an abortive transcription process to a highly processive one, and this greatly enhances viral gene expression and is essential for viral replication [6].
Unlike Cyclins involved in cell cycle progression, the expression level of Cyclin T1 is usually independent of cell cycle stages. However, Cyclin T1 has been shown to be regulated in human peripheral blood lymphocytes (PBLs), primary CD4+ T cells, and monocytes/macrophages. In PBLs and CD4+ T cells, activation from a resting state results in a strong up-regulation of Cyclin T1 protein expression through a mechanism that involves post-transcriptional regulation [7]–[9]. Cyclin T1 expression is low in freshly isolated monocytes and increases significantly when cells are induced to differentiate into macrophages [10]. The induction of Cyclin T1 in macrophages correlates with a permissive state for HIV-1 replication, as monocytes do not support HIV-1 replication [11]. Additionally, Cyclin T1 protein expression is shut-off at late times of macrophage differentiation by proteasome-mediated proteolysis, but it can be re-induced by macrophage activation or HIV-1 infection [10],[12].
The increase in Cyclin T1 protein expression in monocytes plays an important role in macrophage differentiation, as an shRNA depletion of Cyclin T1 in a monocytic cell line prevents the up-regulation of over 20% of mRNAs normally induced when these cells are stimulated to differentiate into macrophages [13]. We have previously observed that although Cyclin T1 protein expression is very low in freshly isolated primary monocytes, Cyclin T1 mRNA levels are high, suggesting that translation of this mRNA may be actively suppressed in monocytes [10],[14]. Given the approximate 5 kb length of the Cyclin T1 3′UTR [15], a miRNA(s) may be responsible for this repression. MicroRNAs are non-coding small RNAs of about 22 nucleotides in length that function in metazoans through base-pairing preferentially with the 3′UTR of target mRNAs, resulting in translational inhibition or in some cases mRNA degradation [16]–[18]. A number of studies have linked miRNAs to the regulation of specific gene functions, cell fate transition, malignant transformation, and especially hematopoietic lineage commitment [19]–[21].
In this study, we examined the miRNA expression profile during the early stage of primary human monocytes differentiation into macrophages. This analysis identified miR-198 as a negative regulator of Cyclin T1 protein expression through targeting sequences in the 3′UTR of Cyclin T1 mRNA. We found that inhibition of miR-198 function in primary monocytes resulted in increased Cyclin T1 protein levels, and overexpression of miR-198 in differentiating monocytes repressed the normal Cyclin T1 up-regulation. Inhibition of Cyclin T1 up-regulation by miR-198 in a monocytic cell line also resulted in decreased HIV-1 proviral gene expression and HIV-1 replication, indicating that miR-198 possesses an anti-HIV-1 function that appears to function through the repression of an essential cellular cofactor.
Cyclin T1 protein expression is induced by a post-transcriptional mechanism when human primary monocytes are cultured in vitro under conditions that allow macrophage differentiation [10]. Given the ∼4.6 kb 3′UTR in the Cyclin T1 mRNA (see below), we wished to examine changes in miRNAs expression during monocyte to macrophage differentiation, as miRNAs that are down-regulated during differentiation are candidates for repressors of Cyclin T1 protein expression. We isolated total RNA from both monocytes and macrophages allowed to differentiate in vitro from two healthy blood donors. A microarray platform was used to examine expression levels of 321 validated human miRNAs in these RNA preparations (see Materials and Methods).
Using criteria and statistical analysis described in Material and Method, we identified several miRNAs that were differentially expressed in monocyte and macrophages from both donors (Figure 1A). Of these miRNAs, nine were up-regulated (red in heat map) and 13 were down-regulated (green in heat map) in macrophages relative to monocytes. Thus, a total of 22 out of 321 miRNAs examined showed reproducible differential expression during monocyte to macrophage differentiation. To monitor the reliability of the miRNA microarray data, we selected four miRNAs for end-point RT-PCR analysis with commercially available primer sets. As shown in Figure 1B, increased amounts of PCR products for miR-155 were detected in RNA samples from macrophages relative to monocytes, in agreement with the microarray data which showed an induction of miR-155. Reductions in the levels of miR-26a and miR-223 in macrophages were detected in the PCR assays, again in agreement with the microarray data in which both of these miRNAs were down-regulated. Finally, the levels of miR-24 were similar between monocyte and macrophages in the PCR assays, in agreement with microarray data that showed a constant level of miR-24 expression. These results indicate that the microarray data are likely to be reliable in general.
To identify miRNAs that might repress Cyclin T1 mRNA translation, we wished to search Cyclin T1 mRNA sequences for potential target sites for the 13 miRNAs found to be down-regulated during monocyte differentiation. Cyclin T1 mRNA is ubiquitously expressed as a ∼8 kb transcript and because the coding sequence of Cyclin T1 only requires 2178 nucleotides and the 5′UTR is approximately 330 nucleotides in length, it has an extensive 3′ untranslated region (3′UTR) that has not been characterized [4],[22]. The alignment of human EST sequences and the full-length goat Cyclin T1 mRNA sequence suggest that the 3′UTR of human Cyclin T1 mRNA is contained within a single exon [15]. An analysis of the 8 kb sequence downstream of the Cyclin T1 stop codon revealed seven potential poly(A) signals (AATAAA) (Figure 2A). To determine the 3′UTR sequences for miRNA target predictions, we first characterized the poly(A) signal usage in the Cyclin T1 mRNA by RT-PCR.
RNA samples from cell lines, primary CD4+ lymphocytes, and monocytes/macrophages were examined to determine if alternative poly(A) signals are utilized for the Cyclin T1 3′UTR. Using an RT primer that anneals to the start of the poly(A) sequences and PCR primers specific for each poly(A) signal (P1 to P5, Figure 2A), we were able to amplify a discrete band of ∼400 bp only with primer P4 (Fig 2B). Extension times in PCR reactions were 45 seconds to maximize the expected ∼400 bp product that would be produced by the primer nearest to the major poly(A) site. These data suggest that the fourth poly(A) signal is utilized to produce the majority of Cyclin T1 mRNA in the cell types examined. The absence of PCR products in reactions without reverse transcription excludes the possibility of genomic DNA contamination in PCR reactions. A BLAST search revealed two cDNA clones with only three or four nucleotide differences with the P4-PCR product (Figure 2C), further supporting the conclusion that the fourth poly(A) signal is used predominantly. Based upon these results, we conclude that Cyclin T1 mRNA arises predominantly from utilization of the fourth poly(A) signal and its length is ∼7.1 kb (plus poly(A) sequences). Additionally, this analysis found no evidence of alternative poly(A) site selection in CD4+ lymphocytes and monocyte/macrophages. This deduced Cyclin T1 3′UTR was used for miRNA target site predictions.
We selected several down-regulated miRNAs, including miR-15b, miR-26a, and miR-198, for initial experiments. These three miRNAs were down-regulated >2-fold in our microarray data with monocytes/macrophages and each are predicted to have >five target sites in the Cyclin T1 mRNA 3′UTR with minimum free energies (mfe) less than -20 kcal/mol. In gain-of-function experiments, precursors for these miRNAs were transfecting into 293T cells and Cyclin T1 protein expression was examined in immunoblots. Cyclin T1 protein levels were reduced in 293T cells transfected with the miR-198 precursor (pre-miR-198) relative to mock-transfected cells, while miR-15b precursor (pre-miR-15b) had no significant effect (Figure 3A). Similar to the results with the pre-miR-15b transfection, the miR-26a precursor did not affect Cyclin T1 protein expression (data not shown). SiRNAs that target the Cyclin T1 coding region were used as a positive control for transfection, and these siRNAs successfully knocked down Cyclin T1 protein expression.
To further validate the effect of miR-198 on Cyclin T1 protein expression, increasing amounts of pre-miR-198 were transfected into HeLa cells, and a dose-dependent reduction in Cyclin T1 protein level was seen (Figure 3B). A control miRNA precursor, pre-miR-Control, did not affect Cyclin T1 expression. Additionally, miR-198 did not affect Cyclin T1 mRNA levels, which remained constant in cells transfected with pre-miR-198 (Figure 3C). The levels of β-actin and GAPDH mRNAs also remained unchanged following transfection of pre-miR-198, suggesting that miR-198 overexpression has no impact on overall mRNA levels. These data demonstrate that miR-198, a miRNA which is down-regulated during monocyte to macrophage differentiation, is capable of down-regulating Cyclin T1 protein expression. Because the ectopic expression of miR-198 reduces Cyclin T1 protein levels without affecting Cyclin T1 mRNA levels, it is likely that miR-198 acts through translational inhibition as is common for miRNAs.
We found that miR-198 is predicted to have >10 potential target sites in the 3′UTR of Cyclin T1 mRNA with mfe lower than −20 kcal/mol (see Materials and Methods). To determine if the Cyclin T1 3′UTR can be regulated by miR-198, we inserted the full length 3′UTR between the luciferase coding sequence and the poly(A) signal of the pGL3 firefly luciferase vector driven by CMV immediate early promoter (pU3-full, Figure 4B). An shRNA vector was generated to express miR-198 (shmiR-198) from the pCL-based retroviral vector. An shRNA designed to target green fluorescence protein, shGFP, was also generated as a control. Pools of HeLa cells expressing shGFP or shmiR-198 were generated by infection of cells with shRNA retroviruses and subsequent puromyocin selection. Reporter plasmids were then transfected into shGFP- or shmiR-198-expressing HeLa cells along with a renilla luciferase reporter plasmid (pTK-RL) as an internal control for transfection efficiency. Additionally, p198T, which contains sequences with perfect complementarity to miR-198, was included as a positive control that should be repressed through an siRNA pathway. As shown in Figure 5A, the relative expression of the positive control p198T was repressed in cells expressing shmiR-198 relative to cells expressing shGFP, showing more than a 60% reduction. The parental pVector expressed at a slightly higher level in shmiR-198-expressing cells. Expression of pU3-full was also repressed by ∼40%, suggesting that there are miR-198 target sites in the Cyclin T1 3′UTR.
To identify the locations of these target sites, we divided the full-length 3′UTR into four overlapping fragments (U3-1 to U3-4) and inserted them into to the pGL3 vector (pU3-1 to pU3-4, Figure 4A and 4B). Only expression from the pU3-1 and pU3-4 reporters showed a statistically significant repression in shmiR-198-expressing cells, demonstrating a 30% and a 20% reduction for pU3-1 and pU3-4, respectively (Figure 5A). It appears therefore that target sites for miR-198 reside in these two fragments, consistent with recent findings that miRNA target sites are generally locate away from the center of a 3′UTR [23].
There were three predicted target sites in fragment U3-1 and U3-4 with mfe<−25 kcal/mol, and they were designated site 2478, site 2867, and site 6502 according to their locations in the Cyclin T1 mRNA (Figure 4A, first nucleotide in coding sequence defined as +1). All three predicted targets show extensive of complementarity with miR-198 at their 5′ ends. However, site 6502 contains additional eight nucleotides at its 3′ end complementary to the miR-198 seed sequence (Figure 4C), which closely resembles the structure of the 5′-dominant canonical miRNA target site [24]. Expression of reporter plasmids containing two or three copies of site 2478 (p2478), site 2867 (p2867), or site 6502 (p6502) in the 3′UTR of the luciferase gene was assayed in cells expressing either shmiR-198 or shGFP (Figure 5A). A control reporter plasmid, pCtrl, which contains three copies of unrelated sequences, expressed at similar levels in both shmiR-198- and shGFP-expressing cells. Expression of p2867 was reduced a relatively modest 20% in shmiR-198-expressing cells, while expression of p2478 was unaffected perhaps due to a sub-optimal seed sequence. Expression of p6502 was significantly repressed by shmiR-198, showing a 40–50% reduction in shmiR-198-expressing cells relative to control cells. Therefore, site 2867 in fragment U3-1 and especially site 6502 in fragment U3-4 appear to be responsible for repression mediated by miR-198 in Cyclin T1 mRNA 3′UTR. Although both of the sites have similar binding energies, site 6502 showed greater sensitivity to shmiR-198 when compared to site 2867. When site 6502 along with its adjacent sequences was clone into the reporter plasmid (p6502g-WT, Figure 4B) and assayed, a ∼50% reduction was observed in shmiR-198-expressing cells relative to shGFP-expressing cells (Figure 5A). In addition, mutation of three nucleotides in the site 6502 which are complementary to the mRNA seed sequence abolished the inhibition by shmiR-198 (Figures 4C and 5A), strongly suggesting the direct targeting of miR-198 to site 6502 and highlighting the importance of the seed match present in site 6502 during miRNA-mediated repression.
To verify that repression of Cyclin T1 protein expression by miR-198 requires target sequences in the Cyclin T1 3′UTR, we carried out a plasmid transfection experiment with a HA-tagged Cyclin T1 expression plasmid lacking the Cyclin T1 3′UTR. The HA-Cyclin T1 plasmid was co-transfected with a precursor for either miR-198 (pre-miR-198) or a pre-miR-Control. To verify that the transfected pre-miR-198 was biologically active, we co-transfected the p198T luciferase control plasmid (Figure 4C) that is repressed by miR-198 by a siRNA pathway. Additionally, a renilla luciferase reporter plasmid was co-transfected to monitor transfection efficiency and siRNAs against Cyclin T1 were included in a control co-transfection. The siRNAs against the HA-Cyclin T1 were very effectively in reducing protein expression but had little effect on expression of the p198T reporter plasmid (Figure 5B). Although the pre-miR-198 was able to reduce expression of the p198T reporter plasmid about 20-fold relative to the pre-miR control, it had no inhibitory effect on Cyclin T1 expression and even appeared to stimulate Cyclin T1 expression. These data indicate that the repression of Cyclin T1 protein expression by miR-198 requires target sequences in the Cyclin T1 mRNA 3′UTR.
The data presented above demonstrate that miR-198 can target sequences in the Cyclin T1 mRNA 3′UTR and can repress expression of endogenous Cyclin T1 protein expression when ectopically expressed. To examine whether miR-198 regulates Cyclin T1 expression in primary monocytes, we isolated primary monocytes and macrophages from two donors (Donor3 and Donor4) and first re-examined the correlation between miR-198 expression and Cyclin T1 protein levels. In agreement with our previous results [10],[14], Cyclin T1 protein levels were strongly up-regulated in macrophages relative to monocytes (Figure 6A). Additionally, although Cyclin T1 protein expression is induced in macrophages, Cyclin T1 mRNA levels were reduced from 20 to 60% when compared to levels on monocytes (Figure 6B), similar to our previous results [10],[14]. In agreement with the microarray data shown in Figure 1A, miR-198 levels were reduced in macrophages relative to monocytes, showing a 14-fold reduction in Donor3 and a 19-fold reduction in Donor4. Although both Cyclin T1 mRNA and miR-198 levels were decreased in macrophages, the ratio between Cyclin T1 mRNA and miR-198 increased in macrophages. This observation indicates that in macrophages there are more copies of Cyclin T1 mRNA for each copy of miR-198.
In loss-of-function experiments, we examined the effects of a miR-198 inhibitor on Cyclin T1 expression in monocytes. We transfected a chemically modified single-stranded nucleic acid inhibitor of miR-198 (anti-miR-198) into freshly isolated monocytes. As shown in Figure 6C, anti-miR-198 was able to induce Cyclin T1 levels from two- to six-fold in two donors examined (Donor 3, 4). In an additional gain-of-function experiment, transfection of pre-miR-198 to overexpress miR-198 in monocytes induced to differentiate with GM-CSF treatment resulted in about a 3-fold decrease in Cyclin T1 protein levels (Figure 6C, right panel). The effects of miR-198 on Cyclin T1 expression are consistent with recent quantitative proteomic studies that have shown that in most cases miRNAs influence proteins levels about 1.5- to 2-fold [25],[26]. These data in primary monocytes further support the proposal that miR-198 plays a role in repressing Cyclin T1 protein expression in monocytes, and the reduction in miR-198 levels during macrophage differentiation contributes to induction of Cyclin T1 protein expression. The induction of Cyclin T1 has been shown to be important for the program of gene expression in macrophages [13].
In addition to cellular genes that are dependent on Cyclin T1 for expression, HIV-1 Tat-mediated gene expression is highly dependent upon Cyclin T1. We therefore examined if down-regulation of Cyclin T1 by miR-198 could result in decreased HIV-1 proviral expression. We used a promonocytic cell line, Mono Mac 6 (MM6) for these experiments, as we previously determined that this cell line is a useful model to investigate the role of Cyclin T1 function in monocyte differentiation [13]. Cyclin T1 is up-regulated in MM6 cells treated with PMA for 48 hours (Figure 7A, left panel), similar to its induction during primary monocyte to macrophage differentiation. Ectopic expression of miR-198 in MM6 cells from a transfected shmiR-198 plasmid, followed by PMA treatment, largely prevented Cyclin T1 up-regulation (Figure 7A, right panel). In contrast, transfection of a control shGFP plasmid resulted in similar Cyclin T1 levels as those seen in mock-transfected cells.
To examine if HIV-1 proviral gene expression is affected by the down-regulation of Cyclin T1 by miR-198, a HIV-1 NL4-3 luciferase proviral reporter plasmid was co-transfected with shGFP or shmiR-198 plasmid into MM6 cells. A renilla luciferase reporter plasmid, pTK-RL, was also co-transfected as an internal control. Cells were treated with PMA, harvested 48 hours after transfection, and divide into two equal portions for luciferase assays and immunoblot analysis. The results of two independent experiments are shown in Figure 7B; in one of these experiments, duplicate transfections were performed. Luciferase assays revealed that expression of the HIV-1 proviral reporter plasmid was approximately nine-fold lower in shmiR-198-expressing cells than in shGFP-expressing cells, and the results of immunoblots indicated that this reduction correlated with decreased Cyclin T1 protein levels. These data suggest that miR-198 is capable of repressing HIV-1 proviral expression by targeting its cellular cofactor Cyclin T1.
To examine if HIV-1 replication is also repressed by ectopic expression of miR-198, MM6 cells were transfected with a shGFP control or shmiR-198 plasmid and cells were treated with PMA immediately after transfection. Cultures were infected with M-tropic HIV-1 strain SF162 after one day of PMA treatment and p24 expression in culture supernatants was measured at either three or four days post-infection in two independent experiments (Figure 7C). A 25% reduction in p24 expression was observed when expression was examined at three days post-infection, while a 50% reduction was observed when p24 expression was examined at four days post-infection. The greater inhibition of proviral reporter plasmid expression by miR-198 in Figure 7B than in the HIV-1 infection in Figure 7C is likely the result of transfection efficiency. In the proviral reporter plasmid experiment, both the reporter plasmid and the miR-198 shRNA vector were co-transfected and it is therefore likely that the majority of transfected cells contained both plasmids. In the HIV-1 infection experiment, it is likely that a significant portion of infected cells were not transfected with the miR-198 shRNA vector. The data in Figure 7C further suggest that miR-198 is capable of repressing HIV-1 replication through down-regulation of the essential cofactor Cyclin T1.
It has been reported that miRNA expression shows distinct patterns in HIV-1 provirus plasmid-transfected HeLa cells [27]. Therefore, we wished to examine if HIV-1 infection affects miR-198 expression in macrophages. Macrophages isolated from healthy blood donors were allowed to differentiate for four days, and were infected with a M-tropic HIV-1 SF162 strain for seven days, and RNA was isolated for miR-198 quantification. Measurements of p24 expression were performed at day four and seven post-infection to verify productive infections (data not shown). As shown in Figure 8A, miR-198 levels increased modestly to 1.9-fold in Donor 6 and 3.8-fold in Donor 7 when normalized to U6B snRNA. These data suggest that HIV-1 infection up-regulates miR-198 levels in macrophages. Cyclin T1 mRNA levels also increased 1.8-fold in Donor 6 and 1.6-fold in Donor 7, while α-tubulin mRNA levels remained the same (Figure 8B). Cyclin T1 protein expression is induced in HIV-1-infected macrophages, and this has been shown to involve an inhibition of proteasome-mediated proteolysis of Cyclin T1 [12]. The data shown on Figure 7B indicates that a relatively modest increase in Cyclin T1 mRNAs may contribute to the induction in Cyclin T1 protein expression in infected macrophages. The significance in the modest increase in miR-198 levels in HIV-1 infected macrophages remains to be elucidated.
Our finding that miR-198 represses Cyclin T1 protein expression in monocytes provides mechanistic insight into previous observations that translation of Cyclin T1 mRNA is inhibited in these cells [10],[14]. Because Cyclin T1 is a component of P-TEFb, a general RNA polymerase II elongation factor, miR-198 is likely to broadly regulate gene expression in monocytes through its own targeting functions and through indirect effects on Cyclin T1-dependent genes. It is important to note that P-TEFb function is not deficient in monocytes, as these cells express high levels of Cyclin T2, an alternative Cyclin subunit of P-TEFb [13],[28]. Other miRNAs have been shown to target transcription factors and are also capable of broadly regulating gene expression through indirect mechanisms. For example, miRNA lin-4 targets lin-14 in C. elegans [29],[30] and in neuronal cells, miR-124 targets the RNAP II CTD phosphatase SCP1, a component of the REST transcription repressor complex [31].
Freshly isolated monocytes do not support viral replication and must undergo a program of macrophage differentiation to become permissive for HIV-1 replication [11]. Viral entry is not limiting in monocytes, but reverse transcription and nuclear import of the pre-integration complex are defective [32],[33]. Because Tat is essential for HIV-1 replication [6] and Cyclin T1 is a critical Tat cofactor, miR-198 can function in monocytes as an additional repressor of viral gene expression and replication (Figure 6B and 6C). MiRNAs have previously been reported to regulated Tat function, as the miRNA cluster miR-17/92 is capable of inhibiting HIV-1 replication through repression of the histone acetyltransferase PCAF that also participates in Tat function [34].
We observed that HIV-1 infection of macrophages modestly induces both miR-198 and Cyclin T1 mRNA expression levels (Figure 7). Because HIV-1 infection induces Cyclin T1 protein expression in macrophages [12], it is possible that following up-regulation of Cyclin T1 mRNA and protein by infection, a cellular negative-feed back loop is activated that results in elevated levels of miR-198 and a subsequent dampening of the induction of Cyclin T1.
The up-regulation of Cyclin T1 protein expression is an important event during monocyte differentiation, as shRNA depletions and transcriptional profiling have shown that Cyclin T1 is required for expression of more than 20% of the mRNAs that are induced during the differentiation program [13]. Because miR-198 restricts Cyclin T1 expression in monocytes, this miRNA may contribute to the prevention of a transcriptional program of differentiation. The reduction of expression of miR-198 is therefore likely to be important for monocyte differentiation. Furthermore, genes involved in immune responses are overrepresented in the set of Cyclin T1-dependent mRNAs in macrophages [13], suggesting that proper macrophage function may require the down-regulation of miR-198. A topic for future research will be the study of signals and mechanisms that lead to the down-regulation of miR-198 in macrophages.
It is notable that Cyclin T1 mRNA levels are reduced ∼30–60% when Cyclin T1 protein levels are up-regulated following monocyte differentiation (Figure 5) [10],[14]. The explanation for this phenomenon is not clear, but it is possible that transcription of the Cyclin T1 gene is not increased during differentiation, as the promoter for Cyclin T1 appears to be constitutively active [22]. The active translation of Cyclin T1 mRNA following the relief of repression by miR-198 may reduce the mRNA half-life, and if a transcriptional induction of the Cyclin T1 gene does not occur, a decrease in the amount of Cyclin T1 mRNA will result.
miR-198 is in the 3′UTR (exon 11) of FSTL1, a gene that has recently been shown to enhance inflammatory cytokine production in a monocytic cell line after mitogen stimulation [35]. The correlation between the FSTL1 transcript and miR-198 levels has not yet been established, but FSTL1 transcripts are not detectable in peripheral blood leukocytes [36], suggesting that the processing of the FSTL1 primary transcript for protein production or miR-198 might be mutually exclusive.
Little information is available concerning potential functions and mRNA targets for miR-198. Although expression in monocytes was not examined in a previous study in 40 normal human tissues, miR-198 expression was found to be generally low and restricted to only a few tissues [37]. In hepatic tumors, expression of miR-198 was found to be down-regulated relative to normal liver parenchyma [38]. A genetic analysis identified a SNP in the miR-198 gene that has a nominally significant allelic association with schizophrenia [39]. We used both PicTar [40] and TargetScan [41],[42] to identity mRNAs that are predicted to be targets of miR-198. This analysis identified 41 mRNAs in common between the two programs that are potential targets for miR-198. However, no common features from the proteins encoded by these mRNAs were apparent. An intriguing predicted target for miR-198 is in the 3′UTR of PTEN, a negative regulator of the PI3-kinase pathway, and PTEN is involved in LPS signaling in monocytes/macrophages [43]. Interestingly, the HIV-1 Tat protein has been shown to decrease expression of PTEN in primary human macrophages [44], which is consistent with our findings that HIV-1 infection induces miR-198 expression in macrophages (Figure 8). Because overexpression of PTEN enhances HIV-1 expression [45], it is possible that the anti-HIV-1 function of miR-198 involves targeting cellular cofactors in addition to Cyclin T1.
We note that miR-223,which has been shown to target HIV-1 in the 3′ end of the viral genome and repress its expression [46], was identified in our miRNA profiling as a miRNA that is down-regulated during macrophage differentiation (Figure 1). This further emphasizes the power of miRNA profiling in identifying miRNAs associated with specific processes. Finally, given the ∼5 kb of 3′UTR sequence in Cyclin T1 mRNA, it is possible if not likely that miRNAs in addition to miR-198 are involved in repression of Cyclin T1 expression in monocytes and other cell types. Our miRNA profiling only assayed 321 miRNAs and given that it is estimated that there are >800 human miRNAs, it is possible that additional miRNAs that regulate Cyclin T1 in monocytes and other cell types await discovery.
Peripheral blood mononuclear cells (PBMCs) were isolated from healthy blood donors (Gulf Coast Regional Blood Center, Houston, TX) by Isolymph density gradient centrifugation (Gallard-Schlesinger). Primary monocytes were isolated from PBMCs by negative selection using monocyte Isolation Kit II (Miltenyi Biotec). Macrophages from the same donors used for monocyte preparations were obtained by adherence of PBMCs to dishes in RPMI medium supplemented with 1% human serum for one hour as described previously [10]. Adhered cells were washed three times with phosphate buffered saline (PBS), incubated with complete RPMI medium (10% fetal bovine serum and 1% Penicillin-Streptomycin liquid, GIBCO) for two hours, washed three times with PBS, and cultured with complete RPMI medium containing GM-CSF (10 units/ml) for four to five days before RNA isolation. Purities of monocyte and macrophage preparations were evaluated by flow cytometry using CD14 as a monocyte marker and CD71 as a macrophage marker. The purity of monocyte and macrophage preparations was determined to be ≥ 92% (data not shown). Resting CD4+ T cells were isolated and activated as previous described [9].
To prepare RNA samples for miRNA microarray or miRNA quantification, total RNA enriched with small RNAs was isolated using mirVana miRNA isolation kit (Ambion) according to the manufacturer's protocol. RNA preparations from Donor1 and Donor2 were sent to LC Sciences (Houston, TX) for miRNA microarray analysis using the μParaflo microfluidic chips and detailed process can be found at http://www.lcsciences.com. Briefly, probes for miRNAs were in situ synthesized on chips using photogenerated reagent chemistry with repeats for each probe to allow statistical analysis. Version 7.1 arrays were used to detect 321 unique mature miRNAs. RNA samples from the same donor for comparison were labeled with Cy3 or Cy5 for hybridization, and the labeling was reversed between the same cell populations from the two donors. Quality control was done by hybridizing a group of control oligos to evaluate array quality as well as by mixing a fixed amount of 20-mer RNA oligos to samples as external controls. Data were processed by subtracting background and normalized the signals using a LOWESS filter [47]. A detectable transcript was defined by fulfilling the following criteria: (1) signal intensity >3×background standard deviation; (2) spot CV (standard deviation/signal intensity) <0.5; (3) signals from at least 50% of the repeating probes are above detection level. The ratio of the two colors of detected signals and p-values of the t-test were calculated. A cut-off at p-value < 0.01 was used to define a differentially expressed miRNA. A heat maps for all differentially expressed miRNAs was generated by Cluster [48]. The primary data can be accessed at http://www.bcm.edu/molvir/labs/herrmann-rice-lab/Rice-Herrmann_Pages-Data.htm.
End-point PCR analysis for miR-26a, miR-155, miR-223, and miR-24 were used to verify microarray results of Donor1 and Donor2 by mirVana qRT-PCR detection kit and primer sets (Ambion). A 25-cycle of PCR was preformed and the PCR products were resolved by gel electrophoresis using a 3.5% agarose gel to examine the ∼90 bp products. Quantification of band intensity was done by analysis of the gel pictures with ImageQuant (Molecular Dynamics). TaqMan MicroRNA assays (Applied Biosystems) were use to quantify mature miR-198, U6B snRNA, and U3B snRNA. Target prediction was done using primarily RNAhybrid [49] and also microInspector [50].
Total RNA was isolated from cells using RNeasy mini kit (Qiagen) and treated with DNase I (Invitrogen) for reverse transcription using ThermoScript RT-PCR system (Invitrogen) and primer T7-(TTT)6-V (5′-TTGTAATACGACTCACTATAGGGC-TTTTTTTTTTTTTTTTTT-A/G/C-3′). PCR was performed using reverse primer for T7 and forward primers specific for each poly(A) signal (P1: 5′-gggccaatggtcacaacacg-3′, P2: 5′-TGTATATCCGAAGGGAAACAGC-3′, P3: 5′-GCTGAACTGTAGTGAAGCATCG-3′, P4: 5′-CAGCATATTATTCTGCCATTGC-3′, and P5: 5′-ATTGTTACATGAATACCTGG-3′). PCR products were resolved on a 2% agarose gel and bands of ∼400bp were extracted from the gel, purified with QIAquick gel extraction kit (Qiagen) for TOPO TA cloning (Invitrogen) for DNA sequencing.
An shRNA strategy was utilized to express miR-198 (shmiR-198) or a control shRNA designed to target GFP (shGFP) from a pCL-based retroviral vector [51]. BOSC cells were used to generate viruses by co-transfecting shGFP or shmiR-198 plasmid and an amphotropic packaging vector using lipofectamine 2000 (Invitrogen). Viruses were harvested 24 hours after transfection for infection of HeLa cells for another 24 hours. Infected HeLa cells were then selected by puromyocin treatment (1 µg/ml) for two days to obtain cell pools expressing shGFP or shmiR-198. As quantified by TaqMan MicroRNA assays (Applied Biosystems), miR-198 was expressed >400-fold in cell pools expressing shmiR-198 relative to pools expressing shGFP.
Full-length or fragments of Cyclin T1 3′UTR, or copies of three predicted target sequences were inserted between the luciferase gene and the poly(A) signal of a pGL3-based firefly luciferase reporter driven by CMV immediate early promoter (Promega). The full-length Cyclin T1 3′UTR was cloned by PCR reactions using BAC clones as templates and extra ∼500bp genomic sequence downstream of the poly(A) signal was included for cloning into the luciferase reporter (pU3-full). Fragments of Cyclin T1 3′UTR were cloned by PCR reactions using pU3-full as template and primers containing XbaI site at the 3′ ends for subsequent cloning. Individual predicted target sequence was obtained by annealing complementary oligos containing copies of the predicted sequence and 3′ overhang for XbaI site ligation. Mutagenesis was done using QuikChange Site-Directed Mutagenesis Kit (Stratagene) according to manufacturer's instructions.
For reporter plasmid experiments, 25 ng of each firefly luciferase reporter plasmid was co-transfected into shGFP- or shmiR-198-expressing HeLa cell pools along with 10 ng of thymidine kinase promoter-driven renilla reporter plasmid (pTK-RL, control for transfection efficiency) using Lipofectamine 2000 (Invitrogen). To further boost miRNA expression, 1ug of shGFP- or shmiR-198 plasmids were also co-transfected. Cells were harvested 24 hours after co-transfection and luciferase activities were measured using Dual-Luciferase Reporter Assay System (Promega). Relative luciferase activity was calculated by dividing firefly luciferase activity by renilla luciferase activity (FL/RL). Relative expression was obtained by normalization to FL/RL in shGFP-expressing cells transfected with the same firefly reporter plasmid.
For the transfection experiment to examine the requirement of the Cyclin T1 3′UTR for repression by miR-198, pHA-Cyclin T1 [4] (50 ng), p198T (50 ng) and pTK-renilla luciferase (10 ng) were co-transfected into HeLa cells with pre-miR-Control (50 nmol), pre-miR-198 (50 nmol) or siCyclin T1 (50 nmol) in a 24-well culture dish using Lipofectamine (Invitrogen). Cells were harvested at 24 hours post-transfection and split into two portions for immunoblot analysis and dual-luciferase assays.
miRNA precursors (6.25 pmole to 100 pmole, Applied Biosystems) or an siRNA targeting Cyclin T1 coding region (Dharmacon) were transfected into HeLa or 293T cells (∼30% confluent at times of transfection) using oligofectamine (Invitrogen). Cells were harvested by direct lysis with 1×loading buffer for immunoblot analysis at 72 hours post-transfection. Independent transfections were performed in HeLa cells for total RNA isolation using RNeasy mini kit (Qiagen) and RNA were subjected to RT-real-time-PCR assays for quantification of Cyclin T1, β-actin, GAPDH, and α-tubulin mRNA levels using the Bio-Rad MyIQ single color detection system as previously described (Sung and Rice, 2006). Primer sequences designed by the Beacon Designer 2.0 (Premier Biosoft) are: β-actin-F 5′-AGCAAGCAGGAGTATGACGAGTC-3′, β-actin-R 5′-AGAAAGGGTGTAACGCAACTAAGTC-3′, GAPDH-F 5′-CGCCAGCCGAGCCACATC-3′, GAPDH-R 5′-AAATCCGTTGACTCCGACCTTCAC-3′. miR-198 inhibitor (anti-miR-198) or control miRNA inhibitor (anti-miR-Control) was transfected into primary monocytes (250 pmole inhibitors/3×106 cells) obtained from healthy donors using the Amaxa Nucleofector system. Cells were harvested by direct lysis at 18 or 24 hours after transfection for immunoblot analysis to detect Cyclin T1 and β-actin protein expression. Pre-miR-198 or pre-miR-Control was also transfected with the Amaxa system into primary monocytes (250 pmole inhibitors/3×106 cells) and cells were treated with GM-CSF to induce differentiation for 48 hours. Cyclin T1 protein levels were quantified by scanning X-ray films with Personal Densitometer SI and analyzed with ImageQuant (Molecular Dynamics), followed by normalization to β-actin protein levels. Antibodies for immunoblot were purchased from Santa Cruz Biotechnology.
The promonocytic cell line MM6 was cultured and transfected according to the protocols described at http://www.monocytes.de. Briefly, MM6 cells were propagated in RPMI 1640 (Invitrogen) supplemented with L-glutamine (2 mM, Invitrogen), penicillin-streptomycin (400 U/ml and 400 µg/ml, Invitrogen), 1×non-essential amino acids (Invitrogen), OPI media supplement (Sigma), and 10% certified fetal bovine serum tested for endotoxin (Invitrogen). Low passage MM6 cells were used and cells were kept at a concentration of no more than 5×106/ml to maintain a low endogenous Cyclin T1 protein expression level. For HIV-1 wild type LTR-luciferase reporter assays, 5×106 MM6 were washed with RPMI 1640 and incubated with 3 µg shGFP- or shmiR-198-expressing plasmids, 1 µg HIV-1 luciferase reporter plasmid, 1 µg pTK-RL, and 250 µg/ml DEAE dextran (Sigma) for 90 minutes. The HIV-1 luciferase reporter plasmid contains a deletion in the env gene, a replacement of the nef gene with firefly luciferase gene, and a deletion in the vpr gene. Cells and plasmids were mixed every 30 minutes by gently rocking the tube during incubation and then treated with 10% DMSO for 3 minutes. After washing out DMSO, cells were treated with PMA (10 µg/ml final concentration), harvested 48 hours later, and divided into two equal portions for dual-luciferase assays and immunoblot analyses as described above. HIV-1 infections with strain SF162 (100 TCID50 units per culture in 6 well dishes) were performed one day after PMA treatment. Measurements of p24 expression (RETROtek, ZeptoMetrix Corporation) in culture supernatants were performed at day three or four post-infection as indicated.
Macrophages were allowed to differentiate for five days from monocytes and incubated with HIV-1 SF162 strain at 6000 TCID50 per 10 cm2 dish for two hours (37°C, 5% CO2) with intermittent shaking. Following virus adsorption, cells were washed three times with PBS. Cells were then supplied with complete RPMI medium and incubated for seven days to allow at least two rounds of infection. Measurements of p24 expression (RETROtek, ZeptoMetrix Corporation) at day four and day seven of infection were performed to verify productive infections. At day seven post-infection, cells were washed with PBS three times for total RNA isolation for miR-198 quantification and mRNA quantification as described above.
|
10.1371/journal.pgen.1007939 | Regeneration of the zebrafish retinal pigment epithelium after widespread genetic ablation | The retinal pigment epithelium (RPE) is a specialized monolayer of pigmented cells within the eye that is critical for maintaining visual system function. Diseases affecting the RPE have dire consequences for vision, and the most prevalent of these is atrophic (dry) age-related macular degeneration (AMD), which is thought to result from RPE dysfunction and degeneration. An intriguing possibility for treating RPE degenerative diseases like atrophic AMD is the stimulation of endogenous RPE regeneration; however, very little is known about the mechanisms driving successful RPE regeneration in vivo. Here, we developed a zebrafish transgenic model (rpe65a:nfsB-eGFP) that enabled ablation of large swathes of mature RPE. RPE ablation resulted in rapid RPE degeneration, as well as degeneration of Bruch’s membrane and underlying photoreceptors. Using this model, we demonstrate for the first time that zebrafish are capable of regenerating a functional RPE monolayer after RPE ablation. Regenerated RPE cells first appear at the periphery of the RPE, and regeneration proceeds in a peripheral-to-central fashion. RPE ablation elicits a robust proliferative response in the remaining RPE. Subsequently, proliferative cells move into the injury site and differentiate into RPE. BrdU incorporation assays demonstrate that the regenerated RPE is likely derived from remaining peripheral RPE cells. Pharmacological disruption using IWR-1, a Wnt signaling antagonist, significantly reduces cell proliferation in the RPE and impairs overall RPE recovery. These data demonstrate that the zebrafish RPE possesses a robust capacity for regeneration and highlight a potential mechanism through which endogenous RPE regenerate in vivo.
| Diseases resulting in retinal pigment epithelium (RPE) degeneration are among the leading causes of blindness worldwide, and no therapy exists that can replace RPE or restore lost vision. One intriguing possibility is the development of therapies focused on stimulating endogenous RPE regeneration. For this to be possible, we must first gain a deeper understanding of the mechanisms underlying RPE regeneration. Here, we develop a transgenic zebrafish system through which we ablate large swathes of mature RPE and demonstrate that zebrafish regenerate RPE after widespread injury. Injury-adjacent RPE proliferate and regenerate RPE, suggesting that they are the source of regenerated tissue. Finally, we demonstrate that Wnt signaling may be involved in RPE regeneration. These findings establish a versatile in vivo model through which the molecular and cellular underpinnings of RPE regeneration can be further characterized.
| The RPE is a polarized monolayer of pigment-containing cells that separates the retina from the choroid and performs many critical functions for vision. Microvilli extend from the apical RPE surface and interdigitate with photoreceptor outer segments, enabling the RPE to support photoreceptor health [1]. The basal surface of the RPE abuts and helps to form Bruch’s membrane (BM), which, along with tight junctions between RPE cells, creates the blood-retina barrier and facilitates nutrient and ion transport between the retina and choriocapillaris [2–4]. Additionally, RPE pigment prevents light scatter by absorbing stray photons. Due to its importance in maintaining retinal function, diseases affecting the RPE have dire consequences for vision. Age-related macular degeneration (AMD) is one such disease, and is the third leading cause of blindness in the world [5,6]. AMD is commonly divided into two types: atrophic (dry) and exudative (wet). In the early stages of atrophic AMD, RPE cells in the parafovea become dysfunctional and progressively degenerate, and this is thought to result in death of parafoveal rods [7–9]. Progressively, RPE dysfunction and degeneration spread to the fovea, resulting in loss of cone photoreceptors, and ultimately, loss of high-acuity vision [10–12]. Exudative AMD occurs in a subset of atrophic AMD cases when choroidal vasculature invades the retina [11,13].
Transplantation of stem cell-derived RPE has emerged as a possibility for treating AMD [14–16], and clinical trials are currently underway [17–23]. However, little is known about the fate of transplanted RPE, and whether their survival and integration can be improved. An unexplored complementary approach is the development of therapies that stimulate endogenous RPE regeneration. In mammals, RPE regeneration is limited and dependent upon the size of the injury [24]; small lesions can be repaired by the expansion of adjacent RPE [25,26], but existing RPE are unable to repair large lesions [24,27–30]. In some injury paradigms, RPE cells proliferate but do not regenerate a morphologically normal monolayer (e.g. [26,31,32]). Indeed, RPE often overproliferate after injury, such as during proliferative vitreoretinopathy (PVR), where proliferative RPE invade the subretinal space and lead to blindness [33–35]. Recently, a subpopulation of quiescent human RPE stem cells was identified that can be induced to proliferate in vitro and differentiate into RPE or mesenchymal cell types [30,36], suggesting that the human RPE contains a population of cells that could be induced to regenerate.
Little is known about the process by which RPE cells respond to elicit a regenerative, rather than pathological, response. Indeed, no studies have demonstrated regeneration of a functional RPE monolayer following severe damage in any model system. The development of such a model is a critical first step to acquiring a deeper understanding of the molecular mechanisms underlying RPE regeneration. Zebrafish offer distinct advantages for this purpose: the development, structure and function of the zebrafish eye is similar to human, including a cone-rich larval retina; they are amenable to genetic manipulation and imaging, and they can regenerate neural tissues (e.g.[37–39]). However, it is unknown whether the zebrafish RPE is capable of regeneration. Here, we demonstrate that the zebrafish RPE possesses a robust capacity for regeneration and identify cellular and molecular mechanisms through which endogenous RPE regenerate in vivo.
To develop an RPE injury model, we utilized a transgenic line in which a promoter element from rpe65a drives expression of the nfsB-eGFP fusion protein in mature RPE [40] (rpe65a:nfsB-eGFP; Fig 1). nfsB is an E. coli nitroreductase that converts the ordinarily benign prodrug metronidazole (MTZ) into a potent DNA crosslinking agent, leading to apoptosis in expressing cells [41–44]. rpe65a:nfsB-eGFP transgenic embryos were treated with phenylthiourea (PTU) [45] to suppress melanin synthesis. To ablate RPE, 5dpf larvae were removed from PTU and exposed to 10mM MTZ for 24 hours. After treatment, eGFP+ cells degenerate (Fig 1D), nuclei in the outer nuclear layer (ONL) adjacent to ablated RPE become disorganized (Fig 1D’) and photoreceptor outer segment morphology is disrupted (Fig 1D”). Degeneration of eGFP+ cells was accompanied by the absence of pigmentation recovery after removal of PTU. To quantify this, eyes were enucleated from ablated and control larvae, and brightfield images were taken to provide an en face view of the RPE (S1 Fig). Quantification of the mean pigment intensity showed that pigmentation in ablated eyes was significantly reduced compared to controls by 2dpi (p<0.0001).
To characterize the temporal dynamics of RPE and photoreceptor (PR) degeneration following MTZ treatment, sections were taken from larvae at 3, 6, 12, 18, 24 and 48 hours post-injury (hpi) and stained for TUNEL (Fig 2; S2 Fig). At 3hpi, TUNEL+ nuclei were detected in the RPE (S2A and S2B Fig) while the ONL appeared normal. At 6hpi, nuclear organization in the ONL began to deteriorate and by 12hpi, apoptosis significantly increased in the RPE (Fig 2E, p = 0.016) and ONL nuclei became delaminated. By 18hpi, apoptosis in the ONL increased significantly (Fig 2F, p<0.0001) and eGFP accumulated in blebs, a process which left regions of the RPE devoid of eGFP signal (S2G and S2H Fig). RPE apoptosis peaked at 24hpi (Fig 2E, p<0.0001). Apoptosis, while remaining significantly elevated when compared to controls, began to decrease in both layers by 48hpi (Fig 2E and 2F, p<0.0001 in RPE, p = 0.0301 in ONL). However, by 48hpi, all remaining eGFP signal was contained in irregular eGFP+ blebs, likely consisting of RPE cell debris. ONL nuclear lamination remained severely disrupted (S2I–S2L Fig). Non-transgenic siblings treated with MTZ showed no significant increase in apoptosis (S3 Fig).
To characterize degeneration further, RPE-ablated larvae were stained with markers for RPE (ZPR2) [46], red/green cone arrestin (ZPR1) [47], and F-actin (phalloidin) (Fig 2G–2O). In unablated larvae, nfsB-eGFP colocalized with ZPR2 in the central RPE, confirming fidelity of the transgene (Fig 2G). rpe65a:nfsB-eGFP was expressed in mature RPE cells while ZPR2 signal extended further into the periphery, labeling both mature eGFP+ RPE as well as less-mature eGFP- RPE closer to the ciliary marginal zone (CMZ) (Fig 2G). Between 1 and 3 days post-injury (dpi), changes to ZPR2 staining recapitulated disruption of eGFP+ RPE, including degeneration of the RPE cell body. ZPR1+ cones also began degenerating at 1dpi (Fig 2J), and F-actin bundles in photoreceptor outer segments became more diffuse and lost their perpendicular orientation (Fig 2M). By 3dpi, both eGFP and ZPR2 signals were absent from the central RPE, confirming degeneration of RPE in the central injury site (Fig 2I). PR degeneration in the central retina also peaked at this time, displaying aberrant cone morphology (Fig 2L), and significant degeneration of photoreceptor outer segments throughout the injury site (Fig 2O). Despite rigorous screening, some variability in ablation severity was observed, likely from variations in transgene expression and ablation efficiency. To mitigate variability, only larvae with high levels of eGFP signal disruption in the eye (severe ablation) were utilized in subsequent experiments. In severely ablated larvae, ablation-mediated degeneration reliably peaked between 1-2dpi (i.e. as in Fig 2).
Immunohistochemical data strongly supported RPE and PR degeneration following ablation and this was confirmed by transmission electron microscopy (TEM) analyses (Fig 3). In unablated larvae, central RPE cells containing pigmented melanosomes were easily observable (Fig 3A). The PR layer was also properly laminated and contained readily identifiable cone and rod outer segments (Fig 3A). Analysis of ablated larvae at 3dpi revealed severe degeneration of the RPE, which was occupied by debris that was either distributed throughout the injury site or collected in membrane-enclosed structures that may be macrophages (Fig 3C, arrow). Bruch’s membrane (BM) underlying the ablated RPE was also significantly thinner than in controls (Fig 3B, 3D and 3E; p<0.0001) and contained obvious gaps (Fig 3D). Consistent with defects detected by histology (Fig 2), the PR layer of ablated larvae was severely degenerated, showing reduced size and integrity of photoreceptor outer segments, and containing degenerated outer segment material and other cellular debris (Fig 3C).
Taken together, these data indicate that RPE ablation via the rpe65a:nfsB-eGFP transgene causes specific loss of central RPE cells in larval zebrafish, with morphological defects beginning at 3hpi and destruction of the central RPE peaking at 2dpi. Further, immunohistological analyses demonstrate that underlying photoreceptor cells also degenerate rapidly after RPE ablation.
Visual function of larvae was evaluated by analyzing the optokinetic response (OKR) to determine whether ablation of the RPE results in vision defects [48–50] (Fig 4). A cohort of ablated and control larvae were exposed to a rotating full-field visual stimulus at 1dpi, 2dpi and 3dpi, and visual responses were recorded (Fig 4A–4C). At 1dpi, ablated larvae exhibited a modest reduction in stimulus tracking gain relative to controls, and this reduction in gain became significant at 2dpi (Fig 4D, p = 0.0055) indicating that visual function is disrupted after ablation. By 3dpi, ablated larvae demonstrated a recovery of stimulus tracking gain (Fig 4C and 4D). Likely, this rapid recovery is due to new photoreceptors being generated from the continually proliferative CMZ (see below). Collectively, these data demonstrate that ablation of large swathes of mature RPE cells in rpe65a:nfsB-eGFP transgenics results in the rapid degeneration of underlying PRs and BM, and a loss of visual function.
As discussed above, a subset of RPE possess a latent ability to proliferate in vitro [36] and various degrees of RPE repair have been documented (e.g. [25,26,31,32,51,52]) but in none of these systems is the RPE able to recover a functional monolayer following a large injury. Zebrafish possess a remarkable ability to regenerate a multitude of tissues [37,38,53], but it is unknown if they can regenerate RPE. Thus, we analyzed the regenerative capacity of ablated larvae at 4, 6, 7, and 14dpi with ZPR2, ZPR1, and phalloidin (Fig 5). At 4dpi, ZPR2+ cells extended into the injury site (Fig 5D) and RPE pigmentation significantly increased when compared to 2dpi levels (S1 Fig), suggesting that RPE cells have begun to regenerate. Although ZPR1-labeled cones and photoreceptor outer segments remained degenerated in the central ablation site, morphologically normal ZPR1-positive cones reappeared in the periphery, and these were always in direct apposition to regenerated eGFP+ RPE (Fig 5E and 5F; S4 Fig). At 6dpi, morphologically normal eGFP+/ZPR2+ RPE cells populate the periphery and approach the central injury site (Fig 5G), and PR morphology improves in a similar pattern (Fig 5H and 5I; S4 Fig). Interestingly, ZPR2+/eGFP- cells always appeared at the advancing tip of the regenerating monolayer (Fig 5G). While the rpe65a:nfsB-eGFP transgene is expressed specifically in mature RPE, ZPR2 labels less-mature RPE, suggesting that these ZPR2+/eGFP- cells are RPE that have not yet fully differentiated. By 7dpi, the injury site was populated by ZPR2+ RPE (Fig 5J). Although ZPR1-labeled cones continued to possess aberrant outer segment morphologies compared to controls in the central retina at 7dpi (Fig 5K), photoreceptor outer segment architecture began to improve at this time (Fig 5L). By 14dpi, ZPR2+/eGFP+ cells populated the entire RPE layer, and these displayed proper RPE cell morphology (Fig 5M and 5N). While most ZPR1+ cones displayed proper morphology, ONL disorganization persisted, particularly in the injury site, where cones failed to align perpendicularly to the RPE (Fig 5N). Seven months post-ablation, the RPE of ablated larvae were morphologically similar to those in unablated siblings (Fig 6).
Regeneration appeared to proceed in a periphery-to-center fashion in fixed samples. We utilized optical coherence tomography (OCT) to quantify the spatial and temporal dynamics of RPE degeneration and regeneration in individual larva over time. The RPE in OCT images presents as a bright line due to the density and pigment present in intact tissue; in ablated eyes, the intensity of the signal decreases as a result of tissue disruption (Fig 7A and 7B). Intensity of RPE signal (backscatter) can be quantified by determining the pixel intensity at each position of the RPE; here, we quantified the intensity from the optic nerve to the dorsal periphery, and examined changes in intensity in individual larvae over time (Fig 7, S1–S6 Videos). This analysis revealed that backscatter was significantly decreased in ablated larvae compared to controls in the central-most three quintiles of the RPE at 1dpi, and that all but the central-most quintile recovered to unablated levels by 5dpi (Fig 7C, p<0.0001). These results further support a model in which RPE regeneration occurs in a peripheral-to-central manner.
To confirm that the regenerated RPE is morphologically normal, TEM analyses were performed on 14dpi larvae (Fig 8), a time point at which eGFP+/ZPR2+ RPE cells populate the injury site (Fig 5N). Both unablated and ablated RPE contained melanosomes (Fig 8A and 8C). Moreover, BM thickness was restored in ablated larvae (Fig 8B, 8D and 8E; p = 0.3402). Despite this apparent recovery, subtle differences still existed between regenerated and unablated RPE: RPE in the regenerated region appeared to contain more melanosomes and had thicker cell bodies (Fig 8A and 8C). Consistent with immunohistochemical results, at 14dpi ONL lamination was improved but not completely recovered. Taken together, these data demonstrate that larval zebrafish are capable of regenerating a functional RPE monolayer following widespread RPE ablation and that regeneration is rapid, occurring within 1–2 weeks post-ablation.
In larvae, the eye undergoes significant growth, making it possible that RPE regeneration is the result of a permissive growth environment rather than an ability of the RPE to regenerate per se. Thus, we determined whether RPE regeneration also occurs in the adult eye. Transgene expression in unablated adults is restricted to mature central RPE as it is in larvae (Fig 9A). At 3dpi, there were clear signs of RPE degeneration that mirrored those in RPE-ablated larvae, including disruption of cell body cohesion and deterioration of apical processes, as indicated by the aberrant expression of ZPR2 and eGFP (Fig 9B). Degeneration extended from the central RPE (Fig 9B”) to the periphery (Fig 9B’). At 7dpi, adults showed signs of RPE regeneration in the peripheral injury site, such as recovery of contiguous eGFP+/ZPR2+ RPE (Fig 9C, arrowheads; Fig 9C’), however central RPE had not yet recovered (Fig 9C”). By 14dpi (Fig 9D) and 35dpi (Fig 9E), adult zebrafish showed restoration of peripheral eGFP+/ZPR2+ RPE (Fig 9D’ and 9E’) as well as successful regeneration of central RPE with apically localized ZPR2 expression (Fig 9D” and 9E”), similar to sibling controls (Fig 9A, 9A’ and 9A”). Quantification of RPE recovery based on contiguous eGFP+/ZPR2+ expression showed significant degeneration occurred by 3dpi (p = 0.0286), and that RPE fully regenerated by 35dpi (Fig 9F). Taken together, these results demonstrate that the adult zebrafish is also capable of regenerating the RPE, and in a similar periphery-to-center mechanism as occurs in larvae (Fig 9B–9E, arrowheads). Given these similarities and the technical advantages of using larvae over adults (e.g. comparatively rapid regeneration, access to a large number of samples, the ease of in vivo imaging and genetic manipulations, and utility in high-throughput drug screens), we focused further efforts on characterizing the mechanisms underlying RPE regeneration in larvae.
The rate and periphery-to-center pattern of RPE regeneration suggest that regeneration is driven by cell proliferation, and not simply the expansion of individual RPE cells, a response noted in several systems after small RPE injuries [24,31,54]. Proliferative cells are a major component of regeneration in diverse tissues, and they often derive from a resident pool of progenitor cells, e.g. as in blood and skin [55,56], or from differentiated cells that are stimulated to respond to injury, e.g. as in heart and retina [53,57,58]. Moreover, RPE cell proliferation results from the loss of BM contact in several injury contexts, and pathologically, during PVR [24,30,59,60]. Thus, we hypothesized that uninjured peripheral RPE cells respond to injury by dedifferentiating and proliferating to replace lost tissue.
To test this hypothesis, we first performed 24-hour BrdU incorporation assays to characterize the total number of proliferative cells within the RPE layer throughout regeneration (Fig 10). Proliferative cells appeared in the RPE as early as 1dpi, largely appearing immediately adjacent to the CMZ or in the center of the ablation site (Fig 10G and 10U, p<0.0001). Between 2-3dpi, more proliferating cells localized to the center of the eye, within the injury site (Fig 10I), and the number of proliferative cells in the RPE peaked between 3-4dpi (Fig 10J and 10U). During this period, proliferative cells populated much of the central eye in ablated larvae, with many localizing adjacent to or within the injury site (Fig 10J and 10Q–10T). In contrast, unablated eyes showed eGFP+ RPE throughout the central RPE (Fig 10D and 10M–10P) and sparse BrdU+ nuclei (Fig 10M–10P). As regeneration continued, eGFP+ RPE cells appeared closer to the center of the injury site and the number of proliferative cells in the RPE layer decreased (Fig 10L and 10U), with most remaining proliferative cells localizing to the injury site. As expected, BrdU+ cells were also observed in the retina, and these are likely Müller glia-derived progenitor cells (MGPCs) generated in response to PR degeneration [58,61,62]. Quantification of the number of BrdU+ cells in the central retina demonstrated that the kinetics of retinal regeneration largely overlapped that of the RPE (Fig 10V).
Next, we sought to obtain greater spatial and temporal resolution in our analysis of proliferative cells in the RPE layer. Therefore, larvae 1-7dpi were exposed to short 2-hour pulses of EdU and subsequently eyes were enucleated, stained for ZPR2/EdU, and imaged to acquire en face views of the entire RPE (Fig 11). To quantify the spatial dynamics of RPE cell ablation and regeneration, we divided the RPE of each eye into four regions based on cell location and two markers of differentiated RPE: pigmentation level and ZPR2 staining. The four RPE regions were delineated as follows: (1) peripheral RPE cells are pigmented and dimly ZPR2+, (2) differentiated RPE cells are highly pigmented and ZPR2+ (3) transition zone RPE cells are lightly pigmented but ZPR2+ and therefore likely consist of differentiating RPE extending into the injury site, and (4) the injury site, which contains no identifiable RPE cells, and which is often filled with aggregates of what are likely GFP+ and/or ZPR2+ debris (Fig 11K). Using these criteria to quantify RPE layer composition, our analysis confirmed that a large proportion of the RPE degenerates rapidly after ablation (Fig 11F and 11L). Strikingly, these analyses also revealed that differentiating RPE cells form a transition zone as soon as 1dpi (Fig 11G inset), and newly-formed differentiated RPE reappear in the periphery at 2dpi (Fig 11H and 11L). As regeneration proceeded, ZPR2+ transition zone cells always appeared in the periphery, stretching between the region of differentiated RPE and the central injury site. Furthermore, the proportion of the RPE encompassed by the transition zone at each time point correlated with the proportion of differentiated RPE cells added at the following time point, strongly suggesting these transitional RPE cells differentiate into regenerated RPE (Fig 11L). These analyses confirm earlier analysis showing that new RPE is added to the peripheral injury site, and that regeneration of a pigmented ZPR2+ RPE is completed by 7dpi.
Analysis of EdU+ cells revealed that there are more EdU+/ZPR2+ cells in the peripheral RPE of ablated larvae at 0.5dpi and 1dpi, and though this increase did not achieve significance (Fig 11O, p = 0.076 and p = 0.078, respectively), cryosections of 1dpi eyes showed peripheral EdU+ZPR2+ cells similar to those observed after BrdU exposure (compare Fig 11M and 11N to Fig 10G). During these early time points, EdU+/ZPR2+ cells were largely restricted to the peripheral retina, with only a few EdU+ cells appearing in the injury site (Fig 11R). During intermediate time points, when RPE cells reappear and the transition zone extends centrally, EdU+/ZPR2+ cells were present in both. As regeneration proceeded, the density of EdU+/ZPR2+ in regenerated RPE decreased, while increasing in the transition zone (Fig 11P and 11Q). By 4dpi, proliferative cells were largely restricted to the center of the eye (Fig 11I), and the majority of the remaining proliferative cells were located either in the injury site or the transition zone. Interestingly, the transition zone and regenerated RPE contained an even mix of EdU+ and EdU+/ZPR2+ cells, which may suggest that some differentiated RPE cells remain proliferative in this region and continue to generate new EdU+/ZPR2- cells that later enter the transition zone and differentiate. As expected, proliferative cells were also observed in the CMZ, particularly during early time points (e.g. Fig 11A, 11F, 11M and 11N and Fig 10A–10H); however, there appeared to be fewer proliferative CMZ cells beginning at 2dpi. As part of our experimental paradigm, embryos were incubated in PTU until 5dpf, and therefore it is possible that PTU withdrawal elicits a proliferative response throughout the retina, or that CMZ proliferation may ordinarily decelerate starting at 7dpf. Since both ablated and unablated larvae have fewer proliferative CMZ cells at later time points, it is unlikely that this phenomenon is a critical factor influencing RPE regeneration. Taken together, these results strongly suggest that peripheral RPE respond to injury by proliferating, that proliferative RPE cells and/or their progeny move into the injury site, and that proliferation continues within newly-generated RPE cells adjacent to the injury site until the lesion is repopulated.
We were interested in the EdU+/Zpr2+ differentiated RPE and the possibility that they continue to proliferate after injury. Thus, to determine whether early-proliferative cells enter the injury site and continue proliferating, we pulsed ablated larvae with BrdU between 0-1dpi, and with EdU at 3dpi before fixation and analysis (Fig 12A and 12B). Transverse sections revealed a significant increase of BrdU+/EdU+ cells and BrdU+ cells within the RPE of ablated fish (Fig 12C, p<0.0001). Interestingly, BrdU+/EdU+ cells often appeared at the interface between pigmented RPE and the unpigmented injury site, and some appeared to be pigmented (Fig 12B”). We next sought to determine whether early-proliferative cells ultimately integrate into the regenerated RPE. To do this, we exposed ablated larvae to BrdU between 0-1dpi and fixed them at 7dpi for analysis (Fig 12D and 12E). Transverse sections revealed a significant increase of BrdU+ cells within the RPE (Fig 12E and 12J, p<0.0001). These data suggest that early-proliferative cells enter the injury site at the leading edge of the regenerating RPE layer and either continue proliferating or give rise to proliferative cells there. Were this the case, we hypothesized that early-proliferative cells would integrate into both the peripheral and central RPE layer, while later-proliferative cells would form RPE only within the central RPE. To assess this, we pulsed ablated larvae with BrdU at 3-4dpi or 5-6dpi before fixing at 7dpi (Fig 12F–12I). BrdU+ cells were distributed throughout the RPE after a 0-1dpi pulse, but became more restricted to the central RPE after 3-4dpi and 5-6dpi pulses (Fig 12K). This analysis demonstrated that proliferative cells at early time points were distributed throughout the RPE, while later-proliferative cells were restricted to the central RPE. Finally, to determine whether early-labeled proliferative cells ultimately differentiate into RPE by 7dpi, we quantified the number and centrality of BrdU+/eGFP+ cells in 7dpi larvae that had been pulsed with BrdU between 0-1dpi (Fig 12L and 12M). Our analysis revealed that significantly more BrdU+ cells in the RPE were eGFP+ than eGFP- (Fig 12L, p = 0.0005), and that BrdU+/eGFP+ cells preferentially integrated toward the center (Fig 12M, p<0.0001). In summary, these data indicate that early-proliferating cells in the RPE layer ultimately differentiate into regenerated RPE, and strongly suggest that these proliferative cells are located in the periphery and that they or their progeny migrate into the injury site.
Our results thus far provide the first demonstration in any model system that RPE can endogenously regenerate after widespread injury. Next, we wanted to leverage this in vivo system to begin to identify the molecular underpinnings of the regenerative response. Previous studies have identified Wnt signaling as a regulator of tissue regeneration in multiple contexts [63–69], including the retina [70–72], and possibly RPE [73]. Thus, we examined Wnt signaling to begin to gain mechanistic insight into the molecular mechanisms underlying RPE regeneration. To assess Wnt pathway activity after RPE ablation, we examined expression of the Wnt target gene, lef1 [69,74]. lef1 was upregulated in ablated larvae at 1dpi (Fig 13B and 13B’), but not in unablated siblings (Fig 13A and 13A’) or in sense controls (Fig 13C, 13C’,13D and 13D’). Closer analysis of lef1 expression in ablated eyes revealed transcripts distributed in and adjacent to the RPE layer (Fig 13B’), suggesting the Wnt pathway is activated post-ablation. We next utilized IWR-1, which stabilizes Axin2 and promotes destruction of ß-catenin [75], to determine if disrupting Wnt pathway components impedes RPE regeneration. Larvae were pre-treated 24 hours prior to ablation (4dpf/-1dpi) with 15μM IWR-1 or with a vehicle control (0.06% DMSO) and kept in drug or vehicle until fixation at 4dpi (the time at which peak proliferation is observed in the RPE layer (Fig 10U)). Quantification of BrdU+ cells/section revealed a significant decrease in proliferation in IWR-1-treated RPE when compared to controls (Fig 13E–13G, p<0.0001). Further, there was a noticeable lapse in recovery of a pigmented monolayer in IWR-1-treated larvae (Fig 13I, arrowheads) relative to DMSO controls (Fig 13H). ZPR2 staining overlapped with pigmented RPE in both ablated DMSO- (Fig 13K) and IWR-1-treated (Fig 13L) larvae, indicating the lapse in pigment recovery was not simply a pigmentation deficiency, but rather a failure of the RPE to regenerate. Quantification of percent RPE recovery indeed showed a significant decrease in the IWR-1-treated larvae (Fig 13J, p<0.0001). These data suggest that components of the Wnt signaling pathway may be involved in RPE regeneration.
The stimulation of endogenous RPE regeneration is an appealing possibility for treating degenerative RPE diseases. However, the development of such a therapy is constrained by the paucity of data regarding the cellular and molecular underpinnings of regeneration. While the mammalian RPE possesses a latent proliferative ability, the process by which RPE cells respond to damage by proliferating and regenerating a functional monolayer, remains largely unknown.
The development of an animal model of RPE regeneration following specific and widespread RPE damage is a critical first step towards elucidating the regenerative process. Here, we developed a zebrafish model to ablate mature RPE and assess its regenerative capacity. In this model, ablation of a large contiguous stretch of RPE led to apoptosis and degeneration of the majority of mature RPE, which was rapidly followed by BM and PR degeneration and loss of visual function. In comparison, most RPE injury/regeneration models create small lesions using non cell-specific injury techniques (e.g. debridement or laser photocoagulation; [32,76,77]), or ablate a diffuse subpopulation of RPE cells via sodium iodate [78–82]. In mouse, a genetic RPE ablation system expressing diphtheria toxin in a subpopulation of RPE did not cause BM degradation or RPE proliferation [54]. Indeed, many RPE injury models preserve an intact BM or spare large regions of RPE (e.g. [54,83]). In contrast, our zebrafish model creates RPE and photoreceptor degeneration, which more closely resembles defects observed in late-stage AMD, wherein RPE dysfunction and degeneration precedes PR loss [12,84] [85,86], and thus may represent a more clinically-relevant starting point than other extant models for studying RPE regeneration.
Remarkably, we found that zebrafish are capable of regenerating after such a severe injury: within 7-14dpi in larvae, and within 1 month in adults. To our knowledge, these data provide the first evidence of RPE regeneration after widespread injury in any model system. Mammals largely fail to regenerate a functional RPE monolayer following injury [25,26]. One exception to this is in “super healer” MRL/MpJ mice, which regenerate the RPE within ~30 days after administration of mild doses of sodium iodate that elicit degeneration of the central RPE [51]. Beyond this example, mammalian RPE are incapable of regenerating after severe injuries (e.g. 27–29, 31). Our zebrafish RPE ablation model differs significantly from Xenopus [87], newt [88,89], and embryonic chick [90–92] retinectomy models wherein the entire retina is surgically removed and subsequently regenerates from remaining RPE tissue that that transdifferentiates, proliferates, and regenerates retinal tissue. Studies in these models have focused on the RPE-to-retina transdifferentiation process, and RPE-specific regeneration remains unexplored.
We present data here demonstrating that both larval and adult zebrafish possess the capacity to regenerate their RPE. However, due to the technical advantages of using larvae in studying regeneration (i.e. rapid regeneration, large sample sizes, feasibility of in vivo imaging, utility of the available genetic toolkit, and the ability to perform high-throughput drug screens) we mainly focused on characterizing RPE regeneration during larval stages. In larvae, regenerated RPE appeared at the periphery of the injury site at 2dpi, and the entire lesion was repopulated with differentiated RPE cells within 1 week. Our data support the following model of larval RPE regeneration (Fig 14): injury-adjacent RPE expand into the injury site, where they encounter degraded BM and proliferate to form daughters that enter the injury site and differentiate into RPE. RPE commonly expand to fill territory vacated by lost RPE [24,54,93], and contact with a degenerated BM induces RPE proliferation in many contexts [24,30,35,94–96]. Supporting this, we found that early-dividing cells (0-1dpi) often appear in the RPE periphery, localize to the injury site during peak phases of regeneration, and ultimately form RPE that integrate into the regenerated RPE monolayer. Wholemount analyses indicated that proliferative cells appear in the peripheral RPE soon after injury, and proliferative cells differentiate into RPE in distinct zones: (1) newly differentiated injury-adjacent RPE, (2) a transition zone, containing actively differentiating RPE cells, and (3) the injury site, which contains cellular debris as well as some proliferative cells that do not yet express RPE markers. Further experiments are necessary to determine whether all injury-adjacent RPE are capable of proliferating in response to injury, or if proliferation occurs within a subpopulation. Several lines of evidence suggest the latter possibility, and highlight the important role played by peripheral RPE: in mouse, a subpopulation of mature RPE in the periphery remain in the cell cycle and respond to microscopic photocoagulation injuries by proliferating at a higher rate than central RPE [32,97], while experiments in pig have shown that peripheral RPE respond to debridement of central RPE by proliferating [98]. Indeed, preservation of the peripheral RPE is also a prerequisite for successful RPE regeneration in the MRL/MpJ mouse model, which fails to regenerate RPE when high doses of sodium iodate cause degeneration of both central and peripheral RPE [51,99]. Finally, the discovery of a subpopulation of RPE stem cells [36] suggests that an endogenous regeneration-capable population of RPE could exist in the human eye, and these might be analogous to injury responsive cells in zebrafish.
While we present strong evidence supporting a model in which regenerated RPE derives from injury-adjacent RPE, we cannot definitively establish the source of regenerated RPE without performing lineage tracing. In response to retinal injury, Muller glia proliferate and generate MGPCs that differentiate into new neurons [37,58]. MGPCs are present at identical time points and in regions adjacent to ablated RPE. While it is possible that MGPCs transdifferentiate into RPE, this ability is not supported by any published studies. Another possible source for regenerated RPE is the CMZ, which generates new neurons throughout the life of the animal [100–102]. It was recently shown that rx2+ stem cells in the CMZ generate both RPE and retinal neurons [103]. Thus, rx2+ cells in the CMZ could potentially respond to RPE injury by generating RPE in the periphery that migrate and proliferate within the injury site. Attempts at lineage tracing to date have been unsuccessful and therefore, it will be necessary to develop new genetic tools to unambiguously identify the source of regenerated RPE.
We were surprised by the recovery of the OKR at 3dpi, well before the complete regeneration of the RPE layer (Fig 4). The threshold number of PRs required for a positive OKR in zebrafish is unknown. However, since the OKR is elicited by a large-field stimulus, activating PRs throughout the eye, this early recovery could be driven by CMZ-derived PRs that integrate into the peripheral retina by 3dpi. Consistent with this hypothesis, we observed CMZ-derived BrdU+ PRs in the retinal periphery at 3dpi (Fig 12B). It is also possible that recovered PRs adjacent to newly regenerated RPE in the peripheral injury site contribute to the OKR before regeneration of the central injury is complete. Alternatively, it is possible that a sub-population of central photoreceptors is functionally compromised, but survives ablation of the RPE, and these recover function by 3dpi. More work will be required to elucidate which of these processes might contribute to the rapid recovery of visual function at 3dpi.
Wnt signaling plays a known role in multiple regenerative contexts [63–69], including in the eye [70–72]. We show here that the Wnt pathway is activated in a subset of cells after RPE injury and that chemical inhibition using IWR-1 impairs regeneration. While we have not identified the lef1-expressing cell type, two possible sources are: 1) apoptotic cells in the injury site and 2) cells of the immune system. First, there are notable similarities between lef1 expression and TUNEL staining in the ONL and the RPE, and we detect lef1 expression (Fig 13B and 13B’) at the same time that TUNEL+ cell numbers peak in ablated larvae (Fig 2D–2F). A previous study in Hydra showed that apoptotic cells are a source of Wnt3, and that this is required for head regeneration [104]. The immune system has also been shown to play a critical role in influencing the regenerative response [105–107] and Wnt signaling regulates the inflammatory properties of immune cells. Wnt signaling is also important for RPE development in vivo [108–110], and it is possible that Wnt activation is important at multiple time points during RPE regeneration for initiation and to stimulate progenitor proliferation and/or differentiation. Further work is required to distinguish between cells in which Wnt is activated and regenerating RPE, and ultimately, these experiments may determine how the Wnt signaling pathway modulates RPE regeneration.
Finally, as noted above, we observed signs of BM degeneration in the central injury site following ablation. That RPE ablation leads to BM breakdown may provide insight into the mechanisms underlying the initiation of CNV at the onset of exudative AMD. During CNV, choroidal endothelial cells penetrate the BM and grow into the subretinal space; whether this process is initiated by the degeneration of the choriocapillaris or the RPE remains controversial [105,111,112]. Our results suggest that the RPE is required for BM maintenance. A logical next step would be to determine if choroidal vasculature invades the subretinal space following degeneration, as this would provide evidence that RPE degeneration is causative of CNV. Additionally, we show that zebrafish RPE are capable of repairing the BM during regeneration. In human, the BM undergoes a series of changes during aging that are thought to underlie AMD pathogenesis and inhibition of RPE function, changes that are thought to underlie the barrier to development of successful RPE transplant therapies [113,114]. The mechanisms underlying BM repair in zebrafish may provide critical insights into improving transplant survival and reintegration in humans.
Zebrafish were maintained at 28.5°C on a 14-hour light/10 hour dark cycle. Embryos were obtained from the natural spawning of transgenic or wild-type parents in pairwise crosses. According to established protocols [45], embryos were collected and raised at 28.5°C in the dark until they reached appropriate ages for experimentation. rpe65a:nfsB-eGFP was propagated by outcrossing to AB-strain wild-type fish. All animals were treated in accordance with provisions established by the University of Texas at Austin and University of Pittsburgh School of Medicine Institutional Animal Care and Use Committees.
To establish the rpe65a:nfsB-eGFP transgenic, 0.8KB immediately upstream of the coding sequence of rpe65a was cloned into a p5E entry vector, and this was placed upstream of nfsB-eGFP using the Multisite Gateway Cloning system (Thermo Fisher, Waltham, MA) within a backbone vector enabling Tol2-mediated transgenesis. This construct was then inserted into the genome using Tol2 recombinase [115].
Embryos derived from rpe65a:nfsB-eGFP x AB crosses were maintained in PTU-containing (Sigma-Aldrich, St. Louis, MO) fish water between 1-5dpf, and dechorionated at 3dpf. At 5dpf, larvae were exposed to 10mM metronidazole (MTZ; Sigma-Aldrich), dissolved in fish water, for 24 hours in the dark. After treatment, larvae were removed from MTZ and allowed to recover in fish water for 48 hours. At this point, severely ablated embryos were selected based upon the disruption of eGFP signal in the RPE and lack RPE pigmentation. Adult zebrafish (age 11 months) were ablated using the same parameters as for larvae: 24-hour 10mM MTZ treatment in a light-blocking incubator. Post-MTZ treatment, adults were returned to a 14-hour light/10 hour dark cycle and allowed to recover for 48 hours. Post-recovery, all ablated adults and larvae were placed on the recirculating water system (Aquaneering, San Diego, CA).
Larvae were euthanized at 2, 4, or 7dpi and fixed in 4% paraformaldehyde (PFA) (Thermo Fisher) overnight. Eyes were enucleated and the lens was removed before mounting in 3% methylcellulose. Images were acquired at a 112X magnification on a Zeiss (Oberkochen, Germany) Axio Zoom.V16 microscope. MetaMorph Image Analysis Software (Molecular Devices, San Jose, CA) was used to normalize the background of each image, delineate a region of interest (ROI) around each eye, and measure the average intensity (total intensity within the ROI divided by the area of the ROI). To quantify pigment (gray intensity) rather than white intensity, values were inverted by subtracting the generated intensity value from 65,536 (the number of gray levels in a 16-bit image).
For BrdU labeling, larvae were incubated in system water containing 10mM BrdU (Sigma-Aldrich) for 24 hours at various time points. After incubation, larvae were either immediately fixed for analysis or rinsed with fresh system water and then allowed to recover until 7dpi. For EdU labeling, larvae were incubated in system water containing 500μM EdU (Thermo Fisher, #C10338) for 2 hours immediately prior to fixation with 4% PFA in PBS.
Larvae were euthanized with Tricaine (Spectrum Chemical, New Brunswick, NJ) before fixation in 4% PFA overnight at 4°C, and adults were euthanized with Tricaine before removal of the eyes. Larvae and adult eyes were then sucrose-protected and embedded in tissue-freezing medium (TBS, Inc., Waltham, MA) before being sectioned at 14μm (larvae) or 16μM (adults) on a Leica CM1850 cryostat. Sections were rehydrated in 1xPBS for 5 minutes, and blocked in 5% normal goat serum in PBS for 2 hours at room temperature. For BrdU imaging, sections were treated with 4N HCl for 8 minutes at 37 degrees for antigen retrieval. Sections were counterstained with 1:500 DAPI (Life Technologies, Waltham, MA) for 9 minutes at room temperature, washed 3X with PBS, and mounted with Vectashield (Vector Laboratories, Burlingame, CA). Images were obtained with a 40X objective on an Olympus (Tokyo, Japan) FV1200 confocal microscope. Antibodies used in this study include BrdU (Abcam, Cambridge UK, ab6326), ZPR2 (ZIRC, Eugene, OR), ZPR1 (ZIRC), Phalloidin (Thermo Fisher, A22284, 1:33 dilution). TUNEL (Roche, Mannheim Germany, #12145792910) and Click-It EdU staining (Thermo Fisher, C10618) were performed according to manufacturer’s instructions.
After fixation, larvae were prepared using a protocol adapted from [116]: larvae were rinsed in PBST and water before permeabilization using acetone (-20°C for 12 minutes), rinsing in PBST, 1mg/mL Collagenase (Sigma-Aldrich) (30 minutes at RT), and 2mg/mL Proteinase K (Thermo Fisher, cat#BP1700) (30 minutes at RT). Larvae were refixed in 4% PFA (20 minutes at RT), rinsed in PBST, blocked using PBDTX (PBS + 1% bovine serum albumin, 1% DMSO and 0.1% Triton X-100), and stained overnight in PBDTX at 4°C with primary antibodies listed above at 1:250. Larvae were rinsed PBDTX on a shaker at RT (4X for 15 minutes), incubated with secondary antibodies (1:250), rinsed with PBDTX again (4X for 15 minutes) and counterstained with DAPI (1:200 in PBS). Chemically etched Tungsten needles [117] were used to dissect the eye and remove the choroid, and RPE flatmounts were mounted on Superfrost slides in PBS immediately prior to imaging.
Whole mount in situ hybridization was performed as described [118] using a previously reported DIG-labeled RNA probe for lef1 [119]. Post-labeling, larvae were fixed overnight in 4% PFA at 4°C then sucrose-protected, embedded, sectioned (12μm), and mounted using DPX Mounting Medium (Electron Microscopy Sciences, Hatfield, PA). Images were obtained with a 40X objective on a Zeiss Observer.Z1 inverted microscope.
Embryos were fixed in fresh 4% glutaraldehyde, 2% PFA overnight at room temperature, stained in 2% osmium tetroxide (OsO4) and 2% potassium ferrocyanide and 2% uranyl acetate, and microwave-embedded with a modified reduced-viscosity Spurr-Quetol-651 resin using a BDMA accelerator (Electron Microscopy Sciences) via a 30, 50, 75, 100, and 100% resin/acetone infiltration series [120]. Samples were sectioned using a Leica Ultracut UC7 ultramicrotome at a thickness of 70nm, and imaged on a FEI Tecnai transmission electron microscope.
For quantification of BM thickness, three transverse sections of the eye including the optic nerve were taken from the central RPE injury site in n = 3 zebrafish. The central RPE was defined as the region between the optic nerve head and the intersection of the dorsal eye and brain, as this correlated both with the area of highest transgene expression in unablated larvae, and degeneration in ablated larvae. In each section, three images of the BM were collected at 20,500X magnification in each image, and in each image, 3 measurements of the BM were taken by drawing lines perpendicular to the retina connecting the retinal and choroidal surface of the BM. In summary, 3 lines were drawn per image, 3 images taken per section, and 3 sections taken per animal for a total of 27 BM measurements per larva.
Larvae were immobilized in 3% methylcellulose, oriented dorsal up and exposed to a full field rotating stimulus projected onto a screen (NEC, Itasca, IL) that encompassed 180 degrees of the stimulated eye’s field of vision. Responses were captured using infrared light (880nm; Spectrum, Montague, MI) through a Flea3 Camera (Point Grey Research, Richmond, BC, Canada) mounted on a dissecting microscope (Leica Microsystems, Wetzlar, Germany). Videos were recorded by FlyCapture software (FLIR, Richmond, BC, Canada) and quantified using custom MATLAB scripts [50].
Larvae were immobilized dorsal up in 3% methylcellulose and imaged with Optical Coherence Tomography (OCT) (~840nm; Leica Bioptigen R2210 Spectral Domain Ophthalmic Imaging System). After imaging, larvae were rinsed and transferred into petri dishes to recover. OCT scans were analyzed using Bioptigen software (InVivoVue; Bioptigen, Research Triangle Park, NC), FIJI, and MetaMorph.
Larvae were treated with 15μM IWR-1-endo (Sigma-Aldrich) or 0.06% dimethyl sulfoxide (DMSO; Thermo Fisher) as a vehicle control from 4dpf/-1dpi until 9dpf/4dpi with daily replenishing of water and treatment. To assay proliferating cells, 10mM BrdU (Sigma-Aldrich) was added for 24 hours prior to fixation at 9dpf/4dpi. BrdU immunohistochemistry was performed as described.
For quantification of apoptosis, TUNEL+ cells within the RPE layer and ONL were recorded in unablated and ablated larvae at 3, 6, 12, 18, 24 and 48hpi. For quantification of BrdU+ nuclei in the RPE layer, BrdU+ nuclei contained within the RPE monolayer or presumptive RPE space between the retina and choroid were counted. To quantify the number of BrdU+ nuclei in the central retina, a reference line was drawn between the proximal-most BrdU+ cell originating from the CMZ in the dorsal and ventral retina. All BrdU+ cells in all layers of the retina proximal to this line were then counted. To quantify the centrality of BrdU+ cells, an additional perpendicular line bisecting the line demarcating CMZ-generated BrdU+ cells was drawn. BrdU+ cells in the RPE were counted using criteria detailed above, and the angle of each cell relative to these lines (0° ≤ x ≤ 90°) was calculated using FIJI software [121].
For quantification of RPE following OCT imaging, three transverse images of the retina were captured per fish at each time point (MTZ- n = 10, MTZ+ n = 9). Using MetaMorph, a line was drawn on the RPE beginning just dorsal of the optic nerve and terminating at the dorsal periphery, and a linescan of pixel intensity was taken from this line. The average RPE pixel intensity from the optic nerve to the periphery was graphed using GraphPad Prism 7.04 for Windows (Microsoft, Redmond, WA) (error bars = SEM). To quantify positional differences, each graph was divided into quintiles and the area under the curve was calculated for each quintile and compared using the Student’s t test.
For quantification of adult RPE recovery, central sections from adult eyes (MTZ-, 3dpi, 7dpi, 35dpi n = 4, 14dpi n = 3) were obtained and the edges of the injury site were designated based on contiguous-expressed eGFP+/ZPR2+ (e.g. arrowheads in Fig 9B–9E). Measurements for: 1) intact RPE (contiguous eGFP+/ZPR2+ expression) and 2) degenerated RPE were made along the dorsal and ventral retinae using the line segment tool in FIJI (ImageJ). Measurements were made along the basal side of the RPE from the distal tip to the junction of the optic nerve head (ONH). Width of the ONH was omitted from these measurements to avoid variability between adults. Dorsal and ventral measurements were summed and adult RPE regeneration was represented as percent eGFP+/ZPR2+ RPE.
Normality of datasets was assessed by the D’Agostino-Pearson Omnibus test. When analyzing non-normal datasets, Mann-Whitney U test was utilized to determine the significance of differences between unablated and ablated larvae. In normally-distributed datasets, Student’s t-test was utilized. These analyses were performed by GraphPad Prism 7.0c Software for Mac OS (La Jolla, CA, www.graphpad.com) and Microsoft Excel 14.7.3 (Microsoft). All statistical analyses are included in S1 Table.
|
10.1371/journal.pgen.1000094 | Overlapping Protein-Encoding Genes in Pseudomonas fluorescens Pf0-1 | The annotated genome sequences of prokaryotes seldom include overlapping genes encoded opposite each other by the same stretch of DNA. However, antisense transcription is becoming recognized as a widespread phenomenon in eukaryotes, and examples have been linked to important biological processes. Pseudomonas fluorescens inhabits aquatic and terrestrial environments, and can be regarded as an environmental generalist. The genetic basis for this ecological success is not well understood. In a previous search for soil-induced genes in P. fluorescens Pf0-1, ten antisense genes were discovered. These were termed ‘cryptic’ genes, as they had escaped detection by gene-hunting algorithms, and lacked easily recognizable promoters. In this communication, we designate such genes as ‘non-predicted’ or ‘hidden’. Using reverse transcription PCR, we show that at each of six non-predicted gene loci chosen for study, transcription occurs from both ‘sense’ and ‘antisense’ DNA strands. Further, at least one of these hidden antisense genes, iiv14, encodes a protein, as does the sense transcript, both identified by poly-histidine tags on the C-terminus of the proteins. Mutational and complementation studies showed that this novel antisense gene was important for efficient colonization of soil, and multiple copies in the wildtype host improved the speed of soil colonization. Introduction of a stop codon early in the gene eliminated complementation, further implicating the protein in colonization of soil. We therefore designate iiv14 “cosA”. These data suggest that, as is the case with eukaryotes, some bacterial genomes are more densely coded than currently recognized.
| Sequenced bacterial genomes provide a vast resource for research fields such as pathogenesis, drug discovery, and microbial ecology. Once sequenced, the genes within a genome are predicted using computational and manual methods. An assumption underlying both approaches is that any given length of DNA encodes only a single gene. This concept has been challenged by findings in eukaryotic genomes, and in bacterial plasmids and viruses where it is known that some stretches of DNA specify both ‘sense’ and ‘antisense’ RNA molecules. In prokaryotic cells there is little information regarding the potential of the genome to code two genes within the same stretch of DNA. We show that in the bacterium Pseudomonas fluorescens Pf0-1, both strands of DNA are transcribed at six locations in the genome, and that at one of these locations (iiv14), two different proteins are specified by the same piece of DNA. At the iiv14 locus, we demonstrate that the newly identified gene (antisense to the predicted gene) functions to promote colonization of soil, and name this gene cosA. Our findings indicate that bacterial genomes have more genes than currently thought, and important genes that have escaped detection occupy the same stretch of DNA as known genes.
| The genetic basis for ecological success is not well understood, yet has practical and fundamental significance. The environmental versatility of P. fluorescens, coupled with secondary metabolism that enables strains to antagonize plant pathogenic fungi or degrade organic pollutants, makes this species an important and relevant model for investigating environmental fitness and applications such as biocontrol and bioremediation. Furthermore, insight into complex ecosystem interactions is enhanced by knowledge of fitness determinants of the individual players.
To expand the understanding of genes functioning to promote soil fitness, we examined gene activity by evaluating expression when introduced into soil [1]. It has been proposed that elevated gene expression in a particular environment likely contributes to fitness of that organism within that environment [2]. Consistent with this suggestion, several studies have identified niche-specific gene activation and subsequently demonstrated that mutations in some of those genes reduced fitness in the environment in question [3]–[6]. Using in vivo expression technology (IVET) to directly examine gene expression of P. fluorescens Pf0-1 in soil, we identified 22 genes which were up-regulated during growth in soil. Mutations were subsequently introduced into three of these genes, and in each case the mutant showed a defect in soil colonization. Interestingly, ten soil-induced antisense genes were discovered, none of which was predicted by computational annotation of the Pf0-1 genome sequence [1],[7]. These have previously been termed ‘cryptic’ genes as they had escaped detection by gene hunting algorithms [7]. Herein, these are termed ‘hidden’ or ‘non-predicted’ genes.
Although antisense transcription has been reported in eukaryotic systems [8]–[11], antisense genes in prokaryote genomes have received limited attention, usually as antisense RNA regulators [12],[13]. Previous IVET experiments have suggested the existence of additional sense/antisense transcriptional pairs in bacteria [e.g. 3,6], but these suggestions have not been further investigated.
We chose for further study six loci at which non-predicted antisense genes were reported opposite predicted genes. Here we report the confirmation of sense/antisense transcription at each of these six loci. We demonstrate that both the hitherto unknown gene iiv14, and the putative membrane protein gene found opposite, specify proteins. Further, we show that the gene iiv14 is important for colonization of soil, and therefore name the iiv14 gene cosA.
Our analysis of sense/antisense transcripts in P. fluorescens dramatically increases the number of experimentally verified sense/antisense pairs in bacteria. Our data suggest that bacterial genomes are more densely coded than currently known, and that key traits pertaining to microbial ecology can be specified by hidden genes found antisense to those predicted during genome annotation.
We previously reported the discovery of ten DNA sequences expressed during growth of P. fluorescens Pf0-1 in soil, that were antisense to predicted genes in the genome of Pf0-1 [1]. These ten antisense sequences are not physically linked to each other, and have no similarity to known protein-coding genes [1]. We carried out RT-PCR experiments at six loci using gene specific primers to generate the cDNA, and thus distinguish transcripts produced from the two DNA strands. In laboratory cultures we detected a basal level of expression from both the predicted coding and antisense sequences (Figure 1A), lower than that required for observable dapB reporter gene activity in the initial IVET screen and for survival on minimal medium [1]. In a control experiment at the rpoS locus, transcription was only detectable from the rpoS gene, not the opposite DNA strand (Figure S1). In all RT-PCR experiments, controls in which reverse transcriptase was omitted produced no products. The direct demonstration of transcription of the non-predicted antisense genes confirms their existence, and transcription of the predicted genes on the opposite strand indicates these are not simply mis-annotated.
The gene iiv14 was chosen for further investigation. The transcribed region of iiv14 identified by RT-PCR is within a potential open reading frame complementary to bases 1092441–1093457 in the Pf0-1 genome sequence, the conceptual translation product of which has no significant matches in GenBank (GenBank accession number CP000094). If the iiv14 transcript spanned the ORF, it would suggest that iiv14 could be translated. The transcription start site was mapped by 5′ RACE to 160 bp upstream of the candidate ORF. RT-PCR experiments with gene-specific primers were used to determine that the 3′ end of the transcribed sequence was at least 210 bp downstream of the presumed stop codon. In addition, the TransTermHP website (http://transterm.cbcb.umd.edu/tt/Pseudomonas_fluorescens_PfO-1.tt) showed a predicted terminator spanning bases 1092229–1092199 of the Pf0-1 genome (starting 193bp downstream of the presumed stop codon), with a confidence score of 41. Thus, the iiv14 transcript is at least 1388 nucleotides long, and the candidate ORF is within the iiv14 transcript (Figure S2).
To provide direct evidence for translation of the candidate iiv14 ORF, and the predicted opposite gene (Pfl_0939) (Figure 1B) we added codons for six histidines to the 3′ end of both putative genes, cloned each in plasmid pME6000 (about 15 copies per cell), and transferred these to Pf0-1. Both constructs included the ORF plus upstream sequence likely to contain the native promoter (747 bp for iiv14 and 148 bp for Pfl_0939). Inclusion of the native promoter and expression in Pf0-1 ensured that observed proteins were not artifacts resulting from expression in a heterologous host or under control of a non-native promoter. Western blot analysis using an antibody directed to polyHis demonstrated that both the iiv14 ORF and Pfl_0939 specify proteins, of approximately the expected size (Figure 1C and D). The iiv14 protein was 37 kDa, consistent with the predicted molecular weight (MW) of 38.788 kDa, while the MW of the Pfl_0939 protein was about 70 kDa, slightly less than the expected 80.745 kDa, but not unusual for membrane proteins. Consistent with the fact that iiv14 is upregulated in soil and only expressed at a basal level in laboratory culture, the His-tagged iiv14 protein had to be purified from 1L of culture and concentrated prior to western blot analysis, and films were exposed for 14–16 hours to obtain a clear signal.
The proposed translation product of iiv14 has no significant matches in BlastP searches of GenBank, providing no functional clues. We therefore sought to examine its importance in a sterile soil growth assay, where its expression is elevated. The iiv14 ORF was deleted by SOE PCR [14] and allele exchange as described [1]. Because iiv14 and Pfl_0939 overlap, the deletion results in a double mutant. Further, deletion of the iiv14 ORF also removes the first 44 bases of Pfl_0940, probably rendering it non-functional. Growth of this deletion mutant in soil was monitored by periodic sampling of colony forming units. The iiv14/0939/0940 mutant was unable to colonize soil as rapidly as Pf0-1 (p<0.001) (Figure 2A, columns 1 and 2). However, between the first and second days when the wild-type had already approached maximum colonization density, the population of the Pf0-1Δiiv14 strain increased 1000 fold (Figure 2A, column 4), such that the cell numbers of both strains were approximately equal after two days of growth. Thus, the iiv14/0939/0940 deletion affected the early part of soil colonization, rather than soil survival per se. In laboratory medium (PMM), both Pf0-1 and the mutant showed approximately equal doubling times (65 min).
For the soil growth experiments, cells grown in laboratory culture were used as the inoculum. Thus, the slow initial growth of Pf0-1Δiiv14 could potentially be explained by a defect in adapting from laboratory media to soil. We therefore transferred wild-type and iiv14/0939/0940 mutant bacteria which had grown in soil for seven days into fresh soil, by diluting the previously colonized soil with fresh sterile soil. After one day in fresh soil, the increase of the mutant population was significantly (at least 100-fold) lower than that of Pf0-1 (p<0.005), demonstrating that the deletion of the iiv14 locus results in a soil colonization defect, not a defect in adapting to growth in soil after growth in laboratory culture medium (Figure 2B). Over the period between 24 and 48 hours, the iiv14-locus mutant population increased more than the Pf0-1 population, as the latter had already approached its maximum density in the first 24 hours.
To be certain that the deletion of the iiv14 locus caused the reduced soil colonization, we used allele exchange to replace the deleted region with wildtype sequence. To achieve this, region 1092301–1093786 in the Pf0-1 genome, which spans iiv14, was cloned in pSR47s. The resulting plasmid was used in allele exchange as described [1]. Recombinants possessing wildtype sequence were confirmed by PCR. In soil colonization experiments, the replacement strain was indistinguishable from Pf0-1, confirming that the deletion was completely responsible for the colonization defect (data not shown).
Because the two genes iiv14 and Pfl_0939 overlap each other, deletion of one results in loss of the other. To test whether loss of iiv14 was sufficient to explain the soil colonization defect, we cloned the iiv14 ORF and upstream region into plasmid pME6000. Two versions of the complementation clone were constructed, one consisting of 1155bp upstream of the iiv14 ORF (called pME14CF1), and the other containing only 329bp of upstream sequence, including 169bp upstream of the iiv14 transcriptional start site (called pME14CF2) (Figure 3A). Neither of these clones includes the full-length Pfl_0939 coding sequence opposite iiv14. Thus, phenotypes attributed to the presence of the complementation clones are related to restoration of iiv14, not the opposite gene. P. fluorescens strains possessing the vector pME6000 grew slower than plasmid-free strains in soil. Therefore, complementation experiments were followed and analyzed on day two, rather than after one day in soil. The findings were compared to controls of Pf0-1 and Pf0-1Δiiv14 carrying the plasmid vector. As seen with plasmid-free strains (Figure 2), the Pf0-1(pME6000) population increased significantly more (about 10-fold; p<0.001) after one day than did the Pf0-1Δiiv14(pME6000) population (not shown). After two days, the Pf0-1(pME6000) population had increased more than the iiv14 mutant carrying pME6000 (p<0.05) (Figure 3B, column 1 and 2), verifying that the phenotypes associated with plasmid-free strains were true for plasmid-bearing strains. Both pME14CF1 and pME14CF2 complemented the defect in Pf0-1Δiiv14 (Figure 3B; compare column 2 with columns 3 and 4). Relative to Pf0-1Δiiv14 harboring the vector alone, the population of complemented strains increased significantly after two days in soil (p<0.005). To further test whether iiv14 or Pfl_0939 was important in soil colonization, the plasmid constructed for expression of His-tagged Pfl_0939 protein (pME0939His) was utilized in complementation experiments. This plasmid lacks the coding sequence for the first 12 amino acids of iiv14, so only Pfl_0939 protein is made. Unlike pME14CF1 and pME14CF2, the pME0939His failed to restore soil colonization, demonstrating that the defect is not due to the loss of Pfl_0939 (data not shown).
As described above, deletion of iiv14 also removed 44 base pairs of the Pfl_0940 coding sequence. Plasmids pME14CF1 and pME14CF2 both contain the coding sequence for Pfl_0940 in addition to iiv14 (Figure 3A). To distinguish between the deletion of iiv14 and disruption of Pfl_0940 as the cause for the soil colonization defect, we constructed two additional complementation clones in pME6000, called pME940c1 and pME940c2, both of which include the full length of Pfl_0940, but not iiv14 or Pfl_0939. (Figure 3A). Neither of these clones was capable of restoring the soil growth phenotype of the iiv14/0939/0940 mutant (data not shown). Taken together, the complementation experiments using plasmid-based constructs demonstrate that of the three genes at the iiv14 locus, it is the non-predicted iiv14 that is important in colonization of soil, and was designated cosA.
Having demonstrated that cosA is important for soil colonization, and that the gene specifies a protein, we created a nonsense mutation in the gene to determine whether it was the cosA-specified protein or some other feature of the sequence that was important for soil colonization. Codon 17 of the cosA gene was changed from AAG to TAG, after which the mutated DNA region was cloned to create plasmids that were identical to pME14CF1 and pME14CF2, apart from the nonsense mutation. The AAG to TAG change in cosA results in a silent change from leucine codon CTT to leucine codon CTA in the Pfl_0939 sequence on the opposite strand. In the sterile soil colonization assay, the mutation-containing plasmids were unable to complement the colonization defect, relative to Pf0-1(pME6000) (p<0.01) (Figure 3B, columns 5 and 6).
Deletion and complementation experiments (above) demonstrated that cosA specifies a soil colonization factor. Multiple copies of cosA (on the complementing plasmids) accelerated soil colonization by wild-type Pf0-1. After one day in sterile soil, the population of Pf0-1 carrying pME14CF1 increased significantly more than that of Pf0-1 carrying the vector alone (p<0.05) (Figure 3C, columns 1 and 2). Over the first two days in soil, the median increase of the Pf0-1(pME14CF1) population was more than 15 times that of Pf0-1(pME6000), while the median increase for Pf0-1(pME'14CF2) was about 10 times that of the control (p<0.05 for both) (Figure 3C, columns 4–6).
When Pf0-1ΔcosAkmr was introduced into sterile soil in competition with wild-type Pf0-1Smr, the mutant was not competitive during the first day, and was unable to increase its relative population over subsequent days (Figure 4). This result is in contrast to competitions between differently marked wild-type strains, where each makes up 50% of the soil population after co-inoculation with equal numbers (not shown). The proportion of Pf0-1ΔcosA in the population did not decline over time, confirming that the fitness defect of the mutant is confined to the early colonization period.
Our results demonstrate that there are at least six loci in P. fluorescens Pf0-1 at which both annotated (‘sense’) and ‘antisense’ DNA sequences are transcribed. Further investigation of a chosen example provided conclusive evidence for overlapping protein-coding genes specified opposite each other by the same stretch of DNA. Importantly, a role for this locus in soil colonization was identified, and of the two overlapping genes present, it was the novel non-predicted gene cosA that was required for efficient soil colonization.
These findings suggest that current genome annotations provide an incomplete view of the genetic potential of a given organism, a proposal that is not without precedent in prokaryotic biology. Genes specifying the small ncRNAs affect processes including transcriptional regulation, mRNA stability, and chromosome replication [15]. These are still difficult to predict ab initio in organisms for which there is little information on promoter consensus sequences, although new computational approaches have recently become available [16]. In the prokaryotic horizontally mobile elements, antisense RNA has long been known to control copy number in plasmids, and play a role in bacteriophage development [reviewed in 17]. For prokaryote chromosomal genes, both trans- and cis-encoded antisense RNA regulators are known: e.g. control of glnA in Clostridium acetobutylicum [18], ompF in E. coli [19], and the photosynthesis gene isiA in Synechocystis sp. PCC 6803 [13].
In eukaryotes, the concept that genomes include numerous sense/antisense gene pairs is becoming increasingly obvious with genome-wide transcriptional studies in yeast [8] and Arabidopsis [10]. Antisense transcripts have been implicated in eye development [20] and control of entry into meiosis in yeast [21]. However, discussion of antisense transcription is limited to possible regulatory roles for antisense RNA [e.g. 8], without consideration of the possibility that they may specify proteins. Genome annotations do not routinely predict the existence of two protein-coding genes on opposite DNA strands, and in fact normally deliberately eliminate predicted overlaps. Moreover, small protein-coding genes can be missed by predictive algorithms. For example, the blr gene in E. coli specifies a 41 residue protein, and was discovered in a sequence believed to be intergenic [22].
The fact that antisense genes have been implicated in important biological functions indicates that more attention should be given to this emerging class of genes. Because they are difficult to predict with existing algorithms, experimental techniques should be adapted to allow their inclusion in genome-wide surveys. Tiling array technology is not biased toward predicted genes, and thus can be used to reveal the existence of non-predicted transcripts, such as those found antisense to known genes. Proteomic studies using accurate mass tags have the capacity to identify any and all proteins produced by a given organism. If data from such experiments are analyzed in an unbiased way, proteins produced from non-predicted genes will be identified. Finally, genetic techniques such as the IVET approach that led to the discovery of the iiv sequences described here [1] are well-suited to the discovery of novel genes. We have argued previously [7] that the frequency with which antisense genes are detected by promoter trapping strongly suggests that they represent real genes. The added advantage of IVET-like experiments is that they provide information regarding up-regulation in a particular environment, which yields clues as to function.
The exact role for cosA in colonization of soil is currently unknown. The cosA deletion mutant has no growth defect in laboratory culture, yet is impaired in soil colonization. Pf0-1 strains possessing the complementing cosA containing plasmids are more rapid colonizers of soil than the control strains (Figure 3C), but do not grow faster than control strains in laboratory media. In fact, Pf0-1 carrying the plasmid pME14CF1 forms colonies on agar surfaces considerably less quickly than control strains.
Two proteins, SetA (20 kDa) and SetB (7 kDa), [23] are thought to be encoded opposite pic, which specifies a serine protease [24] on a pathogenicity island in Shigella flexneri serotype 2 strains [25],[26] and in enteroaggregative E. coli [24],[27]. Thus, this example of probable antisense protein-coding genes evolved in the context of a horizontally mobile element.
We have demonstrated that at one sense/antisense chromosomal locus (cosA), both the predicted ‘sense’ gene (Pfl_0939) and the unpredicted ‘antisense’ gene cosA specify proteins, and it is the non-predicted gene identified initially from gene-expression in soil studies which is important for colonization of soil in a model laboratory system. Thus, antisense genes may be more functionally diverse than simply making regulatory or antisense RNAs. The cosA/Pfl_0939 pair is the first demonstration of overlapping antisense protein-coding genes in a prokaryote genome.
Pseudomonas fluorescens Pf0-1 was used as wild-type [28]. The genome sequence of Pseudomonas fluorescens Pf0-1 is available under GenBank accession number CP000094. Pf0-1Δiiv14/cosA was made by deleting bases 1092441 to 1093457 in the Pf0-1 genome by SOE-PCR [14] and allele exchange using plasmid pSR47s [29], as described [1]. E. coli DH5α (F- φ80lacZΔM15 Δ(lacZYA-argF) U169 recA1 endA1 hsdR17 (rk−, mk+) phoA supE44 λ- thi-1 gyrA96 relA1) (Invitrogen, Carlsbad, CA) was used to propagate plasmids. E. coli S17-1 (recA pro hsdR RP4-2-Tc::Mu-Km::Tn7 λ-pir) [30] served as the donor strain in conjugations. P. fluorescens strains were grown in PMM [31] at 30°C, and E. coli strains were grown in LB at 37°C. Plasmid pME6000 [32] was used for complementation studies and to express His-tagged proteins. Streptomycin resistant miniTn7 was constructed by inserting a Smr cassette from pHRP315 [33] into pUCT-mTn7T [34], while kanamycin resistant MiniTn7 is carried on pHRB2 [35]. MiniTn7 constructs were used to introduce Kmr and Smr markers into P. fluorescens strains for competition experiments, as described [36]. Complementation plasmids pME14CF1 and pME14CF2 contain regions 1092301–1094612 and 1092301–1093786 of the Pf0-1 genome, respectively. Plasmids pME940c1 and pME940c2 contain regions 1093208–1093627 and 1093208–1093786, respectively. Plasmid pME14His contains bases 1092438–1094204 of the Pf0-1 genome, and has codons for 6-His introduced immediately upstream of the cosA stop codon. Plasmid pME0939His contains bases 1091084–1093421 of the Pf0-1 genome and has codons for 6-His introduced immediately upstream of the Pfl_0939 stop codon. Pf0-1 sequences in each clone were amplified by PCR. In both cases, the six histidine codons were added by inclusion in the 3′ primer used to amplify the desired sequences.
Recombinant DNA techniques were carried out as described [37]. Restriction and DNA modifying enzymes were purchased from Invitrogen Inc and New England Biolabs (Beverly, MA). Plasmid DNA was purified using the Qiaprep spin miniprep kit (Qiagen, Valencia, CA). Genomic DNA was prepared using Promega's Wizard Genomic DNA purification kit (Madison, WI). DNA was recovered from agarose gel slices using the Qiaex II gel extraction kit (Qiagen). PCR was carried out with Invitrogen Platinum Taq DNA polymerase High Fidelity. PCR products were cloned with pGEM-T Easy (Promega). Oligonucleotides were purchased from IDT (Coralville, IA) and DNA sequences were determined at the Tufts University Core Facility (Boston, MA).
The complementing regions from pME14CF1 and pME14CF2 were cloned into plasmid pHRB2, which is smaller than the pME6000 vector. Codon 17 was changed from AAG to TAG. This also causes a silent change from leucine codon CTT to leucine codon CTA in Pfl_0939. The nonsense mutation was introduced using the “round the horn” protocol as described (http://openwetware.org/wiki/'Round-the-horn_site-directed_mutagenesis). The DNA polymerase used was KOD Hot Start DNA polymerase (Novagen). Primers used were 14mutF (tgggcTagtccttcgggcttg), which contains the base change to create the nonsense mutation (upper case), and 14wtR2 (tgcctcgtgaaatcgccttcc). The nonsense mutation was verified in resulting plasmids by DNA sequencing. Two mutants of each complementing region were then each recloned into pME6000, and the DNA sequence was again verified. Each of the four resulting plasmids (two for each of CF1 and CF2) were transferred to Pf0-1ΔcosA, and then assessed for complementing ability in the soil assay as described below.
Soil growth and survival assays were carried out as described previously [1] using gamma-irradiated, sandy loam soil, of known composition [38]. Briefly, cultures were grown for 16 hours in minimal medium, and diluted to contain approximately 104 cfu/mL. One mL of diluted culture was mixed with 5 g of soil, achieving approximately 50% water-holding capacity. Cultures for competition experiments in soil were adjusted to equal A600 values prior to dilution and then 500 µL of each competitor were used to inoculate soil as above. Population increase was monitored over time by extraction of cells and cfu determination by colony counting on selective media.
P. fluorescens Pf0-1 RNA was isolated using an RNeasy Mini Kit, including the on-column DNaseI treatment (Qiagen). The RNA was then treated with RQ1 DNaseI (Promega) for 1h at 37°C, and re-purified using a Qiagen RNeasy column. For RT-PCR experiments, cDNA was synthesized from 500 ng of total RNA using Superscript III (Invitrogen) and a gene specific primer, at 52°C for 1 h. The cDNA was amplified by PCR using the gene specific primer, and an appropriate partner primer. All RT-PCR experiments were carried out with a negative control consisting of a reverse transcriptase-free reaction.
5′ RACE was carried out using Invitrogen 5′ RACE system as recommended. Gene specific primers were 14RT-R (5′-ggcctgctgatctttttcag), 14GSP2 (5′-tgttcctgcaaccgaattcg), and 14GSP3 (5′-gggtgaaaagctacctgcac). Products were cloned in pGEM-T Easy, and sequenced using T7 and SP6 primers.
Proteins specified by cosA and Pfl_0939 were modified by addition of six histidine codons immediately before the stop codon, by PCR. In all cases, cultures were grown to A600 = 0.4, in PMM. His-tagged CosA protein was extracted and purified from P. fluorescens Pf0-1 carrying plasmid pME14His, using denaturing conditions, as described for E. coli in the QIAexpressionist handbook (Qiagen). Amicon Ultra-4 (10k) (Millipore, Billerica, MA) centrifugal filters were used to concentrate proteins. His-tagged Pfl_0939 was extracted from Pf0-1 carrying plasmid pME0939His, by sonication of cells, and solubilization in SDS buffer [37], resulting in a crude extract which was used in the western blots. Controls for each extraction (Pf0-1 carrying vector alone) were processed in the same way as each experimental preparation.
For western blot analysis, proteins were separated by SDS-PAGE (12.5% acrylamide for CosA-His protein, and 10% for Pfl_0939-His) and transferred to PolyScreen PVDF membrane (Perkin Elmer, Waltham, MA) using a Biorad Mini Trans-blot Cell (70 V, 1 h). Membranes were processed for immunodetection as described (QIAexpress Detection and Assay Handbook). Mouse monoclonal anti-polyHistidine antibody (Sigma, St. Louis, MO) was diluted 1∶10000 in fresh TTBS (17.2 mM NaCl, 5.1 mM KCl, 24.8 mM tris base, 22 mM HCl (to pH 7.4), 0.1% Tween 20)+3% BSA. HRP-linked antimouse IgG antibody (Cell Signaling Technology, Danvers, MA) secondary antibody was diluted 1∶10000 in TTBS plus 10% non-fat milk powder. Detection of antibodies was carried out using Western Lightning Western Blot Chemiluminescence Reagent Plus (Perkin Elmer). Luminescence was detected with Kodak BioMax MR film after 18 h of exposure. Protein molecular weights were estimated by comparison to the BenchMark Pre-Stained Protein Ladder (Invitrogen).
Data from soil experiments were analyzed by the Mann Whitney test using GraphPad Prism 4 software. |
10.1371/journal.pbio.1001171 | Memory Phenotype CD4 T Cells Undergoing Rapid, Nonburst-Like, Cytokine-Driven Proliferation Can Be Distinguished from Antigen-Experienced Memory Cells | Memory phenotype (CD44bright, CD25negative) CD4 spleen and lymph node T cells (MP cells) proliferate rapidly in normal or germ-free donors, with BrdU uptake rates of 6% to 10% per day and Ki-67 positivity of 18% to 35%. The rapid proliferation of MP cells stands in contrast to the much slower proliferation of lymphocytic choriomeningitis virus (LCMV)-specific memory cells that divide at rates ranging from <1% to 2% per day over the period from 15 to 60 days after LCMV infection. Anti-MHC class II antibodies fail to inhibit the in situ proliferation of MP cells, implying a non–T-cell receptor (TCR)-driven proliferation. Such proliferation is partially inhibited by anti–IL-7Rα antibody. The sequence diversity of TCRβ CDR3 gene segments is comparable among the proliferating and quiescent MP cells from conventional and germ-free mice, implying that the majority of proliferating MP cells have not recently derived from a small cohort of cells that expand through multiple continuous rounds of cell division. We propose that MP cells constitute a diverse cell population, containing a subpopulation of slowly dividing authentic antigen-primed memory cells and a majority population of rapidly proliferating cells that did not arise from naïve cells through conventional antigen-driven clonal expansion.
| The class of immune cells called CD4 T lymphocytes consists of two major cell types: naïve cells that have not yet participated in an immune response and memory cells, which are cells that have responded to antigen, expanded in number, and acquired new characteristics. These two cell types can be distinguished from one another because they display different cell surface marker proteins. In this paper, we argue that many—probably most—of the cells researchers generally characterize as memory cells on the basis of their surface markers are not authentic memory cells. True memory cells—the ones produced, for example, when we immunize a child against a disease—divide very slowly, whereas the bulk of the cells we generally characterize as memory cells divide very rapidly. Mice that have never been exposed to antigens have as many of these “memory-like” cells as normal mice have, implying that these cells arise by a process that does not require foreign antigen. Analysis of the sequence of the antigen recognition receptors on these “memory-like” cells indicates that their replication does not derive from a few cells or clones undergoing multiple rounds of proliferation, thus their division cannot be explained by conventional, antigen-driven clonal expansion. We conclude that this large population of “memory-like” cells has arisen by a mechanism independent of a response to foreign antigen, and that these cells may have a crucial biological function.
| Peripheral non-Treg CD4+ T cells are often divided into two major subpopulations that can be designated naïve-phenotype (NP) and memory-phenotype (MP) cells, respectively [1]. In the mouse, MP cells are characterized by the expression of high levels of CD44 and low levels of CD45RB; they lack Foxp3 and high levels of CD25. MP cells may be either CD62L dull or bright [2]. It is generally assumed that MP cells constitute the aggregate of all antigen-specific memory cells; that is, of all cells that have expanded in response to antigenic stimulation. However, there are some reasons to question the concept that all MP cells are indeed foreign antigen-experienced cells. MP cells proliferate rapidly; estimates of their proliferative rates in lymph nodes range from 4% to 10% per day [2],[3]. By contrast, T-cell receptor (TCR) transgenic [4],[5] or polyclonal [5],[6] CD4 T cells that had responded to immunization with cognate antigens or infection proliferate at <1% to 2.5% per day when examined after the initial expansion and contraction phases have been completed [7]. The proliferation of antigen-primed CD4 T cells is largely driven by cytokines rather than through TCR stimulation [8]–[14]. What drives the rapid, apparently spontaneous, proliferation of MP under normal conditions is unknown, although when transferred to lymphopenic recipients, their proliferation is burst-like (i.e., they divide multiple times in a relatively short period) and appears to be driven by TCR-mediated stimulation.
Understanding the proliferation of MP cells has also been of considerable interest among those studying lymphocyte dynamics in chronic infections, particularly with lentiviruses, where proliferative rates of human or macaque MP cells in HIV- or SIV-infected individuals are much greater than those of comparable cells from noninfected individuals [15],[16]. Indeed, such rapid proliferation has been associated with the state of excessive inflammation that, in turn, has been regarded as a principal driver of the immunodeficiency of AIDS patients [17]–[19]. It has been suggested, on the basis of BrdU labeling and of measurement of Ki-67 expression in SIV-infected macaque CD4 T cells, that much of the proliferation of these MP cells represents recent burst-like divisions, presumably in response to antigenic stimulation, of cells that were undergoing the familiar pattern of clonal expansion and transition from central or effector memory populations to tissue-seeking effector cells [17],[20]. Although this mode of proliferation appears to be the case for SIV-infected macaques and presumably HIV-infected humans, whether it explains the proliferation of MP cells in normal individuals has not been determined. Recognizing that MP CD4 T cells constitute a large and heterogeneous population, we repeated previous experiments establishing the differences in proliferative rates of MP cells from those of authentic antigen-experienced memory cells and also compared the behavior and frequency of MP CD4 T cells in conventional and germ-free (GF) mice. In order to understand whether the proliferation of MP cells in situ (not in transfer models) was driven by antigen and was burst-like or by cytokines and was stochastic, we treated mice with anti-MHC class II antibodies or with anticytokine antibodies. Further, we reasoned that if the expansion of MP cells was burst-like, it should have originated from a small number of precursors and thus proliferating MP cells should have a much more limited TCR sequence diversity than MP cells that were not dividing.
Our results indicate that in situ MP cell division is driven largely by cytokines and not by TCR-mediated stimulation, that the diversity of the CDR3 regions of TCR β chains of particular VβJβ sets is similar in dividing and nondividing cells, and that conventional and GF MP cells are not distinguishable in either frequency, division rate, or, in a preliminary analysis, in sequence diversity. These results imply that the bulk of MP CD4 T cells in young adult mice differ in key respects from authentic antigen-driven memory cells.
To readdress the question of the relative proliferative rates of MP cells and antigen-specific memory cells, we first evaluated the use of Ki-67 as a measure of recent proliferation. C57BL/6 mice received BrdU in a single intraperitoneal (IP) injection and were humanely killed 24 h later or BrdU was administered in their drinking water and mice were humanely killed 3 d later. Figure 1A shows that 11% of CD44bright Foxp3− CD4 lymph node T cells evaluated 24 h after a single injection of BrdU were stained by an anti-BrdU antibody, confirming the rapid proliferative rate of these cells. All of these BrdU+ cells were Ki-67+ and, in addition, 25% of the CD44bright Foxp3− CD4 cells were Ki-67+/BrdU−, as anticipated, since Ki-67 is known to be expressed for a period of time after cells have completed their cycle.
When we examined cells from the mice that had received BrdU for 3 d (Figure 1A), we found that 35% of the cells were BrdU+, reaffirming their rapid proliferative rate. The great majority of the Ki-67+ cells were BrdU+, indicating that most cells do not retain Ki-67 expression for more than 3 d after their last division. Indeed, Pitcher et al. [21] reached a similar conclusion regarding Ki-67 expression as a result of analyzing the proliferation of PBMCs from SIV-infected macaques by simultaneous staining for BrdU and Ki-67 [21]. Accordingly, we used either Ki-67 or BrdU in different experiments; particularly, we have relied on Ki-67 in later experiments in which we examined TCR Vβ sequence diversity among dividing and nondividing MP cells. In those experiments, we also took advantage of the finding (Figure 1A) that Ki-67 mean fluorescence intensity (MFI) was highest in cells that had taken up BrdU during the previous 24 h.
We then compared the frequency of Ki-67+ cells among splenic MP cells and antigen-specific (tetramer+) CD4 T cells obtained 60 d after acute lymphocytic choriomeningitis virus (LCMV) infection (Figure 1B). The frequency of tetramer+ CD4 T cells in LCMV-infected mice is greater among splenic CD4 T cells than lymph node CD4 T cells, so we limited our evaluation to tetramer+ cells from the spleen and compared them to splenic MP cells whose proliferative rates are somewhat less than those of lymph node MP cells. In this experiment, 17%±3% of the splenic MP cells were Ki-67+. Among tetramer+ CD4 T cells (5.2% of the CD44bright CD4 T cells) obtained from mice infected 60 d earlier, only 7%±2% were Ki-67+. This finding implies that ∼2% of the tetramer+ cells divided each day, a frequency similar to the proliferative rates of antigen-specific memory CD4 T cells reported by others [5]–[7].
This difference in proliferative rates could be explained if MP cells have derived quite recently from NP cells and are still dividing relatively rapidly, while the tetramerpos memory cells induced by intentional immunization were examined 60 d after infection, and in experiments reported by others at least 40 d after infection/immunization, when their proliferative rates may have slowed considerably.
If this were the case, we might anticipate that authentic memory cells would be dividing substantially more rapidly when studied relatively shortly after infection. We assessed the expression of Ki-67 in tetramerpos and tetramerneg CD44bright CD25− CD4 T cells 15 d after LCMV infection. At that time, 7.7% (±0.3%) of the CD44bright cells were tetramerpos. Of these, 7.8%±1.9% were Ki-67+ compared to 18.9%±0.6% of the tetramerneg MP cells (Figure 1C, results from one of three mice). This experiment indicates that one cannot account for the differences in the in situ proliferative rate of MP cells and of the antigen-driven memory cells on the basis of the more recent priming of the former than the latter. We did verify that tetramerpos cells examined 6 d after infection were essentially all (97%) Ki-67+, indicating that these cells had undergone rapid proliferation as a result of antigenic stimulation. As we will show later, it is highly unlikely that most of the Ki-67+ MP cells represent a population in the midst of its antigen-driven expansion from naive or memory precursors.
The high proliferative rate of MP cells might be due to a distinct, small subpopulation dividing very rapidly while a large population divides slowly. We thought that explanation unlikely in view of the classic report by Tough and Sprent [2] that more than 60% of CD44bright CD4 T cells became BrdU+ during a labeling period of 30 d, implying that over that period of time the great majority of CD44bright CD4 T cells had divided at least once. We observed an even more rapid proliferation with ∼60% of MP (CD44bright Foxp3−) CD4 T cells having taken up BrdU in a 10-d labeling period (Figure 1D), again arguing that the high proliferative rate of MP cells is not a property of a small subpopulation of these cells.
The presence and proliferation of MP cells in GF mice needs to be considered in assessing the possible role of foreign antigens in stimulating the in situ proliferation of MP cells in normal animals. We reported that the proliferative rate of CD44bright CD25− cells in SW GF mice was ∼4% in 6 h and was no different from that of such cells from conventional SW mice [3]. This implies that the generation and proliferation of MP cells can be achieved in mice with very limited antigenic load. To examine this in greater detail and in the mouse strain that was being studied in our experiments, we injected BrdU into conventional and GF C57BL/6 mice and evaluated the frequency of BrdU+ cells 6 h later. BrdU+ cells constituted 4.7% of the GF CD44brightFoxp3− lymph node CD4 T cells and 5% of the same cells from conventional donors. The proportion of Ki-67+ MP lymph node cells was 38.7% in GF mice and 38% in conventional mice (Figure 2A). The absolute numbers of total lymph node CD4 T cells, of CD44bright CD4 T cells and of Foxp3− CD44bright CD4 T cells were not different and thus there was no difference in the numbers of Ki-67+ or of BrdU+ cells. This finding was the case for both peripheral and mesenteric lymph node cells (Figure 2B and 2C). Thus, numbers and proliferative behavior of MP cells is similar in mice with very limited antigen-exposure (i.e., GF mice) to that in conventional mice, raising the possibility that a substantial proportion of MP cells in conventional mice may develop through a process other than foreign antigen-driven activation and expansion.
One approach to evaluating the importance of antigen stimulation in T-cell dynamics is to determine whether proliferation can be blocked by anti-MHC class II antibodies [22]. To that end, we utilized FcγRγ−/− mice so that the anti-class II antibody would not block T-cell responses by elimination of antigen-presenting cells. In such mice, anti-class II antibodies powerfully inhibit antigen-specific in vivo responses. We transferred CD45.1 OT-2 cells to CD45.2 FcγRγ−/− C57BL/6 mice, treated the recipients with the anti-class II antibody Y3P (1.8 mg IP) and 1 d later immunized them with an ovalbumin peptide plus LPS. BrdU was given to these mice in drinking water from the time of immunization and mice were humanely killed 3 d later. Mice treated with mouse immunoglobulin G (IgG) rather than Y3P showed expansion of the transferred cells. 68.5% of these cells were BrdU+ and 78.2% were Ki-67+. By contrast, in the treated mice, there was no expansion of the transferred cells when compared to unimmunized mice and only 6% were BrdU+ and 12.5% Ki-67+ (Figure 3A and 3B). In the same animals, the frequency and number of MP cells that were BrdU+, Ki-67+ were not affected by treatment with Y3P (Figure 3A and 3B). In a separate experiment, in which BrdU was administered to nonimmunized FcγRγ−/− mice for 6 h prior to humanely killing, Y3P had no effect on the frequency or numbers of BrdU+ or of Ki-67+ MP cells (Figure 3C).
Normal C57BL/6 mice were either untreated or received anti–IL-15, anti–IL-7Rα, or anti–IL-2 antibody on day 1 and day 4 and were humanely killed on day 7. There was no effect on total numbers of CD44bright cells in lymph nodes but the numbers of Ki-67+ cells was significantly reduced among recipients of anti–IL-7Rα (Figure 4), indicating that at least a portion of the proliferative response of MP cells depended on IL-7, or conceivably, thymic stromal lymphopoietin (TSLP). Neither anti–IL-15 nor anti–IL-2 had a significant effect.
Transfer of MP CD4 T cells into Rag2−/− recipients results in burst-like proliferation such that the great majority of the cells present 6 to 7 d later have undergone 7 or more divisions, as judged by carboxyfluorescein diacetate succinimidyl ester (CFSE) dilution [5],[23]–[26]. We carried out such an experiment and confirmed that the majority of the transferred CD4 T cells present 6 d later had undergone multiple divisions. Neither anti–IL-15 nor anti–IL-7Rα had any inhibitory effect, but Y3P almost completely inhibited proliferation (Figure 5A), indicating that the expansion of cells in this lymphopenic setting required recognition of MHC and presumably of peptide/MHC complexes.
If MP cells are normally undergoing proliferation bursts, we reasoned that the dividing cells would have originated from relatively small cohorts of cells that would each go through many divisions. This should have the effect that the dividing cells would have relatively limited sequence diversity compared to nondividing MP cells in the same individual. To test whether this thesis is correct in a system in which we could be confident that cells had undergone multiple cell divisions, we transferred 1 million CFSE-labeled CD44bright CD25− CD4 T cells into Rag2−/− C57BL/6 recipients. We humanely killed the mice 3 d after transfer so that the proportion of cells that had not divided would still be substantial; by 7 d, the cells that had divided multiple times completely dominate the distribution. We purified, by cell sorting, CD4+ CD44bright cells that had completely diluted their CFSE and those that had retained their full amount of CFSE, implying they had not divided. We limited our sequence analysis to one Vβ/Jβ set, Vβ2/Jβ1.1. We chose Jβ1.1 since, as the most proximal Jβ, those TCRβ gene segments that have retained Jβ1.1 are very likely to be using it in their rearranged Vβ chain. Limiting the range of Vβs studied also allowed us to sample a larger fraction of the repertoire among those TCRs with a relatively small number of sequences than would have been the case had we tested all Vβs.
Among 42 sequences of transferred MP CD4 T cells that had not divided (CFSEhigh), 37 were unique; one sequence occurred three times, three occurred twice, and 33 but a single time (Figure 5B). By contrast, among 42 sequences from MP CD4 T cells that had divided seven or more times (CFSElow), there were only three unique sequences, occurring 12 (CASSHDSKNTEVFFGKG), 14 (CASSQEAGRGTEVFFGKG), and 16 (CASSQRGGKGVFFGKG) times, respectively. Interestingly, two of the sequences from the cells that divided multiple times were also found among the cells that had not divided, but in these cases they were represented only two or three times. This experiment validates our expectation that a cell population undergoing burst-like division should have a relatively limited repertoire that should be distinguishable from that of cells of a comparable phenotype that had not divided. It further indicates that only a subset of the MP cells undergo burst-like proliferation in a lymphopenic environment.
To examine the repertoire complexity of proliferating MP cells in situ in comparison to that of nondividing MP cells, we took advantage of our observation that the MFI of Ki-67+ cells was highest among cells that had recently divided. We could not utilize pulse labeling with BrdU because the process of staining for BrdU expression requires the use of DNase, interfering with DNA sequence analysis. We sorted CD4+, Foxp3−, CD44bright, Ki-67bright cells and CD4+, Foxp3−, CD44bright, Ki-67negative cells and sequenced both Vβ2/Jβ1.1 and Vβ4/Jβ1.1 CDR3 segments from two individual mice (Figure 6). We obtained 320 Vβ2/Jβ1.1 sequences from the Ki-67negative cells of mouse 1 and 208 sequences from the Ki-67bright cells of this donor. We plotted sequences against their relative representation as indicated by their percentage frequency (Figure 7). We also listed the number of sequences that occurred one to five times or more than five times in the embedded table. Among the 320 sequences from the Ki-67negative MP cells, 133 were unique. Of these, 103 occurred only once or twice. However, some sequences were much more frequent. The sequence CASSRTGGNTEVFFGKG comprised almost 6% of the Vβ2/Jβ1.1 sequences from Ki-67 negative cells, and five other sequences constituted 3% or more of all the sequences. The pattern of sequence expression among the Ki-67bright cells was quite similar to that of the Ki-67negative cells (Figure 7). Of the 208 sequences, 138 were unique. Of these, 128 occurred once or twice. One sequence occurred ∼6% of the time. Interestingly, this was the same sequence (CASSRTGGNTEVFFGKG) that occurred most frequently among the Ki-67negative cells, suggesting that it represents a clone whose dividing and nondividing members reflect prior clonal expansion, possibly from naïve cells, rather than a process of ongoing burst-like proliferation. However, we cannot exclude the possibility that the high frequency of this sequence represents a late phase of a clonal expansion episode when some of the cells have already stopped dividing and have lost Ki-67 expression.
We also sequenced Vβ4/Jβ1.1 CDR3 regions from the MP cells of the same mouse (Figure 7). In this case, we obtained 79 sequences from Ki-67negative cells and 90 sequences from Ki-67bright cells. The results were quite similar to those observed from the sequences of Vβ2/Jβ1.1 Ki-67negative and Ki-67bright cells. We had 56 unique sequences among those from Ki-67negative cells; of these, 54 occurred once or twice. Two sequences were more common, both representing more than 12% of the total sequences. Among the 90 Vβ4/Jβ1.1 sequences from the Ki-67bright cells, there were 53 unique sequences of which 48 occurred once or twice. One sequence (CASSIFESIGKG) constituted ∼25% of all the sequences; however, that sequence was one of the two that each constituted 12% of the sequences from the Ki-67negative cells, implying that these dividing cells may be accounted for by the over-representation of the same particular clone and may not represent an ongoing burst-like expansion. As stated above, we cannot rule out the possibility that this represents a just-completed burst in which some cells have already ceased dividing. There are three sequences that occurred three times and one that occurred four times among the Ki-67bright cells. Two of those had not occurred among the Ki-67negative sequences and thus could conceivably represent burst-like proliferation. However, the majority of the dividing Ki-67bright cells do not appear to have recently expanded from a precursor by multiple cell divisions.
We sequenced Vβ2/Jβ1.1 and Vβ4/Jβ1.1 CDR3 regions from Ki-67negative and Ki-67bright cells from a second donor (Figure 7, mouse 2). Although the results were generally similar, in this mouse, there were sequences among the Ki-67bright cells that occurred multiple times but which were not found among the Ki-67negative cells, suggesting they may represent burst-like proliferation. Indeed, among the Vβ2/Jβ1.1 sequences from mouse 2, two sequences comprised 18% and 17% of the sequences but were only observed once among the Ki-67negative cells and two sequences observed in 5% of the Ki-67bright sequences and not in the Ki-67negative. However, even among the 77 Vβ2/Jβ1.1 Ki-67bright CDR3 sequences from this mouse, 31 were unique and 23 occurred only once or twice. Among the Vβ4/Jβ1.1 sequences, there was one from the Ki-67bright cells that occurred in 9% of the sequences but was not found among the Ki-67negative cells, suggesting that it may represent burst-like expansion. Here too, there were many unique sequences that occurred rarely. Of the 75 Vβ4/Jβ1.1 sequences from Ki-67bright cells, there were 47 unique sequences of which 43 occurred only once or twice. Furthermore, it should be pointed out that we occasionally observe a dominant sequence among those from Ki-67negative cells (in this mouse, 50% of the sequences from KI-67negative are CASSFESIGKG) that is not (or is only infrequently) represented in the sequences from the Ki-67bright cells.
Overall, we conclude that the complexity of Vβ2/Jβ1.1 and Vβ4/Jβ1.1 CDR3 sequences from Ki-67bright cells cannot be distinguished from that of the Ki-67negative cells. Estimating the maximum percentage of Ki-67bright cells that could have been part of bursts from these data is nonetheless not simple. Taking the most inclusive view, it could be argued that all Ki-67bright sequences that are represented many times should be considered as having originated from burst-like clonal expansion during the period immediately before the mice were humanely killed. To obtain an estimate of the frequency of such events, we summed all the sequences that occurred four or more times in the Ki-67bright cells. In the four groups studied, there were 119 sequences among those that occurred four or more times. Since the total number of Ki-67bright sequences analyzed was 450, this implies that ∼25% of the sequences may represent Ki-67bright cells that were part of burst-like clonal expansion. If we exclude those 88 sequences that occurred multiple times in both the Ki-67negative and Ki-67bright groups, then the proportion of dividing cells that are part of burst-like proliferation is ∼7%. Thus, the majority of dividing cells do not appear to be part of an ongoing process of burst-like clonal expansion from a limited number of precursors, which was the case when we examined the burst-like division of the Vβ2/Jβ1.1 MP cell population that occurred upon transfer to severely lymphopenic recipients.
We also examined sequences from Ki-67negative and Ki-67bright Vβ2/Jβ1.1 CD44bright cells from lymph nodes of GF mice. Overall, the patterns of sequence distribution were remarkably similar to those of conventional mice (Figure 8). There were large numbers of unique sequences, most of which were represented only once or twice. There were some CDR3 sequences that did occur relatively frequently among the Ki-67bright cells and were also frequent among the Ki-67negative cells. In each mouse, one sequence was represented frequently among the Ki-67bright cells but was not observed among the Ki-67negative cells. In mouse one, it constituted ∼26% of the Vβ2/Jβ1.1CDR3 sequences from the Ki-67bright cells; in mouse 2, it constituted 20% of the sequences. Thus, a considerable minority of the proliferation of the GF MP cells may have derived by burst-like expansion.
While the sample size of sequences from the GF donors was relatively small, they showed substantial diversity in both the Ki-67negative and Ki-67bright cells, suggesting that in the GF mice, the CD44bright cells have not arisen by differentiation and expansion of cells with a very limited TCR repertoire.
MP CD4 T cells from normal mice (i.e., mice housed in specific pathogen-free facilities) proliferate quite rapidly. BrdU labeling reveals that ∼10% of these cells from lymph nodes take up BrdU in a single day and more than 30% of MP cells are Ki-67+; on the basis of our estimate that the great majority of Ki-67+ cells have divided within the past 3 d, this implies that more than 30% of lymph node MP cells divide in 3 d, a result that is confirmed by more extended BrdU labeling. The frequency of Ki-67bright MP cells is somewhat less in the spleen. By contrast, NP cells take up BrdU at ∼0.1% per day and very few are Ki-67+.
What drives the rapid proliferation of the MP cells has been a matter of uncertainty. Some have concluded that their proliferation is driven by TCR engagement on the basis of transfer to lymphopenic hosts, where it is observed that by 7 d after transfer the majority of the surviving cells have divided seven times or more. Zamoyska and colleagues [25] and Leignadier et al. [27] have used a tetracycline/off system to delete TCR from mature T cells. In both instances, deleting TCR resulted in a substantial diminution in the proportion of CD44bright CD4 T cells that had gone through multiple divisions when transferred to lymphopenic recipients. Similarly, anti-MHC class II antibody blocked expansion of CD4 T cells introduced into neonatal recipients [23], and we showed here that the rapid proliferation of MP CD4 T cells introduced into Rag2−/− recipients was completely inhibited by the anti-MHC class II antibody Y3P. Furthermore, Surh and colleagues reported that the rapid proliferation of CD4 T cells that occurred when these cells were introduced into scid mice was largely lost if the scid recipients were GF [28]. As a group, these observations clearly indicate that in severely lymphopenic settings, expansion of MP cells depends on TCR recognition of peptide/MHC complexes.
The results we present here indicate that only a portion of the transferred MP cells undergo this striking proliferation. When we sequenced CDR3 segments from the Vβ2/Jβ1.1 TCRs 3 d after transfer of MP cells into Rag2−/− recipients, we found only three sequences among the rapidly dividing cells, whereas there were 37 sequences among those that had not divided, implying that the rapid proliferation was a property of a limited set of cells among the transferred MP population. This result is consistent with the observation that naïve CD4 T cells from most TCR transgenic donors fail to rapidly proliferate on transfer to lymphopenic recipients [1],[24],[29] and on our immunoscope analysis of TCR Vβ complexity in Rag2−/− recipients of numbers of CD4 T cells varying from 10 million to 10,000 [29] suggesting that only ∼3% of the transferred cells undergo rapid expansion. It is interesting that two of the sequences represented frequently in the dividing cells were also found in the nondividing population, implying that not all cells of the same specificity are stimulated in a lymphopenic environment.
However, the results obtained by the study of transfer to lymphopenic environments do not appear to be a valid representation of the mechanisms underlying the rapid proliferation of MP cells in situ. Indeed, survival of MP cells in lymphocyte-sufficient settings has been reported to not require expression of TCRs. Furthermore, there is a large literature demonstrating that survival of antigen-specific memory cells arising during immunization does not require TCR engagement, but rather depends upon the availability of cytokines, particularly IL-7 and IL-15 [10]–[12],[23],[24]. However, we wish to point out that the analysis of antigen-specific memory cells emerging from intentional immunization may not necessarily represent what governs the proliferative behavior of MP cells.
Indeed, both the work presented here and recent studies analyzing antigen-specific CD4 memory T cells at varying times after priming show that antigen-specific memory cells emerging from intentional immunization divide relatively slowly compared to MP cells. Lenz at al. [7] infected mice with LCMV. 50 d later, a 7-d exposure to BrdU resulted in only 15% of BrdU+ spleen cells among those capable of producing interferon gamma (IFNγ) in response to challenge with two different LCMV peptides. Purton et al. [5] transferred TCR transgenic SMARTA CD4 T cells, specific for an LCMV epitope, into C57BL/6 mice that were then infected with LCMV. 72 d after infection, 12% of the transgenic cells took up BrdU during a 5-d labeling period. Jenkins and colleagues [6] infected mice with Listeria monocyogenes expressing an ovalbumin peptide (LM2W1S). 40 d after infection, the mice received BrdU for 14 d; among spleen and lymph node CD4 T cells capable of binding an ovalbumin tetramer, only 11.5% were BrdU+.
Our results are consistent with these reports. We infected C57BL/6 mice with LCMV. Fifteen and 60 d later, the frequency of Ki-67+ cells among tetramer+ cells was measured. At 15 d, 8% of the CD44bright tetramer+ cells were Ki-67+; at 60 d, ∼7%. Collectively, these studies indicate that after the expansion phase following immunization is complete, antigen-specific CD44bright CD4 T cells divide at a rate of <1% to ∼2.5% per day. The possibility that MP cells and authentic memory cells might represent distinct cell types, or rather that the MP pool contains both authentic memory cells and another cell population, was also suggested by our prior study in SW GF mice that showed that their MP CD4 T cells proliferated at a rate similar to MP cells from conventional donors [3]. We have examined this point in greater detail here in GF and conventional C57BL/6 mice and confirm that the proportion and absolute number of non-Treg CD44bright CD4 T cells from peripheral and mesenteric lymph nodes of GF mice are similar to those from conventional mice as are the proportion and number of proliferating MP cells. It should also be pointed out that prior studies of GF mice maintained on elemental diets (i.e., antigen-free mice) had shown the presence of substantial numbers of activated CD4 T cells, equivalent in frequency to those in conventional mice [30]–[32]. While these studies were carried out before the availability of the reagents now used to classify MP cells, they strongly suggest that antigen-free mice have similar numbers of MP CD4 T cells as do conventional mice and thus support the concept that foreign (including commensal) antigens are not critical to the emergence of the majority of MP cells.
Here we have shown that the in situ proliferation of MP cells is not inhibited by anti-MHC class II antibody, using a reagent that strikingly inhibits the proliferation of antigen-specific cells in response to antigen challenge and that blocks the rapid proliferation of MP cells transferred to lymphopenic recipients. Rather, we observe that anti–IL-7Rα antibody diminishes, but does not abolish, proliferation of MP cells, implying that IL-7 or TSLP plays a role in this proliferation.
An alternative way to examine the issue of whether the rapid proliferation of these cells represents an antigen-driven response, during which one would anticipate that limited numbers of precursors give rise to bursts consisting of multiple divisions, is to examine the TCR sequence diversity of proliferating MP cells and to compare that to the sequence diversity of quiescent MP cells. If MP proliferation was primarily due to burst-like clonal expansion stimulated by exposure to antigen, it would be expected that the sequence diversity of the proliferating cells would be substantially less than that of the quiescent cells. Indeed, when we studied proliferating and nonproliferating MP cells in lymphopenic recipients, this is precisely what we observed.
We examined CDR3 sequences from Vβ2/Jβ1.1 and Vβ4/Jβ1.1 MP CD4 T cells that were Ki-67bright or Ki-67negative. We chose to limit our study to these two TCR Vβ sets so that we could sample a larger proportion of these defined subrepertoires than we could with the same number of sequences of total CD44bright Ki-67bright cells. As an estimate of the number of cells under study, we used the following considerations. The total number of lymph node CD4 T cells is ∼8 million. ∼15% of these cells are CD44bright, or ∼1.2 million. Of these, ∼half are CD25+, so that MP cells constitute ∼600,000; ∼5% express Vβ2 or Vβ 4, or ∼30,000. Of the Vβ2 or Vβ 4 expressing cells, ∼10% are Jβ1.1, or ∼3,000. The Ki67bright cells are ∼10% of the MP cells so that the total number of Ki-67bright Vβ2/Jβ1.1 or Vβ4/Jβ1.1 MP cells in all the lymph nodes of the animal is ∼300. Thus, the maximum number of unique sequences would be 300 in this cell population; this would be substantially less if repetitive sequences existing among these cells, which would be anticipated on the basis of the likelihood that the generation of MP cells from naïve precursors involved clonal expansion. Thus, in our initial analysis, involving >200 sequences from the CD44bright Ki-67bright Vβ2/Jβ1.1 cells of mouse 1, our sample, while not complete, is quite substantial. Even the samples of 70 to 80 sequences in the other cases are sufficient to provide useful information about complexity, as judged by our observations of multiply repeated sequences.
A further point is our reliance on CDR3 sequences from the β chain of the TCR as a clonal marker. It is possible that we have overestimated the frequency of repeats since there may be occasions in which the same Vβ is used with different Vα's, but we suspect that in the vast majority of cases the CDR3 sequence of the TCR β chain is indeed a clonal marker.
Our results indicate that the distribution of sequences in the Ki-67bright and Ki-67negative populations was not markedly different. If we made the assumption that any sequences that occurred many times among the Ki-67bright cells represented cells that had recently been in a burst-like expansion, then ∼25% of the Ki-67bright cells would be judged to be in such bursts. To obtain this estimate we arbitrarily assigned any sequence that occurred four or more times among the Ki-67bright cells to the set that occurred “many” times. However, this could easily be an overestimate depending on how one interprets those instances in which a similarly high frequency of the same sequence was found among Ki-67negative cells. On the one hand, this might reflect a large clone in which cell division occurred on a stochastic basis so that the frequency of the clone was similar among the dividing and nondividing cells. If we make this assumption, then the proportion of Ki-67bright cells that were in bursts becomes ∼7%. Alternatively, instances in which a sequence found in the Ki-67bright cells was equally (or over-represented) among the Ki-67negative cells may represent the late-phase of a burst episode in which a portion of the cells had already stopped dividing and lost Ki-67 expression but others continued to divide. Overall, we conclude that the sequence data are consistent with a minority of the Ki-67bright cells being part of burst; whether that minority is small or considerable cannot easily be determined. However, when taken together with the failure of anti-class II antibody to block proliferation of MP cells, it is reasonable to conclude that the proportion of Ki-67bright cells that are part of burst-like expansion is quite small.
There may be circumstances in which clonal expansion/burst-like antigen-driven proliferation plays a much great role than is found among MP cells from normal mice. It has been argued that the dynamics of MP cells from SIV-infected macaques, in which proliferative rates are far higher than proliferative rates of MP cells from noninfected macaques, is best explained by multiple overlapping burst episodes of several cell divisions occurring within a brief period of time [17],[18]. These cells, which exist in a highly inflammatory setting, may well show enhanced sensitivity to their cognate antigens or to self-peptide/MHC complexes. A detailed analysis of their sequence complexity and of the complexity of MP cells derived from chronically infected individuals or individuals that were in a state of chronic inflammation would help to clarify this point.
As discussed above, we found that GF and conventional C57BL/6 mice have comparable numbers of MP CD4 T cells in their peripheral and mesenteric lymph nodes and these cells display similar proliferative rates. We sequenced CDR3 gene segments from Vβ2/Jβ1.1 Ki67bright and Ki-67negative MP cells of two GF mice and found that they, like the comparable cells from conventional mice, were very similar in their sequence diversity. One could argue that GF mice are not antigen free, although there are reports that antigen-free mice show normal numbers of “activated” CD4 T cells. Nonetheless, there can be little doubt that GF mice have a much reduced antigenic experience. If the MP cells of GF mice represent clonal expansions because of an extremely limited set of naïve cells responding to a comparably limited set of antigenic stimuli, it would be anticipated that the MP cells from GF mice would have a much more restricted repertoire than MP cells from conventional mice. While our limited number of sequences may not be sufficient to reach a definitive conclusion on this point, there does not appear to be any major difference in the degree of diversity of the Ki-67bright or Ki-67negative MP cells from the two sets of donors.
On the basis of these observations, we propose that MP cells may be a more diverse population than had been considered and that only some of these cells may have emerged by exposure to conventional foreign antigens. The maintenance of most MP cells and their rapid proliferative rate appear to be largely dependent on cytokines. The origin of the infrequently occurring proliferation bursts remains to be clarified, but an obvious possibility is through recognition and response to self-peptides on competent antigen-presenting cells. It should be pointed out that diversity in CD8 T cells has also been described, with one population being designated “bystanders” and that such cells take on a memory phenotype in mice deficient in the transcription factor KLF2, the signaling kinase itk, or the histone acetyltransferase CBP [33]. Whether such “bystander” CD8 cells bear a relationship to the rapidly dividing CD4 MP cells discussed here remains to be determined.
Overall, one may ask what is the function of the large set of MP cells in normal mice? We have proposed [23],[34]–[37] that they represent a pool of cells capable of making a rapid effector response to cross-reactive antigens of pathogens during a period in which the naïve cells proliferate and differentiate. MP cells might play an even more important role in instances in which naïve cells are limiting and no “authentic” memory cells are specific for an introduced pathogen, such as might be the case in aged individuals. Devising models in which these cells are absent will be essential to testing their function. Finally, why proliferative rates of authentic memory and MP cells are different is not clear.
C57BL/6 (B6), B6 Rag2−/−, B6 FcγRγchain−/−, and OT-II CD45.1 mice were obtained from the National Institute of Allergy and Infectious Diseases (NIAID) contract facility at Taconic Farms. GF mice were maintained at the NIAID GF facility. All other mice were maintained under pathogen-free conditions in NIAID animal facilities. Mice infected with LCMV were inoculated IP with 2×105 PFU Armstrong strain. The care and handling of the animals used in our studies was in accordance with the guidelines of the National Institutes of Health (NIH) Animal Care and Use Committee.
Y3P was obtained from Harlan Bioproducts. Antibodies to IL-15 (5H4), IL-2 (S4B6), CD127 (IL-7Rα; SB/14), CD4 (pacific blue; RM4-5), Ki-67 (PE; B56), Vβ2 (FITC; B20.6), Vβ4 (FITC; KT4) were purchased from BD Biosciences. Anti-CD44 (Alexa-700; IM7) and anti-Foxp3 (PE; NRRF-30) were purchased from eBioscience. The detection of BrdU was carried out according to instructions in the kit provided by BD Biosciences. The I-Ab-GP-66-77 tetramer that recognizes receptors for an immunodominant LCMV epitope was provided by the NIH tetramer facility (Emory Vaccine Center). All flow cytometry analyses were performed using an LSR-II (BD Biosciences).
Inguinal, axillary, cervical, and mesenteric CD4 MP lymph node T cells were obtained by sorting on a FACSAria (BD Biosciences). Purity was >99% CD4, CD44+ or, depending on the experiment, CD4, CD44+, Foxp3−. In some experiments, sorted cells were labeled with CFSE (Molecular Probes) at a final concentration of 1.25 µmol and transferred IP into recipient mice.
From 105 to 0.5×105 KI-67 negative and bright CD4 CD44bright, Foxp3− T cells were FACS-sorted into FCS. Cells were resuspended in lysis buffer (20 mmol Tris-HCl, pH 7.5, 150 mmol NaCl) with 4 µg/ml proteinase K (Fermentas), incubated at 56°C for 50 min and then at 95°C for 10 min. Volumes were adjusted to 30 µl; 10 µl were used to amplify the Vβ2/Jβ1.1 and Vβ4/Jβ1.1 CDR3s with the following primers: Vβ2 5′ CAGTCGCTTCCAACCTCAAAGTTC′ or Vβ4 5′ CGATAAAGCTCATTTGAATCTTCGAATC and 3′ Jβ1.1 AGCTTTACAACTGTGAGTCTGGTTCCTTTACC using 35 PCR cycles of 45 s at 95°C, 45 s at 57°C, and 45 s at 72°C. The PCR products were cloned into the TOPO blunt end vector (Invitrogen) and bacteria were transformed. Single colonies were isolated and suspended into 10 µl of water. PCR was carried out on 3 µl of bacterial suspension from single colonies using the universal M13 primers. PCR products were sequenced by Agencourt (Beckman Coulter) using universal T3 primer. The rate of readable sequences was 70% to 80%.
Lymph node cells were stained with anti-CD4, anti-Vβ2, or anti-Vβ4 followed by single-cell sorting into 96-well plates. Plates were heated to 95°C for 3 min followed by a first round PCR of 40 cycles (45 s 95°C, 45 s at 57°C, 45 s at 72°C) using the 5′ Vβ2 or Vβ4 primers and a 3′ Jβ2.7 AGGCTCACGGTTTTAG primer. 3 µl of the first PCR product were subjected to a second PCR (25 cycle of 45 s at 95°C, 45 s at 57°C, 45 s at 72°C) using the Vβ2 or Vβ4 primers and the 3′ Jβ1.1 primer. PCR products were obtained in eight and seven wells out of 96 for Vβ2 or Vβ4, respectively, indicating a frequency for each of ∼10%.
Analysis of the CDR3 sequences was performed using Matlab as a platform to generate a program to analyze the sequences.
|
10.1371/journal.pmed.1002029 | Financial Relationships between Organizations That Produce Clinical Practice Guidelines and the Biomedical Industry: A Cross-Sectional Study | Financial relationships between organizations that produce clinical practice guidelines and biomedical companies are vulnerable to conflicts of interest. We sought to determine whether organizations that produce clinical practice guidelines have financial relationships with biomedical companies and whether there are associations between organizations’ conflict of interest policies and recommendations and disclosures provided in guidelines.
We conducted a cross-sectional survey and review of websites of 95 national/international medical organizations that produced 290 clinical practice guidelines published on the National Guideline Clearinghouse website from January 1 to December 31, 2012. Survey responses were available for 68% (65/95) of organizations (167/290 guidelines, 58%), and websites were reviewed for 100% (95/95) of organizations (290/290 guidelines, 100%). In all, 63% (60/95) of organizations producing clinical practice guidelines reported receiving funds from a biomedical company; 80% (76/95) of organizations reported having a policy for managing conflicts of interest. Disclosure statements (disclosing presence or absence of financial relationships with biomedical companies) were available in 65% (188/290) of clinical practice guidelines for direct funding sources to produce the guideline, 51% (147/290) for financial relationships of the guideline committee members, and 1% (4/290) for financial relationships of the organizations producing the guidelines. Among all guidelines, 6% (18/290) disclosed direct funding by biomedical companies, 40% (117/290) disclosed financial relationships between committee members and biomedical companies (38% of guideline committee members, 773/2,043), and 1% (4/290) disclosed financial relationships between the organizations producing the guidelines and biomedical companies. In the survey responses, 60 organizations reported the procedures that they included in their conflict of interest policies (158 guidelines): guidelines produced by organizations reporting more comprehensive conflict of interest policies (per additional procedure, range 5–17) included fewer positive (rate ratio [RR] 0.91, 95% CI 0.86–0.95) and more negative (RR 1.32, 95% CI 1.09–1.60) recommendations regarding patented biomedical products. The clinical practice guidelines produced by organizations reporting more comprehensive conflict of interest policies were also more likely to include disclosure statements for direct funding sources (odds ratio [OR] 1.31, 95% CI 1.10–1.56) and financial relationships of guideline committee members (OR 1.36, 95% CI 1.09–1.79), but not financial relationships of the organizations (0 disclosures). Limitations of the study include the use of the National Guideline Clearinghouse as the single source of clinical practice guidelines and the self-report of survey responses and organizations’ website postings.
Financial relationships between organizations that produce clinical practice guidelines and biomedical companies are common and infrequently disclosed in guidelines. Our study highlights the need for an effective policy to manage organizational conflicts of interest and disclosure of financial relationships.
| Clinical practice guidelines are designed to influence the practice of large numbers of providers and are vulnerable to influence from biomedical companies.
To manage conflicts of interest, the disclosure of financial relationships with biomedical companies is recommended for individuals and organizations involved in producing clinical practice guidelines.
It is unknown whether organizations that produce clinical practice guidelines have financial relationships with biomedical companies, whether they disclose these relationships, and what policies they use to manage conflicts of interest.
The majority of organizations (63%) that published clinical practice guidelines on the National Guideline Clearinghouse website in 2012 reported receiving funds from biomedical companies on their website or in response to a survey.
Very few (1%) of the published clinical practice guidelines disclosed financial relationships between the organizations producing the clinical practice guidelines and biomedical companies.
Clinical practice guidelines produced by organizations with more comprehensive conflict of interest policies included relatively fewer positive (−9%) and more negative (+32%) recommendations regarding patented biomedical products and were more likely to include disclosure statements for direct funding sources (+31%) and the financial relationships of the guideline committee members (+36%).
Effective policies for managing conflicts of interest for organizations that produce clinical practice guidelines are needed.
We support complete disclosure of financial relationships with biomedical companies for organizations that produce clinical practice guidelines.
| Financial relationships between biomedical companies and the individuals and organizations involved in the production of clinical practice guidelines are vulnerable to conflicts of interest [1–5]. These types of relationships can have undue influence because clinical practice guidelines are resource intensive to produce [6] and are developed by a small number of expert clinicians who determine the scope of the guidelines, synthesize and interpret the published evidence base, and provide recommendations. The potential impacts of conflicts of interest are large because clinical practice guidelines are designed to be widely disseminated and influence the practice patterns of large numbers of healthcare providers [1,7,8].
To help manage conflicts of interest, the Institute of Medicine Committee on Conflict of Interest in Medical Research, Education, and Practice recommended that all sources of funding, both direct and indirect, be “publicly disclose[d] with each guideline” [3]. Organizations that produce clinical practice guidelines and journals that publish guidelines have responded by requiring disclosure statements from experts who participate in guideline development [9]. However, it is unclear whether organizations that produce clinical practice guidelines have financial relationships with the biomedical industry, whether they have policies to manage conflicts of interest, and whether financial relationships are disclosed within guidelines.
We therefore conducted an observational study to determine whether organizations that produce clinical practice guidelines receive funds from biomedical companies, what policies (and specific procedures) they use to minimize and/or manage conflicts of interest, what disclosures they provide within guidelines, and whether there are associations between organizations’ conflict of interest policies and the recommendations and disclosures provided in guidelines.
We employed the conceptualization of conflict of interest developed by Emanuel and Thompson [10] and defined a conflict of interest as “a set of conditions in which professional judgment concerning a primary interest (such as a patient’s welfare or the validity of research) tends to be unduly influenced by a secondary interest (such as financial gain)” [11]. We used the National Guideline Clearinghouse definition of a clinical practice guideline: “clinical practice guidelines are systematically developed statements to assist practitioner and patient decisions about appropriate health care for specific clinical circumstances” [12,13]. The study was conducted by selecting clinical practice guidelines for evaluation, abstracting recommendations and disclosure statements from the guidelines and collecting information on funding sources and conflict of interest policies for the organizations producing the guidelines from their websites and by survey. The study received ethics approval from the University of Calgary Conjoint Health Research Ethics Board (REB13-0048, see S1 Text for research ethics board application).
Two authors (P. C. and H. T. S.) independently reviewed all clinical practice guidelines posted to the US Agency for Healthcare Research and Quality’s National Guideline Clearinghouse website between January 1 and December 31, 2012 (accessed June 9, 2013) [14]. We included clinical practice guidelines produced by organizations whose membership and scope were national or international. We excluded clinical practice guidelines produced by organizations whose membership was restricted to allied health professionals (e.g., nursing associations) or corporations (e.g., health maintenance organizations, insurance companies, etc.); we also excluded guidelines that provided no specific recommendations (suggestions on the best course of clinical action, e.g., “We recommend…”) [15]. These criteria were used to identify clinical practice guidelines that would include recommendations that targeted clinical practices, were related to biomedical products, and could be widely disseminated. Disagreements about which guidelines met these exclusion criteria were resolved by discussion.
Two authors (P. C. and K. C.) used a standardized and pilot-tested data collection tool to abstract the data (with each author reviewing half of the clinical practice guidelines and associated websites). To evaluate the reliability of the data abstraction process, both authors independently abstracted a random sample of 10% of the guidelines and associated websites.
We abstracted data from clinical practice guidelines and the National Guideline Clearinghouse website on reported sources of funding, conflict of interest policies (or a link to a policy), and disclosures of conflicts of interest. All recommendations (suggestions on the best course of clinical action, e.g., “We suggest…”) [15] provided within each clinical practice guideline were individually classified according to whether or not they were related to a biomedical product (i.e., pharmaceutical product or medical technology product). Those recommendations classified as being related to a biomedical product were further classified as positive (recommending the product), neutral (neither recommending nor advising against use of the product), or negative (advising against use of the product) as part of a clinical action management approach using a classification scheme derived from the American College of Cardiology/American Heart Association Task Force on Practice Guidelines and GRADE [16,17]. Biomedical products were classified as patented/exclusive if the patent/exclusive expiry date for the agent was listed as 2013 or later per Approved Drug Products with Therapeutic Equivalence Evaluations, 33rd edition, published by the US Department of Health and Human Services [18].
We abstracted data from the websites of the organizations producing the clinical practice guidelines on disclosure of funding from biomedical companies, solicitation of funding from biomedical companies (defined as invitation and/or instruction on how a corporation could provide funding support to the organization—website tabs/links labelled “opportunities for corporate sponsorship,” “opportunities for sponsorship at upcoming scientific meetings,” etc.), and conflict of interest policies to manage relationships between the organization and biomedical companies.
The survey instrument (see S2 Text) was designed to obtain information about organizational characteristics (type of organization, membership), funding sources (annual revenue, funding sources), and conflict of interest policies (existence of a policy and specific procedures for managing conflicts of interest) for guideline production. A list of 18 procedures for managing conflicts of interest was derived from those recommended by the Institute of Medicine [3], the Council of Medical Specialty Societies [19], and the American College of Chest Physicians [20]; the survey was used to inquire which of these procedures were included in the organization’s conflict of interest policy. For organizations with a conflict of interest policy, a copy was requested. An assessment of the survey’s face validity, clarity, length, and completeness was performed using semi-structured interviews (pretesting) with physicians with experience in clinical practice guideline development prior to its distribution [21].
We searched the websites of the organizations that produced the clinical practice guidelines to identify organizational contacts. Representatives were contacted via email and telephone to identify the most appropriate person within the organization to complete the survey. The individual designated by the organization was sent an email cover letter explaining the purpose of the study and a link providing access to a secure, web-based survey (see S2 Text). Participation in the survey was voluntary; consent was inferred from survey completion. Reminders (emails at 4, 8, and 12 wk and a telephone call at 12 wk) were sent to those who did not respond [22].
Descriptive statistics (proportions and medians and interquartile ranges [IQRs]) were used to report the data abstracted from the clinical practice guidelines, organizational websites, and surveys. Data were reported for all responses to survey questions, and missing values were not imputed. Agreement on the selection of clinical practice guidelines for inclusion in the study and data abstraction from clinical practice guidelines and websites was assessed with Cohen’s kappa coefficient and Cohen’s weighted kappa coefficient [23]. We tested for associations between the reported number of procedures used by organizations for managing conflicts of interest and recommendations (number and nature) and disclosure statements provided in guidelines using Poisson and logistic regression models, respectively. To account for the interdependence of observations (organizations producing more than one guideline), we used robust estimates of variance (generalized estimating equations) [24]. Analyses were calculated using Stata (version 13.1, StataCorp).
Fig 1 summarizes the selection of clinical practice guidelines and producing organizations. Our search of the National Guideline Clearinghouse identified 426 clinical practice guidelines posted between January 1 and December 31, 2012. We excluded a total of 136 clinical practice guidelines because they were produced by organizations whose membership and scope were primarily at the local or state level (n = 91) or whose membership was restricted to allied health professionals (n = 21); the guidelines were produced by health maintenance/insurance corporations (n = 19); or the guidelines did not provide specific recommendations (n = 5). We identified 290 clinical practice guidelines (see S3 Text) produced by 95 national/international medical organizations (see S4 Text) for inclusion in the study. Websites were identified for all 95 (100%) organizations (see S4 Text). The survey was sent to a representative of each organization between December 5, 2013, and April 21, 2014, of which 24 did not respond, six declined to participate, and 65 (65/95, 68%) responded/completed the survey (see S1 Data). Agreement between reviewers on inclusion of clinical practice guidelines in the study (kappa 0.973) and abstraction of data from guidelines and websites (kappa 0.802 for binary data, weighted kappa 0.997 for ordinal data) was excellent.
The characteristics of the organizations that produced the clinical practice guidelines are summarized in Table 1; these characteristics were broadly similar across the organizations that did and did not respond to the survey. The organizations producing the clinical practice guidelines were primarily professional associations (67%) or disease/condition interest groups (21%). The self-reported yearly revenues of the organizations ranged from less than $1 million to over $50 million.
The characteristics of the clinical practice guidelines are summarized in Table 3. The majority of clinical practice guidelines were produced in the United States (65%) and were focused on internal medicine and its subspecialties (42%). The clinical practice guidelines included a total of 4,057 guideline committee members, with a median of 13 members per guideline (IQR 8–17). The median number of recommendations per clinical practice guideline was 9.5 (IQR 4–24), and these recommendations included recommendations regarding pharmaceutical products and medical technologies. Sixty-two percent of clinical practice guidelines had been published in peer-reviewed journals at the time of data collection.
The majority (55/60, 92%) of organizations that reported procedures for managing conflicts of interest indicated that they had a policy specifically for managing conflicts of interest during guideline development. A minority (50/290, 17%) of clinical practice guidelines made reference to a conflict of interest policy of the producing organization. Three of the procedures reported by organizations for managing conflicts of interest (Table 2) could be compared to the disclosures provided in the guidelines produced by the organizations during the study period (Table 4). First, among organizations that reported that committee member conflicts of interest were published in guidelines, nine (9/55, 16%) produced a guideline that did not include a committee member disclosure statement. Second, among organizations that reported that the majority of committee members must be free of conflicts of interest, over half (25/41, 61%) produced a guideline that disclosed financial relationships with biomedical companies for a majority of the committee members. Third, among the organizations that reported that industry partners were not permitted to directly fund clinical practice guideline development, three (3/47, 6%) produced a guideline that disclosed direct funding/support from a biomedical company.
Table 5 summarizes recommendations and disclosures provided in clinical practice guidelines (n = 158) according to the number of procedures for managing conflicts of interest reported by the producing organization (n = 60). There was an association between the number of conflict of interest procedures reported by an organization (18 potential items, response range 5–17) and the number of guidelines produced (rate ratio [RR] 1.10, 95% CI 1.03–1.17, p = 0.003). Organizations with more comprehensive policies (per additional procedure) produced guidelines that included more recommendations regarding biomedical products (RR 1.05, 95% CI 1.03–1.07) but fewer recommendations regarding patented biomedical products (RR 0.94, 95% CI 0.90–0.98). Clinical practice guidelines produced by organizations reporting more comprehensive conflict of interest policies included fewer positive (RR 0.91, 95% CI 0.86–0.95) and more negative (RR 1.32, 95% CI 1.09–1.60) recommendations regarding patented biomedical products. These clinical practice guidelines were more likely to include disclosure statements for direct funding sources (odds ratio [OR] 1.31, 95% CI 1.10–1.56) and for financial relationships of guideline committee members (OR 1.36, 95% CI 1.09–1.79) but not for financial relationships of the organization (zero disclosures).
Our study described financial relationships between organizations that produce clinical practice guidelines and the biomedical industry, and the policies employed to manage conflicts of interest. The results demonstrated that the majority of organizations reported financial relationships with biomedical companies. Most organizations had policies and procedures to manage conflicts of interest; however, there was variation in the procedures included within the policies, and a minority of policies specifically considered production of clinical practice guidelines. Two-thirds of clinical practice guidelines provided disclosure statements for direct funding sources for the guideline, and half provided disclosure statements for guideline committee members. Only 1% of clinical practice guidelines provided disclosure statements for the organizations producing the guidelines.
A growing body of research over the past two decades has described and explored the implications of financial relationships between the medical community and biomedical companies [5,25–28]. For example, a systematic review by Licurse et al. reported that patients, research participants, and readers of medical journals perceive financial relationships between physicians and biomedical companies as impacting the quality and cost of healthcare and believe that these relationships should be disclosed [29]. Journals [30,31], professional societies [19,20,32], government agencies [33], and the biomedical industry [34] have implemented strategies for managing conflicts of interest. A systematic review by Norris et al. in 2011 reported that financial relationships between clinical practice guideline committee members and biomedical companies are common [4] but may be increasingly disclosed [2]. Our study adds to this body of literature by suggesting three key issues that should be considered to further improve the prevention and management of conflicts of interest in the financial relationships between organizations that produce clinical practice guidelines and the biomedical industry.
First, organizations that produce clinical practice guidelines should develop conflict of interest policies to manage relationships with biomedical companies, should ensure these policies address the production of guidelines, and should make the policies available to guideline users. In our study, approximately one in five organizations did not have a conflict of interest policy, and less than half of policies specifically addressed the production of guidelines. Furthermore, conflict of interest policies were infrequently referenced in the guideline text, leaving readers with uncertainty about how conflicts were prevented or managed. The content of the conflict of interest policy is essential for critical appraisal [35], and policies should be made publicly available. In 2009, the Institute of Medicine recommended that journals and websites that publish clinical practice guidelines require organizations to “describe (or provide an Internet link to) the developer’s conflict of interest policy” [3]. Although it is unclear what an optimal conflict of interest policy should include, the data abstraction elements used in the present study—which were derived from the recommendations provided by the Institute of Medicine [3], the Council of Medical Specialty Societies [19], and the American College of Chest Physicians [20]—could serve as a starting template.
Second, disclosure of financial relationships is necessary for transparency and managing conflicts of interest [36,37]. A clinical practice guideline can be funded directly, whereby an entity provides funds for production of a particular guideline, or indirectly, whereby an entity provides funds to an organization that produces clinical practice guidelines, and these funds are then applied to the organization’s programs including guideline production. Both forms of funding represent conflicts of interest. The Institute of Medicine recommends that organizations disclose “sources and amounts of indirect or direct funding received for development of the guideline” [3]. Our study suggests that while disclosure statements regarding direct funding and support for committee members are increasingly provided in clinical practice guidelines [2,4,38,39], disclosure of financial relationships between the organizations that produce guidelines and biomedical companies is uncommon.
Third, mechanisms are needed to manage breakdowns in policies to manage conflicts of interest. Financial relationships between organizations that produce clinical practice guidelines and the biomedical industry are common and likely complex. We anticipate that even when organizations have policies to manage conflicts of interest, breakdowns in execution of policies are inevitable. For example, among the organizations that responded to the survey, 68% indicated that their policies stipulated that the majority of committee members must be free of conflicts of interest, yet, of these, 61% published a guideline included in our study for which the majority of committee members disclosed financial relationships with biomedical companies. Similarly, although 92% of organizations responded in the survey that they published committee member disclosure statements within their guidelines, 18% of these organizations published at least one guideline included in our study that provided no committee member disclosure statement. These discrepancies reflect the complexity of managing conflicts of interest related to producing a clinical practice guideline. A solution advocated by the Institute of Medicine is that organizations create a standing committee to oversee organizational conflicts of interest as well as procedures to manage violations of the conflict of interest policy [3]. A minority of organizations in our study reported using either of these approaches.
The results of our study need to be interpreted within the context of the study’s limitations. First, we sampled a single source of clinical practice guidelines, the National Guideline Clearinghouse, although this is the largest repository of clinical practice guidelines. Furthermore, only English language guidelines are included on the clearinghouse website, and we restricted our focus to guidelines produced by national/international medical organizations, which may limit the transferability of our results. Second, data abstraction was performed by one reviewer, which could introduce error or bias. To guard against this risk, data abstraction was sequentially performed for clinical practice guidelines and websites prior to survey administration, and independent blinded abstraction of a random sample of 10% of the clinical practice guidelines and websites by a second reviewer demonstrated excellent reliability. Third, we developed a parsimonious survey instrument to encourage participation, but this limited data collection. For example, our study provides no data on how organizations managed conflicts of interest for their staff members. Finally, we depended on organizations’ self-report through survey responses and website postings. This limits our ability to fully describe relationships between organizations producing clinical practice guidelines and biomedical companies, and suggests that our analyses likely underestimate the number of these relationships.
Financial relationships between organizations that produce clinical practice guidelines and the biomedical industry appear to be common. These relationships are important because they may influence, through guideline usage, the practice of large numbers of healthcare providers. We believe that to effectively manage conflicts of interest, organizations that produce clinical practice guidelines need to develop robust conflict of interest policies that include procedures for managing violations of the policy, make the policies publicly available, and disclose all financial relationships with biomedical companies.
|
10.1371/journal.pgen.1008415 | Cytoneme-mediated signaling essential for tumorigenesis | Communication between neoplastic cells and cells of their microenvironment is critical to cancer progression. To investigate the role of cytoneme-mediated signaling as a mechanism for distributing growth factor signaling proteins between tumor and tumor-associated cells, we analyzed EGFR and RET Drosophila tumor models and tested several genetic loss-of-function conditions that impair cytoneme-mediated signaling. Neuroglian, capricious, Irk2, SCAR, and diaphanous are genes that cytonemes require during normal development. Neuroglian and Capricious are cell adhesion proteins, Irk2 is a potassium channel, and SCAR and Diaphanous are actin-binding proteins, and the only process to which they are known to contribute jointly is cytoneme-mediated signaling. We observed that diminished function of any one of these genes suppressed tumor growth and increased organism survival. We also noted that EGFR-expressing tumor discs have abnormally extensive tracheation (respiratory tubes) and ectopically express Branchless (Bnl, a FGF) and FGFR. Bnl is a known inducer of tracheation that signals by a cytoneme-mediated process in other contexts, and we determined that exogenous over-expression of dominant negative FGFR suppressed and tumor growth. Our results are consistent with the idea that cytonemes move signaling proteins between tumor and stromal cells and that cytoneme-mediated signaling is required for tumor growth and malignancy.
| The growth of many types of tumors depend on productive interactions with stromal, non-tumor neighbors, and although there is evidence that tumor and stromal cells exchange signaling proteins and growth factors that they produce, the mechanism by which these proteins move between the signaling cells has not been investigated and is not known. Our previous work has shown that normal cells make transient chemical synapses at sites where specialized filopodia called cytonemes contact signaling partners, and in this work we explore the possibility that tumors use the same mechanism to communicate with stromal cells. We show that cytoneme-mediated signaling is essential for growth of Drosophila tumors that model human EGFR over-expression and RET-driven disease. Remarkably, inhibition of cytonemes cures flies of lethal tumors.
| Human tumors include transformed tumor cells, blood vessels, immune response cells, and stromal cells that together with the extracellular matrix (ECM) constitute a “tumor microenvironment” [1]. The tumor microenvironment is essential for oncogenesis, cell survival, tumor progression, invasion and metastasis [2,3], and its stromal cells produce key drivers of tumorigenesis. Known drivers are growth factors (e.g. HGF, FGF, EGF, IGF-1, TGF-β and Wnts), cytokines (e.g. IL-6, SDF-1) and pro-angiogenic factors (e.g. VEGF). It is not known if these proteins function as autocrine, juxtacrine, or paracrine signals, nor is it known how they might move into or within the tumor microenvironment.
Studies of tumor models in Drosophila exploit the experimental attributes of the fly that provide uniquely powerful ways to investigate tumorigenesis [4]. We tested two models for the roles of cytonemes. Cytonemes are specialized, actin-based filopodia that extend between cells that produce and secrete signaling proteins and cells that receive them. The signaling proteins move along cytonemes and exchange at transient synapses that form where cytonemes contact target cells. These synapses are similar to neuronal synapses in constitution, structure and function [5–7], and are necessary for paracrine FGF/Bnl, BMP/Dpp, Hedgehog, Wnt/Wingless (Wnt/Wg), and Notch signaling during normal development of Drosophila epithelial tissues [5,7–9].
EGFR activating mutations are drivers of several types of human cancers [10]. However, elevated EGFR expression of wild type EGFR is not sufficient for tumorigenesis, and additional genetic changes are necessary, such as over-expression of Perlecan, a heparan sulfate proteoglycan (HSPG) component of the ECM [11]. In Drosophila, ectopic over-expression of Perlecan and EGFR in epithelial cells of the wing imaginal disc drives tumorigenesis [12]. Growth and metastasis of the epithelial cells require crosstalk with closely associated mesenchymal myoblasts, which also proliferate abnormally when Perlecan and EGFR are over-expressed in epithelial neighbors. The crosstalk includes BMP/Dpp signaling from the epithelial cells to the mesenchymal myoblasts [12].
The RET gene is the primary oncogenic driver for MEN2 (multiple endocrine neoplasia type 2) syndrome. MEN2 is characterized by several types of neoplastic transformations, including an aggressive thyroid cancer called medullary thyroid carcinoma (MTC). A fly model that overexpresses RETMEN2 phenocopies aspects of the aberrant signaling in MEN2-related tumors, such as activation of the SRC signal transduction pathway, which promotes migration and metastasis of tumorigenic cells. The relevance of the fly model has been established by screens for small molecule suppressors of Drosophila tumors driven by RETMEN2 over-expression. Several compounds that were identified are more effective than the drugs that are currently used for patients [13,14].
In the work presented here, we examined the role of cytoneme-mediated signaling in the EGFR-Pcn and RETMEN2 models. Genetic inhibition of cytonemes by downregulation of five genes that were shown previously to be essential in cytoneme-mediated signaling, reduced tumor growth, and we describe genetic conditions that suppress lethality by as much as 60% in the EGFR-Pcn tumor and by as much as 30% in the RETMEN2 tumor. Our results are consistent with the possibility that cytoneme-mediated signaling is necessary for tumor growth and that interfering with cytoneme-mediated tumor-stromal cell signaling might be a therapy for tumor suppression.
Most of the wing imaginal disc is a columnar epithelium that will generate the wing and cuticle of the dorsal thorax of the adult fly. The disc also includes myoblasts that grow and spread over much of the dorsal basal surface of the columnar epithelium; these mesenchymal cells will generate the flight muscles of the adult. Tracheal branches (respiratory tubes) also adjoin the basal surface of the columnar epithelium, and one branch, the transverse connective, sprouts a bud (the air sac primordium (ASP), Fig 1A) that initiates growth during the third instar period. The ASP is dependent on Dpp and Bnl signaling proteins produced by the wing disc [7]. The myoblasts relay Wg and Notch signaling between the disc and ASP [9]. Cytonemes mediate and are essential for the Dpp, Bnl, Wg, and Notch signaling [15].
To investigate whether cytonemes are also essential in tumorigenesis, we tested a cancer model that requires tumor-stroma interactions in which neoplastic transformation is driven by interactions between the wing disc epithelial cells and myoblasts [12]. Overexpression of wild type EGFR and Perlecan (pcn, a secreted heparan sulfate proteoglycan) in the columnar epithelium drives proliferation of the genetically modified epithelial cells, as well as their genetically wild type myoblast neighbors. Tumorigenesis depends on Dpp signaling from the epithelial cells to the myoblasts.
We first investigated if cytonemes are present in EGFR-Pcn overexpressing tumor cells. We induced the EGFR-Pcn tumor model (with ap-Gal4, an epithelial cell-specific driver) together with CD8:GFP, a membrane-tethered GFP protein (Fig 1C and 1D), and independently expressed membrane-tethered mCherry in the myoblasts (with 1151-lexA, lexO-mCherry-CAAX; a myoblast-specific driver). In this system, the epithelial cell membranes are marked with GFP fluorescence and the myoblast membranes are marked with mCherry fluorescence. We observed that, as previously reported [12], the EGFR-Pcn tumor induces overgrowth and proliferation, producing multilayered masses of disorganized disc epithelial cells and myoblasts ([12], Fig 1D). Higher magnification imaging detected both epithelial cell cytonemes and myoblast cytonemes. Some of the cytonemes appear to extend between the tumor and mesenchymal populations (Fig 1E–1F’). These results show that tumor cells and tumor-associated cells extend cytoneme-like structures and are consistent with the possibility that cytonemes may facilitate signaling between these cell populations.
To monitor the tracheal branches that are associated with the tumorous discs, we induced the EGFR-Pcn tumor and labelled the trachea with membrane tethered GFP (with LHG lexO-CD2:GFP, a tracheal-specific driver [16]). In the EGFR-Pcn tumor discs, the associated trachea were more extensive and branched than normal (Fig 1G and 1H). Their overgrowth was presumably a response to the disc tumor.
In normal development, Dpp produced by wing disc cells at the anterior/posterior compartment border is transported by cytonemes to target cells in both the wing disc and ASP, and cytoneme deficits caused by Capricious (Caps), Neuroglian (Nrg), or Diaphanous (Dia) loss-of-function lead to developmental defects [7]. In the EGFR-Pcn tumor model, Dpp signals from the genetically altered epithelial cells to drive myoblast expansion [12]. Dpp expression is upregulated in the epithelial cells (Fig 2A and 2A’) and pMAD, the phosphorylated form of the Dpp signal transducer MAD, is enriched in the myoblasts (Fig 2A” and 2A’”). This indication of Dpp signal transduction in the myoblasts is consistent with previous results showing that Dpp signaling in this stromal compartment is required for tumor growth [12].
To investigate how Dpp moves in the EGFR-Pcn model, we used CRISPR mutagenesis to tag the endogenous Dpp protein with mCherry. mCherry was inserted at codon 465 as described in [17]. Homozygous Dpp:mCherry flies that lack a wild type dpp gene are viable and develop as wild type, indicating that the Dpp:mCherry chimera has normal function. We induced the EGFR-Pcn tumor in Dpp:mCherry flies, and labeled the EGFR-Pcn tumor cells with CD8:GFP. Dpp:mCherry fluorescence was present in the cytonemes of the epithelial tumor cells (Fig 2B, 2B’ and 2B”), consistent with the possibility that Dpp signaling is mediated by cytonemes in the EGFR-Pcn tumor model.
To assess the role of cytonemes in tumorigenesis, we examined discs in which cytonemes are impaired. Downregulation of Nrg, Caps, SCAR, or dia decreases the number and length of cytonemes, and decreases signaling in tracheal cells, myoblasts and wing disc cells [7,9,18]. Nrg and Caps are cell adhesion proteins and SCAR and Dia are actin-binding proteins. Although severe loss-of-function conditions for Nrg, caps, SCAR or dia are lethal, the partial loss-of-function conditions we used and previously characterized do not perturb cell polarity, cell viability, or cell cycle during normal development [7,9,19,20]. Previous studies of the wing disc and associated tracheal cells and myoblasts identified cytonemes that either “send” signaling proteins from producing cells or “receive” signaling proteins from target cells, and reported cytonemes that link disc cells to each other or to tracheal cells or myoblasts [9,18,20–22]. Available genetic tools can be used to impair cytoneme function but they do not distinguish among these types of cytonemes.
For the tumor discs with diminished Nrg, Caps, SCAR, or Dia, we compared disc morphology, Dpp signaling (monitored by anti-pMad antibody staining), and myoblast distribution (monitored by anti-Cut antibody staining, a marker of myoblasts [23]) in three types of wing discs: control non-tumor discs, EGFR-Pcn tumor discs and EGFR-Pcn tumor discs that also expressed CapsDN or RNAi constructs targeting Nrg, Dia, or SCAR. These genotypes were generated from two crosses. In the first, EGFR-Pcn tumor discs were generated from a cross between ap-Gal4,UAS-psqRNAi/CyO;UAS-EGFR,tub-Gal80ts and UAS-CD8:GFP that produces equal numbers of animals with the tumor genotype and non-tumor controls that have the CyO balancer and lack ap-Gal4,UAS-psqRNAi. The animals were incubated to the 2nd instar stage at low temperature (18°C) to permit repression of the transgenes by Gal80ts and were incubated at non-permissive temperature (29°C) thereafter (Fig 1C). The CyO control animals develop to late 3rd instar within one day and eclose in approximately four days as curly wing adults. All remaining animals developed tumors and were developmentally-delayed, and were analyzed after seven days of culture at 29°C. The second cross mated ap-Gal4,UAS-psqRNAi/CyO;UAS-EGFR,tub-Gal80ts to flies with the respective “tester chromosome” carrying UAS-CapsDN or UAS-RNAi, and were incubated with the same regimen involving removal of CyO balancer adults. The remaining larvae had tumor phenotypes to varied degrees.
Tumor discs were misshapen and approximately 6.3 times larger than control discs, their number of Cut-expressing cells increased by four times, and their anti-pMad staining was not patterned normally (Fig 3A and 3B). In contrast, discs with tumor cells that expressed NrgRNAi in addition to EGFR and Pcn were morphologically less distorted, only 1.8 times larger than controls, and the number and distribution of Cut-expressing cells was close to normal (Fig 3C and 3C’). In these animals, expression of EGFR, Pcn, and NrgRNAi is driven by ap-Gal4 continuously after the second instar, but the noxious effects of NrgRNAi suppress the tumor phenotype induced by EGFR and Pcn overexpression and are tolerated by the disc cells in which NrgRNAi is expressed. The implication is that tumor cells are more sensitive to the consequences of Nrg downregulation than are normal cells. Hyper-sensitivity to sub-lethal levels of toxic conditions is a common hallmark of tumor cells.
Expression of CAPSDN, SCARRNAi, or diaRNAi in the epithelial cells of the EGFR-Pcn model also reduced tumor growth, pMAD expression and number of Cut-expressing cells (Fig 3D–3F’). Expression of diaRNAi also suppressed excessive tracheation in the tumor discs (Fig 3S). To test whether the suppressive, ameliorative effects might be additive, we expressed NrgRNAi and diaRNAi simultaneously in EGFR-Pcn tumor cells. We did not observe that the degree of tumor suppression changed relative to expression of either NrgRNAi or diaRNAi alone (Fig 3G). We also tested the roles of three genes that are essential for planar cell polarity [24]: the dachsous (ds) and fat (ft) genes that encode cadherin family proteins, and four-jointed (fj) that encodes a transmembrane kinase. Expression of dsRNAi, ftRNAi or fjRNAi does not perturb cytoneme-mediated signaling between wing disc and tracheal cells [19], and expression of these RNAi lines in the tumor cells had no apparent effect on tumorigenesis (S2 Fig). Fig 3H summarizes the growth suppression we observed in the genetic conditions we tested.
The presence of cytonemes in both the tumor columnar epithelial and mesenchymal myoblast cells, and the essential role of the myoblasts for tumor progression raises the possibility that myoblast cytonemes might also play an essential role in tumorigenesis. To investigate the role of myoblast cytonemes, we expressed diaRNAi (with 1151-lexA lexO-diaRNAi) in the myoblasts of discs that overexpress EGFR and Pcn in the columnar epithelial cells. The morphology, Dpp signaling pattern and myoblast growth characteristic of the EGFR-Pcn tumors were suppressed (Fig 3I and 3I’). This result is consistent with the idea that the myoblasts signal to the epithelial tumor cells [12], and that this signaling is mediated by cytonemes.
We also analyzed the apical-basal organization of the disc cells by monitoring the distribution of Discs large (Dlg), which associates with the septate junction and localizes to the apical compartment of the columnar epithelial cells. Sagittal optical sections of discs stained with anti-Dlg antibody revealed that the specific apical distribution of Dlg characteristic of wild type cells is disorganized in EGFR-Pcn tumor discs (Fig 3J and 3J’). Expression of diaRNAi in the tumor cells restored the Dlg distribution to normal (Fig 3J”). This demonstrates that expression of diaRNAi suppresses a critical feature of tumor cells and that downregulation of Dia is compatible with normal cellular morphology and behavior.
Although EGFR and Pcn expression in the EGFR-Pcn model (driven by ap-Gal4) is restricted to the dorsal compartment of the wing disc, the tumors grow extensively and metastasize (Fig 4A). The tumorous condition is 100% lethal; animals with these tumors do not mature beyond the larval stage [12]. However, the conditions of Nrg, Caps, SCAR, or dia downregulation that suppress tumor growth also suppressed lethality: the number of EGFR-Pcn tumor-bearing larvae that pupated and that reached the pharate adult stage increased, and for the animals that expressed diaRNAi, approximately 60% survived to adult stage (Fig 4B). These surviving adults were fertile, and wing blade morphological defects were the only visible phenotype (Fig 4C and 4D). Given that cytoneme-mediated signaling is reduced by downregulation of Nrg, Caps, SCAR, or Dia, these results are consistent with the possibility that cytoneme-mediated signaling is necessary for tumor growth and that interfering with signaling either between tumor cells or between tumor and stromal cells suppresses many if not all aspects of tumorigenesis.
The disc-associated ASP branch of the tracheal system is dependent on and sensitive to signals produced by the disc [25,26], and Bnl signaling from the disc to the ASP is cytoneme-mediated and cytoneme-dependent [7,21]. Because tracheal branches grow excessively in the EGFR-Pcn tumor model (Fig 1H), we investigated if Bnl signaling is upregulated in tumor discs. Bnl is normally produced by a small, discrete group of disc cells (Fig 1A). Disc cells do not express Btl, but tracheal cells express Btl and not Bnl [26]. To monitor Bnl signaling in the EGFR-Pcn tumor model, we examined a Bnl reporter that expresses mCherry:CAAX in Bnl-expressing cells [27]. The number and location of Bnl-expressing cells increased in tumor discs (Fig 5A and 5B). We also examined fluorescence of Btl:mCherry (with a CRISPR-generated knock-in [21]). Whereas Btl:mCherry fluorescence was not detected in the epithelial cells of normal wing discs (Fig 5C and 5C’), Btl:mCherry fluorescence was present in many epithelial cells of the tumor (Fig 5D and 5D’). These results are consistent with the possibility that the tumor induces ectopic expression of Btl and that ectopic activation of the Bnl signaling pathway might correlate with excessive growth of the tracheal branches in this tumor.
To investigate the role of Bnl signaling in EGFR-Pcn tumorigenesis, we overexpressed a dominant negative FGFR mutant in the tumor cells to block Bnl signaling (UAS-BtlDN). We monitored wing discs for morphology, Cut expression, and pMAD in EGFR-Pcn tumor discs, and EGFR-Pcn tumor discs that also express BtlDN. Experimental crosses were carried out with the regimen described previously and produced either tumor larvae or suppressed tumor larvae; both crosses generate control (CyO balancer) and tumor-containing animals in a Mendelian ratio of 1:1. In the cross with UAS-BtlDN, 50.6% (76/151) of the larvae pupated and eclosed within 4 days as curly wing adults. The presence of the balancer chromosome indicates that the genotype of these flies lacked ap-Gal4, as expected of control, non-tumor animals. The remaining larvae do not develop beyond the pupal stage, consistent with their having the tumor genotype. Larvae analyzed after 7 days of Gal4 expression at the non-permissive were compared to tumor discs of the same age (Fig 1C). In the EGFR-Pcn tumor discs that also express BtlDN, characteristics of tumor morphology, size, pattern of Dpp signaling, and distribution of myoblasts were suppressed (Fig 5E and 5F).
To confirm the identity genotype of the suppressed tumor discs, RNA isolated from wild type, EGFR-Pcn tumor, and BtlDN-expressing tumor discs was quantified by QPCR. This analysis confirmed the overexpression of EGFR in both tumor and suppressed tumor discs (Fig 5G). We also examined EGF and Bnl signal transduction in tumor and suppressed tumor discs by staining with anti-dpERK antibody. The presence of dpERK was observed in control, tumor, and BtlDN -over-expressing control and tumor discs, and whereas the pattern of dpERK in the tumor discs was expanded and disordered in the tumor discs, the patterns and levels in the suppressed discs was close to normal (Fig 5H–5K). These findings are consistent with the idea that tracheogenesis is necessary for tumor growth and with a previous report that describes comparable findings in studies of a lethal giant larvae Drosophila tumor model [28]. In this tumor, ectopic tracheal sprouting is associated with hypoxic responses and tracheal differential of wing disc tumor cells, a process that may be analogous to “sprouting angiogenesis” and vascular co-option in mammalian tumors [29].
We investigated the role of cytonemes in the Drosophila RET-MEN2 tumor model developed by the Cagan lab [14]. This model mimics the mis-regulation of signaling pathways that have been implicated in MEN2-related tumors. Overexpression of RETMEN2 in a discrete set of wing disc epithelial cells (with ptc-Gal4) resulted in a >4X increase in the number of ptc-expressing cells and a 7X increase in the portion of the disc that consists of ptc-expressing cells (Fig 6A and 6B) [14]. Approximately one-half of the animals survive to the pupal stage, but none survive to adult. We tested whether expression of Irk2DN (an inwardly-rectifying potassium channel required for cyteneme-mediated signaling [5]), diaRNAi, or SCARRNAi in the RET-mutant cells affects tumor growth and survival. We observed that excessive growth of the ptc-expressing cells was suppressed by more than 2X in all three genotypes (Fig 6C–6F). Approximately two-thirds of the animals developed to the pupal stage, and survival to adult also increased (Fig 6G). These flies have normal morphology, and with the exception of small wing vein abnormalities, the wings are indistinguishable from wild type (Fig 6H). These results are consistent with a general role for cytonemes in tumorigenesis and tumor progression.
The tumor microenvironment is a niche that responds to signaling proteins produced by tumor cells and supplies growth factors that support tumor growth and metastasis [30]. Much ongoing work seeks inhibitors of tumorigenesis that target the signaling molecules and growth factors, their signal transduction pathways, and the stromal cells of the microenvironment [31,32]. Two previous studies reported cellular extensions of human tumor cells in ex vivo co-cultures with non-metastatic cells and in vivo, and have been implicated these structures in material transfer between tumor and non-tumor cells [33,34]. In this work, we also investigated the mechanism that transfers signaling molecules and growth factors between tumor cells and stromal cells in vivo, and report the first evidence for their essential role in tumorigenesis.
Previous work established that during Drosophila development, paracrine signaling by the signaling proteins/growth factors Dpp, Bnl, Wg, Notch and Hedgehog, is mediated by cytonemes [35–37]. Cytonemes are specialized filopodia that extend between signal producing and signal receiving cells, making synaptic contacts where the signaling proteins transfer from producing to receiving cells. To extend this work to tumorigenesis, we applied the strategies and tools we developed for previous studies to ask if cytonemes are present in the tumor microenvironment, and if genetic conditions that inhibit cytoneme function and cytoneme-dependent signaling in normal development also inhibit tumorigenesis. In a EFGR-Pcn tumor model, we found that cytonemes extend from both Drosophila tumor and stromal cells (Fig 1). This is consistent with previous studies that reported increased signaling between tumor and stromal cells in this model [12], and with the presence of cytonemes in many other contexts of paracrine signaling [7,18,20,38–41]. We confirmed that Dpp is expressed by the tumor cells (Fig 2; [12]), and found that ectopic Bnl signaling also has an essential role in this tumor (Fig 5). These results imply functional connections between the EGF, Dpp, and Bnl signaling pathways in this tumor, and although we did not identify regulatory interactions between the pathways, our results show that ectopic activation of the Bnl pathway is essential to tumorigenesis.
We also found conditions that impair cytonemes and rescue flies of lethal tumors in both EGFR-Pcn and RET models. We selected five genes from among the more than twenty that are known to be essential for cytoneme-mediated signaling [5,7,18,19]. nrg, caps, Irk2, SCAR, and dia are recessive lethal genes whose functions can be partially reduced in genetic mosaics without affecting viability, cell shape, or the cell cycle, but are necessary for cytoneme function. Downregulating any one of these genes improved viability in the tumor models. dia downregulation is the most effective inhibitor of cytoneme-mediated signaling in other contexts [7,19,20], and it is the most effective in both tumor models. The cures that downregulation effected suggest that cytoneme-mediated signaling, which might be a general mechanism for tumorigenesis in a variety of cancers, might also be a potential target for therapy.
The high degree of evolutionary conservation of Drosophila and human proteins makes Drosophila a clinically relevant platform for understanding mechanisms human disease, and Drosophila tumor models have successfully identified new therapeutic candidates for colorectal, lung and thyroid and stem-cells derived cancers [42–44]. Our work provides proof principle for tumor suppression by interfering with cytoneme-mediated signaling.
Flies were reared on standard cornmeal and agar medium at 29°C, unless otherwise stated. ap-Gal4 UAS-psqRNAi/CyO; UAS-EGFR tub-Gal80ts from S. Cohen [12], UAS-RETMEN2 from R. Cagan [14], Btl:mCherry and Bnl-lexA, from S. Roy [21,27], lexO-diaRNAi from H. Huang, UAS-CapsDN [45] (deletion mutant lacking the intracellular domain), UAS-BtlDN from B. Shilo [46] (dominant negative construct lacking a functional cytoplasmic tyrosine-kinase domain), Irk2DN from E. Bates [47] (a subunit predicted to block the channel). btl-LHG,lexO-CD2:GFP, a tracheal-specific driver [16]; ptc-Gal4 enhancer is an enhancer trap line that mimics ptc expression [48], lexO-mCherry:CAAX from K. Basler; lines from Bloomington Stock Center: 15B03-lexA (#52486), UAS-CD8:GFP (#5137), UAS-diaRNAi (#28541 and #35479), UAS-NrgRNAi (#37496), UAS-dsRNAi (#32964), UAS-ftRNAi (#34970), UAS-fjRNAi (#34323); and UAS-SCARRNAi (#21908) from Vienna Drosophila Research Center Stock Center.
The Dpp:mCherry transgene has mCherry inserted C-terminal to Dpp amino acid 465 [17], with Leu-Val linkers inserted before and after a mCherry coding sequence deleted of its stop codon. The transgene was generated by CRISPR mutagenesis as follows:
Left homology arm fragment contains overlapping sequence with PBS-SK vector and mCherry. The mCherry fragment contains overlapping sequence with the left homology arm and right homology arm. The right homology arm fragment contains overlapping sequence with mCherry and PBS-SK vector. The three fragments were stitched together and cloned into PBS-SK vector using Gibson Assembly (NEB). The resulting vector is designated as Dpp:Cherry donor vector.
Left arm homology sequence was amplified from wild-type genomic DNA using:
L-arm-fwd: cggtatcgataagcttgatcaccttgccgcacaaatacatatac
L-arm-rev: CCTCGCCCTTGCTCACCATCTCCAGGCCACCGCCCTCTCCGGCAGACACGTCCCGA
The mCherry tag was amplified using:
mCherry-fwd:TGTCTGCCGGAGAGGGCGGTGGCCTGGAGATGGTGAGCAAGGGCGAGGAGGATAAC
Cherry-rev:CGCTTGTTCCGGCCGCCCTTCTCTAACTTGTACAGCTCGTCCATGCCGC
The right arm homology sequence was amplified from wild-type genomic DNA using:
R-arm-fwd:GGACGAGCTGTACAAGTTAGAGAAGGGCGGCCGGAACAAGCGGCAGCCGA
R-arm-rev:ccgggctgcaggaattcgatGTCATTATTCGGTTATGCTCTCGCTAG
pCFD-3 gRNA vector
gRNA sequence: CGCTCCATTCGGGACGTGTCTGG
The gRNA sequence without the PAM was cloned into pCFD-3 vector obtained from Addgene.
pCFD-3 gRNA vector and Dpp:mCherry donor vector were co-injected into Cas9 expressing flies (nanos-Cas9) by Rainbow Transgenics. The resulting CRISPR-generated flies were screened and verified by sequencing. The Dpp:mCherry homozygous fly is viable and has normal morphology. The distribution of Cherry fluorescence in the wing disc is consistent with images in [17,49], and the gradient of Cherry fluorescence in the columnar epithelial cells of the disc is intracellular (S1 Fig).
EGFR-Pcn tumors were induced as described by Herranz et al, [12] by overexpression of EGFR and down-regulation of pipsqueak (psq), which leads to increased levels of Pcn. Female flies from the stock ap-Gal4,UAS-psqRNAi/CyO;UAS-EGFR,tub-Gal80ts were crossed to males of the corresponding genotypes at 18°C, and were cultured at 18°C to maintain Gal80 repression of Gal4 and allow normal development. After 5 days larvae were transferred to 29°C to induce Gal4 expression and tumor growth. 4 days after the temperature shift CyO/+ flies eclosed and were removed from the vial. Tumor growth was induced for 7 days, unless otherwise indicated, whereupon larvae were dissected for live imaging or immunostaining, or were maintained at 29°C for survival studies. To control for possible effects on Gal4 expression, all tested genotypes had three UAS transgenes–either UAS-EGFR, UAS-psqRNAi and UAS-CD8:GFP for tumor flies, or UAS-EGFR, UAS-psqRNAi and additional RNAi for comparisons. Experimental and control crosses were carried out in parallel.
Female flies from the RETMEN2 stock [14] were crossed at room temperature to either ptc-Gal4, 2xUAS-CD8:GFP males or either with UAS-diaRNAi, or SCARRNAi or Irk2DN males. For analysis of discs, embryos from one day collections were transferred to 29°C and cultured to third instar stage. For survival comparisons, animals were cultured at 25°C.
Wing discs with trachea attached were dissected in cold phosphate-buffered saline (PBS), placed on a coverslip and mounted upside-down on a coverslip on a depression slide as described [9]. Samples were imaged with a Leica TCS SPE confocal or an Olympus FV3000 inverted confocal laser scanning microscope.
Wing discs were dissected in cold PBS and fixed in 4% formaldehyde for 20 minutes. After extensive washing, the samples were permeablized with PBST (PBS + 0.3% TritonX-100), blocked for 1h with PBST+3%BSA blocking buffer, and incubated with primary antibodies previously diluted in blocking buffer overnight at 4°C. The following primary antibodies were used: α-pMad (Abcam), α-Discs large (Dlg), α-Cut and α-β-galactosidase (Developmental Studies Hybridoma Bank). Secondary antibodies were conjugated to Alexa Fluor 405, 488, 555, or 647. Samples were mounted in Vectashield and imaged with a Leica TCS SPE confocal or an Olympus FV3000 inverted confocal laser scanning microscope.
All measurements and quantifications of wing discs were done in z-section stacks of confocal images using Fiji software from 15–20 discs for each genotype. Total wing disc area and Cut-expressing cells in the EGFR-Pcn tumor or GFP-expressing cells in the RET tumor, were quantified by measuring the mean intensity of fluorescence relative to the total area of the wing disc. Data was normalized to control. Statistical significance values were calculated with Student’s t test.
Total RNA was extracted from 5 wing discs of either wild type, EGFR-Pcn tumor or EGFR-Pcn tumor + BtlDN larvae using the RNeasy Micro Kit (Quiagen). Larvae corresponding from 3 genotypes were under the same temperature conditions (5 days of tumor induction at 29°C). Reverse transcription was carried out using the Applied Biosystem High Capacity RNA-to-cDNA. qPCR reactions were performed with a BioRad C1000 Touch Thermal Cycler and SYBR Green (Bioline). qPCR results were analyzed according to the comparative threshold cycle (Ct) method, where the amount of target, normalized to an endogenous actin reference and relative to an experimental control, is given by 2–ΔΔCt. Ct represents the PCR cycle number at which the amount of target reaches a fixed threshold. The ΔCt value is determined by subtracting the reference Ct value (rp49) from the target Ct value. ΔCt was calculated by subtracting the ΔCt experimental control value.
|
10.1371/journal.pcbi.1005702 | Revisiting chemoaffinity theory: Chemotactic implementation of topographic axonal projection | Neural circuits are wired by chemotactic migration of growth cones guided by extracellular guidance cue gradients. How growth cone chemotaxis builds the macroscopic structure of the neural circuit is a fundamental question in neuroscience. I addressed this issue in the case of the ordered axonal projections called topographic maps in the retinotectal system. In the retina and tectum, the erythropoietin-producing hepatocellular (Eph) receptors and their ligands, the ephrins, are expressed in gradients. According to Sperry’s chemoaffinity theory, gradients in both the source and target areas enable projecting axons to recognize their proper terminals, but how axons chemotactically decode their destinations is largely unknown. To identify the chemotactic mechanism of topographic mapping, I developed a mathematical model of intracellular signaling in the growth cone that focuses on the growth cone’s unique chemotactic property of being attracted or repelled by the same guidance cues in different biological situations. The model presented mechanism by which the retinal growth cone reaches the correct terminal zone in the tectum through alternating chemotactic response between attraction and repulsion around a preferred concentration. The model also provided a unified understanding of the contrasting relationships between receptor expression levels and preferred ligand concentrations in EphA/ephrinA- and EphB/ephrinB-encoded topographic mappings. Thus, this study redefines the chemoaffinity theory in chemotactic terms.
| This study revisited the chemoaffinity theory for topographic mapping in terms of chemotaxis. According to this theory, the axonal growth cone projects to specific targets based on positional information encoded by chemical gradients in both source and target areas. However, the mechanism by which the chemotactic growth cone recognizes its proper terminal site remains elusive. To unravel this mystery, I mathematically modeled a growth cone exhibiting concentration-dependent attraction and repulsion to chemotactic cues. The model identified a novel growth cone guidance mechanism in topographic mapping, highlighting the importance of the growth cone’s unique ability to alternate between attraction and repulsion. Furthermore, an extension of the model provided possible molecular mechanisms for contrasting two types of topographic mappings observed in the retinotectal system.
| During development, neurons extend axon and dendrites [1–3] and axonal growth cones chemotactically migrate in response to extracellular guidance cue gradients and connect to their target sites. Because this axon guidance is a fundamental process in wiring neural circuits, many guidance cues and receptors have been identified and their functional roles (e.g., attraction or repulsion) have been extensively investigated [4–6]. The growth cone’s chemotactic properties are thus being unveiled at the molecular level, but the chemotactic mechanisms of neural circuit construction remain mysterious at the macroscopic level. I addressed this issue by investigating topographic maps, the ordered axonal projections ubiquitous in the sensory nervous system. The best-studied example is in visual system, where retinal ganglion cells (RGCs) project their axons to the optic tectum and/or superior colliculus (SC) while keeping an initial positional relation [7].
The most important concept of topographic map formation is the “chemoaffinity theory” proposed by Roger Sperry in 1940s [8]. Sperry proposed that chemical labels form gradients in source and target areas, allowing a projecting axon to recognize its target site. The theory’s molecular basis was identified with the discovery of gradients of erythropoietin-producing hepatocellular (Eph) receptors and their ligands, ephrins, in the retina (source area) and tectum (target area) [9,10]. Ephs and ephrins are classified into two families, A and B, that encode orthogonal topographic maps in the retina and tectum (Fig 1).
The EphA receptor gradient along the retina’s nasal-temporal axis topographically corresponds to the ephrinA gradient along the tectum’s rostral-caudal axis (Fig 1A). On the orthogonal coordinates, the EphB receptor gradient along the retina’s dorsal-ventral axis corresponds to the ephrinB gradient along the tectum’s medial-lateral axis (Fig 1C). These facts suggest that RGC growth cones chemotactically migrate to their terminal zones guided by ligand concentrations reflective of receptor expression levels. Because ephrinA and ephrinB act as both attractants and repellents in a concentration-dependent manner [11–13], it is possible that growth cones switch between attraction and repulsion around the terminal zone, but the chemotactic mechanism for decoding destination from dual gradients (i.e., receptor and ligand) is unknown.
The EphA/ephrinA- and EphB/ephrinB-encoded topographic maps differ in that the RGCs with higher EphA receptor expression prefer lower tectal ephrinA concentrations (Fig 1B), whereas the RGCs with higher EphB receptor expression prefer higher tectal ephrinB concentrations (Fig 1D). In other words, the retinotectal system’s two kinds of topographic mapping have opposite receptor expression-dependent ligand concentration preferences. How the growth cone’s chemotactic system implements these opposite preferences is also unknown.
Topographic mapping has been extensively investigated with computational models for four decades [14], but all previous models featured growth cones reaching their terminal zones by heuristically-designed chemoaffinity [15–25]. While these models provided insights into the outcomes of surgical experiments in the retinotectal system [15–17] and the abnormal maps resulting from misexpression of Ephs or ephrins [15,18–25], none addressed how the intracellular mechanism of growth cone chemotaxis achieves chemoaffinity.
I sought to determine the underlying mechanism of topographic mapping implemented by growth cone chemotaxis. To this end, I focused on the growth cone’s unique chemotactic property of being attracted and repelled by the same guidance cues in different biological environments [26,27]. By mathematically modeling growth cone migration regulated by intracellular signaling, I attempted to demonstrate how the growth cone reaches its terminal zone in the tectum by switching attraction and repulsion around a preferred ligand concentration. Through this model, I redefined Sperry’s chemoaffinity theory in terms of chemotaxis.
I first studied the projecting growth cone’s preference for a specific ligand concentration associated with the correct terminal zone in the target area. The basic idea is that a growth cone switches between attraction and repulsion around a specific preferred concentration; if the growth cone exhibits attraction and repulsion to lower and higher concentrations, respectively, then it ultimately reaches a location with the preferred concentration. To examine this idea, I mathematically modeled intracellular signaling in chemotactic growth cones.
The model growth cone was equipped with an intracellular activator (A) and inhibitor (I) of their effector (E), where A and I were upregulated by guidance cues and E regulates the growth cone motility (Fig 2A and 2B). This activator-inhibitor framework has been commonly observed in both neural and non-neural chemotactic cells [28–31]. For simplicity, a one-dimensional coordinate (x) across the growth cone was modeled as {x| − L/2 ≤ x ≤ L/2}, where L indicates its length. The reaction-diffusion dynamics of A and I were described by
∂A∂t=DA∂2A∂x2−kAA+cA+αAG(x)∂I∂t=DI∂2I∂x2−kII+cI+αIG(x)
(1)
with reflecting boundaries at both ends (x = ±L/2), where A and I represent the activities of A and I, respectively, DZ, kZ, cZ, and αZ (Z ∈ {A,I}) denote the diffusion constant, decay rate, constant input, and efficacy, respectively, of the guidance cue’s signal transmission, and G(x) represents the guidance cue concentration at x. The activity of E was determined by the ratio of A’s activity to I’s, i.e., E(x) = A(x)/I(x), which is reasonable if E is regulated by a push-pull enzymatic reaction between A and I [32,33]. The growth cone’s migration was driven by the relative spatial polarity of E as ΔE/E*, where ΔE and E* indicate the spatial difference of E across the growth cone (i.e., E(L/2) − E(−L/2)) and the baseline activity of E (i.e., E(0)), respectively. This property was stated as the Weber-Fechner law, in which the detectable spatial polarity of E varies because of the scale of the concentration of E [34]. Indeed, the Weber-Fechner law has been found in several types of chemotactic cells [35–40]. By analytically solving the model (see Methods), I demonstrated that it produced opposite polarities for ΔE depending on the parameters (Fig 2C and 2D and Table 1); when ΔE > 0, the growth cone was attracted and migrated along the gradient, but when ΔE < 0, the growth cone was repelled and turned against the gradient.
I examined how chemotactic responses vary with absolute concentrations in the gradient. My previous study [27] showed that the steady-state response of ΔE/E* was presented by
ΔEE*=ΔAA*−ΔII*,
(2)
where A* and I* denote the baseline activities of A and I, respectively (i.e., A* = A(0) and I* = I(0)), and ΔA and ΔI denote the spatial differences of A and I, respectively, across the growth cone (i.e., ΔA = A(L/2) − A(−L/2) and ΔI = I(L/2) − I(−L/2)) (see Fig 2D). Z* and ΔZ (Z ∈ {A,I}) were analytically derived (see Methods). By substituting these into Eq (2), I found that four chemotactic response patterns were generated depending on parameters (Fig 3A and Table 1): unidirectional repulsion, unidirectional attraction, bidirectional repulsion-to-attraction, and bidirectional attraction-to-repulsion (BAR). In the former two patterns, the growth cone always exhibited attraction or repulsion, meaning that it preferred higher or lower concentrations, respectively (Fig 3C and 3D). In bidirectional repulsion-to-attraction, the growth cone preferred either higher or lower concentrations depending on the initial concentration (Fig 3B). Finally, in BAR, the growth cone avoided both higher and lower concentrations but preferred a specific concentration by switching attraction and repulsion at that concentration (Fig 3E). I hypothesized that this BAR pattern could play a fundamental role in topographic map formation.
Assuming that the growth cone exhibited the BAR pattern, I studied how receptor expression levels affected the preferred concentration. To this end, the receptor was incorporated into the model as follow:
∂A∂t=DA∂2A∂x2−kAA+cA+αAf(R,G(x))∂I∂t=DI∂2I∂x2−kII+cI+αIf(R,G(x)),
(3)
where R represents the expressed receptor’s density, and f(R,G) represents the density of the receptor’s active form depending on the guidance cue concentration. By analyzing this model based on Eq (2) (see Methods), I found that whether the preferred concentration, Gpref, decreases or increases with R was determined by the sign of derivatives of f(R,G) with respect to R and G:
dGprefdR=−∂f/∂R∂f/∂G.
(4)
Therefore, f(R,G), which represents how the guidance cue signal is transmitted to A and I through the receptor, is a crucial factor in the receptor expression level-dependent preferred ligand concentration. I next studied specific examples of f(R,G).
I considered a scenario in which the receptors were activated by guidance cue binding (Fig 4A), which is described by f(R,G) = RG/(K + G), where K is the dissociation constant of binding reaction between the receptor and guidance cue (i.e., a ratio of unbinding rate to binding rate). I then calculated the preferred concentration based on Eq (2) and found that it decreased with the receptor expression level (Fig 4B) as
Gpref∝1R−γ,
(5)
where γ is a positive constant determined by the model parameters. This is consistent with type 1 topographic mapping in which higher EphA levels result in the growth cone preferring smaller ephrinA concentrations (Fig 1A and 1B). If the receptor expression level is greater than γ, this relationship produces a linearly ordered topographic map with exponential distributions of retinal EphA and tectal ephrinA (Fig 4C).
For the mechanism of type 2 EphB/ephrinB-encoded topographic mapping, I tested two biologically plausible hypothetical f(R,G) expressions. First, guidance cue-unbound receptors might trigger intracellular signaling, which can be expressed by f(G) = RK/(K + G) (Fig 4D(i)) (see Discussion for its biological relevance). For this hypothesis, I found that the preferred concentration increases with the receptor expression level (Fig 4E) as
Gpref∝R−γ.
(6)
This is consistent with the fact that higher EphB levels result in the growth cone preferring higher ephrinB concentrations (Fig 1C and 1D) (see Methods). This linear relationship (Eq (6)) produced a linearly ordered topographic map with exponential distributions of retinal EphB and tectal ephrinB (Fig 4F).
In the second hypothesis, I assumed that two kinds of receptor competitively bind the limited ligands (Fig 4D(ii)) (see Discussion for its biological relevance). One kind is uniformly expressed across the retina and the guidance cue-bound form triggers intracellular signaling. The other is expressed in gradients across the retina and indirectly inhibits the uniformly expressed receptor by competitively binding the ligand. This case is described by f(R,G) = RcG/(K + Rc + R) (Methods), where R and Rc indicate densities of the receptors expressed in gradients and uniformly, respectively, and K indicates the dissociation constant of the receptor and ligand. For this hypothesis, I also found that the preferred concentration increases with the receptor expression level as
Gpref∝R+ρ,
(7)
where ρ is a positive constant determined by the model parameters. Thus, this hypothesis also explained the type 2 EphB level-dependent preferred ephrinB concentration.
I presented a mathematical model of chemotactic response of the growth cone to reveal how topographic map is formed by the growth cone chemotaxis. In my model, for the sake of simplicity, I assumed that the migration direction of the growing axon was determined by polarity of the growth cone signaling. The real mechanism must be more complicated than what assumed in my model. However, the minimalist model I developed was very informative and provided a novel chemotaxis-based logic of chemoaffinity theory for topographic mapping. I demonstrated that the model could generate both attractive and repulsive responses depending on absolute concentrations along the gradient. Such bidirectionality endows the growth cone with the preference for a specific guidance cue concentration by switching between attraction and repulsion around that concentration. I also determined the conditions of EphA/ephrinA- and EphB/ephrinB-encoded topographic mapping, in which the preferred concentration decreases and increases, respectively, with the receptor expression level. This study therefore redefined Sperry’s chemoaffinity theory in terms of chemotaxis.
If ephrinA is a repellent, as classically thought [41], then all RGC growth cones must project to the tectum’s rostral end, which has the lowest ephrinA concentration. However, this is not the case; even without tectal space competition between projecting axons, the RGC axons project to the correct terminal zone in the tectum [42]. This contradiction can be resolved simply by regarding ephrinA as both an attractant and a repellent. In fact, ephrinA has been reported to be an attractant or a repellent in a concentration-dependent manner [11]. EphrinB has been regarded as both an attractant and a repellent [12,13]. However, their underlying mechanism was largely unknown. In this study, I demonstrated how ephrinA and ephrinB could indeed work as both attractants and repellents for the chemotactic growth cone.
I demonstrated that whether the preferred ligand concentration decreases or increases with the receptor expression level is determined by whether the guidance cue and the receptor positively or negatively affect intracellular signaling (Eq (4)). As the mechanism of EphA/ephrinA-encoded type 1 topographic mapping, I reasonably assumed that ephrinA-bound EphAs trigger intracellular signaling (Fig 4A), but for type 2 topographic mapping, I tested two hypothetical EphB/ephrinB regulation schemes. The first hypothesis was that ephrinB-unbound EphBs, rather than bound ones, trigger intracellular signaling (Fig 4D(i)). This seems inconsistent with a property of tyrosine kinase-type receptors, which are activated by ligand binding through phosphorylation [43,44], but it has recently been reported that Ephs can be ligand-independently activated by hemophilic Eph-Eph interactions [45], suggesting that ephrinB-bound and -unbound EphBs could generate different signals. The first hypothesis was thus biologically feasible, but further experimental investigation is needed. The second hypothesis was that two kinds of receptor, which are expressed uniformly or in gradients across the retina, competitively bind the ligand (Fig 4D(ii)). This fits the expression profiles of EphB subtypes in the chicken retina well; EphB2 and EphB3 are expressed in gradients across the retina, whereas EphB1 is uniformed expressed [7]. My hypothesis thus offers experimentally testable predictions concerning EphB/ephrinB regulation.
It is worth mentioning functional difference between type 1 and type 2 of topographic mappings. I deduced that local accuracy of axonal projection is determined by multiplication of three factors: 1. spatial derivatives of receptor expression profile in retina (upper panels in Fig 4C and 4F), 2. steepness of mapping function from receptor expression to preferred ligand concentration (Fig 4B and 4E) and 3. spatial derivatives of ligand in tectum (right panels in Fig 4C and 4F). In type 1 topographic mapping, while multiplication of the first and third factors, i.e., (dRNT/dxNT)(dGRC/dxRC), is constant (Fig 4C), the second factor, i.e., the steepness of mapping function, increases as EphA expression decreases (Fig 4B). On the other hand, in type 2 topographic mapping, while the second factor, i.e., the steepness of mapping function, is constant (Fig 4E), multiplication of the first and third factors, i.e., (dRDV/dxDV)(dGML/dxML), increases with EphB expression. Thus, it can be predicted that axonal projection from nasal ventral retinal region associated with lower EphA and higher EphB expression could be more precise than that other retinal region.
RGCs’ axonal projection patterns in the optic tectum or SC are species-dependent. In higher vertebrates (i.e., mammals and birds), the axons overshoot their terminal zones and subsequently form branches [7], while in lower vertebrates (i.e., fish and amphibians), the growth cones directly reach and stop in their terminal zones [7] despite being initially misrouted [46]. The latter case suggests that the chemotactic system implements chemoaffinity, which I investigated as the mechanism of topographic mapping. The growth cone’s chemotaxis might therefore play a fundamental role in topographic mapping, while axonal overshoot and branching might facilitate exploration of the terminal zone. My model could understand the axonal overshoot by incorporating transient dynamics of activator and inhibitor, instead of steady state assumption. On the other hand, how the axon generates branches is out of scope of my model.
Chemotactic gradient sensing has been computationally studied mainly for non-neural chemotactic cells [40,47–51] like Dictyostelium discoideum and immune cells, though attraction to guidance cues has been only paid attention. On the other hand, there are a couple of computational models for the growth cone chemotaxis alternating attraction and repulsion [27,52]. These models, whether applied to neural or non-neural cells, primarily addressed intracellular signaling consisting of activators and inhibitors. In the non-neural cells, the activator and inhibitor were thought to be PI3K and PTEN [28], respectively, or RasGEF and RasGAP [29], respectively. In the growth cone, CaMKII and PP1 were thought to work as the activator and inhibitor, respectively [27,30,31,52], which regulate cellular motility via Rho GTPases [53]. In short, chemotactic responses could be understood from the activator-inhibitor framework [54], so I hypothesized that RGC chemotaxis is also regulated by an activator-inhibitor system, although the intracellular signaling pathway of Eph/ephrin has not been fully identified.
There have been many computational studies on topographic mapping [14]. These studies did not focus on the intracellular mechanism of growth cone chemotaxis, but instead developed models with heuristically designed chemoaffinity (e.g., optimization of energy function) by which the growth cone reaches its correct terminal zone. Given such chemoaffinity, these models potentially gave insights into more system-level phenomena, such as abnormal maps resulting from surgical experiments in the retinotectal system [15–17] and from the misexpression of Eph or ephrin [15,18–25]. These models included several factors not included in my model, such as axon competition for tectal space [55] and counter-gradients of Ephs and ephrins in the retina and tectum [56]. Several models have also addressed a question of how synaptic connection is refined by activity-dependent synaptic plasticity mechanism after activity-independent axon guidance [20,57–59]. Therefore, I must stress that my model does not compete with previous models, but rather can explain the underlying mechanism by which growth cones can chemotactically implement the previous models’ heuristically designed chemoaffinity.
Suppose a shallow extracellular gradient because growth cones are known to detect few percent difference of concentrations across the growth cone [60–64]. I then assumed that the intracellular gradients of A and I, A(x) and I(x), were shallow and slightly perturbed from their activities at x = 0. The activity of E at x could be linearized as
E(x)≃E*+1I*[A(x)−A*]−A*I*2[I(x)−I*],
(8)
where A* = A(0), I* = I(0), and E* = E(0) = A*/I*. The relative spatial difference of E across the growth cone was calculated by
ΔEE*≡E(L/2)−E(−L/2)E(0)=ΔAA*−ΔII*,
(9)
where ΔA and ΔI indicate the spatial differences of A and I, respectively, across the growth cone.
For both A and I, I calculated the intracellular distribution exposure to an extracellular gradient, G(x). Green’s function of ∂Z/∂t = Dz(∂2Z/∂x2) − kzZ was analytically derived using the method of separation of variables:
H(x,ξ,t)=1Lexp(−kZt)+2L∑n=1∞cos[nπL(ξ+L2)]cos[nπL(x+L2)]exp[−{kZ+(nπL)2DZ}t].
(10)
A steady-state solution of Eq (3) was thus obtained by
Z∞(x)=∫0∞dτ∫−L/2+L/2dξH(x,ξ,τ){cZ+αZf(R,G(ξ))},
(11)
where Z represents either A or I. Note that f(R,G(x)) = G(x) in Eq (1). Because the growth cone is so small that G(x) could be modelled as a shallow linear gradient, f(R,G(x)) can be linearized by f(R,G*) + gx, where G* = G(0) and g = (∂f/∂G|G = G*)(dG/dx|x = 0). This led to
Z∞(x)=Z*+2gL∑n=1∞(L/nπ)2[(−1)n−1]kZ+(nπ/L)2DZcos{nπL(x+L2)},
(12)
where Z* indicates baseline activity, i.e., Z* = Z∞(0):
Z*=αZf(R,G*)+cZkZ.
(13)
By numerical simulation of the reaction-diffusion dynamics, I confirmed that Eq (12) was exact. The spatial difference of Z then becomes
ΔZ=Z∞(L/2)−Z∞(−L/2)=8gL3π4h(DZ/kZ)kZ,
(14)
where
h(s)=∑n=1∞1/(2n+1)2(2n+1)2s+(L/π)2,
(15)
which is a monotonically decreasing function converging to 0 (inset of Fig 3A).
I calculated the growth cone’s concentration-dependent chemotactic responses. By substituting Z* as described by Eq (13) for A* and I* in Eq (2) and substituting ΔZ as described by Eq (14) for ΔA and ΔI in Eq (2), I obtained
ΔEE*=8gL3π4[h(DA/kA)αAf(R,G*)+cA−h(DI/kI)αIf(R,G*)+cI].
(16)
Eq (16) exhibits four response patterns to G*: all positive, all negative, negative-to-positive, and positive-to-negative, which correspond to unidirectional attraction, unidirectional repulsion, bidirectional repulsion-to-attraction, and BAR, respectively (Fig 3B–3E). The response patterns’ parameter regions were derived under the condition of ∂f/∂G > 0 (Fig 3A). For example, the BAR response pattern is characterized by attraction at lower concentrations (i.e., ΔE/E*|G* = 0 > 0) and repulsion at G* = ∞ (i.e., ΔE/E*|G* = 0 < 0), which leads to
cAcI<η<αAαI,
(17)
where η = h(DA/kA)/h(DI/kI).
Growth cones with the BAR response pattern prefer a specific concentration of G* at which ΔE/E* = 0. In the Eq (3) model, setting ΔE/E* = 0 in Eq (16) leads to
f(R,Gpref)=γ,
(18)
where γ = (ηcI − cA)/(αA − ηαI). The preferred concentration with a specific f(R,G) can be calculated with Eq (18). In the Eq (1) model, f(R,G) = G, thus Gpref = γ. If f(R,G) = RG/(K + G), Gpref = γK/(R − γ) (Eq (5); Fig 4A). If f(R,G) = RK/(K + G), Gpref = (K/γ)(R − γ) (Eq (6); Fig 4D(i)). If f(R,G) = RcG/(K + Rc + R), Gpref = (γ/Rc)/(R + K + Rc) (Eq (7); Fig 4D(ii)). Total differentiation of Eq (18) leads to (∂f/∂R)dR + (∂f/∂Gpref)dGpref = 0, which in turn leads to
dGprefdR=−∂f/∂R∂f/∂G.
(19)
I assumed a scenario in which two kinds of RGC-expressed receptors competitively bind limited ligands with identical kinetics. Note that the two assumed kinds are expressed either uniformly or in gradients across the retina. Such dynamics are described by
dRc*dt=kf(Rc−Rc*)Gf−kbRc*dRg*dt=kf(Rg−Rg*)Gf−kbRg*,
(20)
where Rj, Rj*, and Gf (j ∈ {c,g}) indicate densities of the total receptors, guidance cue-bound receptors, and free guidance cues, respectively, and kf and kb indicate forward and backward reaction rates, respectively. The total guidance cue concentration is conserved as G=Gf+Rc*+Rg*. At steady state, Rj*=RjGf/(K+Gf), where K = kb/kf. If K ≫ Gf, Rj* can be approximated as (Rj/K)Gf, and the steady state of Rc* depending on G is then described by
Rc*=RcGK+Rc+Rg.
(21)
|
10.1371/journal.ppat.1006151 | Characterization of Early-Phase Neutrophil Extracellular Traps in Urinary Tract Infections | Neutrophils have an important role in the antimicrobial defense and resolution of urinary tract infections (UTIs). Our research suggests that a mechanism known as neutrophil extracellular trap (NET) formation is a defense strategy to combat pathogens that have invaded the urinary tract. A set of human urine specimens with very high neutrophil counts had microscopic evidence of cellular aggregation and lysis. Deoxyribonuclease I (DNase) treatment resulted in disaggregation of such structures, release of DNA fragments and a proteome enriched in histones and azurophilic granule effectors whose quantitative composition was similar to that of previously described in vitro-formed NETs. The effector proteins were further enriched in DNA-protein complexes isolated in native PAGE gels. Immunofluorescence microscopy revealed a flattened morphology of neutrophils associated with decondensed chromatin, remnants of granules in the cell periphery, and myeloperoxidase co-localized with extracellular DNA, features consistent with early-phase NETs. Nuclear staining revealed that a considerable fraction of bacterial cells in these structures were dead. The proteomes of two pathogens, Staphylococcus aureus and Escherichia coli, were indicative of adaptive responses to early-phase NETs, specifically the release of virulence factors and arrest of ribosomal protein synthesis. Finally, we discovered patterns of proteolysis consistent with widespread cleavage of proteins by neutrophil elastase, proteinase 3 and cathepsin G and evidence of citrullination in many nuclear proteins.
| Urinary tract infections (UTIs) are one of the world’s most widespread infectious diseases, with an estimated number of 150 million cases per year. Neutrophils play an important role in the defense of human patients against microbes causing UTIs. Molecules produced by neutrophils that migrate into the urinary tract can kill the invading microbes and resolve an infection, often without a need to treat patients with an antibiotic. Our work shows strong support for a mechanism called the formation of neutrophil extracellular traps (NETs), previously described for other infections and autoimmune conditions, which are involved in killing pathogens that have invaded the urinary tract. We show evidence of extracellular chromatin-containing structures using immunofluorescence microscopy and identified proteins that bind to the chromatin DNA and have functions to damage and kill bacterial cells or stop their growth.
| Urinary tract infections (UTIs) are common bacterial infections estimated to cause disease in 150 million patients globally per year [1]. They are classified as uncomplicated when they affect the lower urinary tract, are not linked to structural and functional urinary tract abnormalities or catheterization, and occur in immunocompetent hosts [2]. All other UTIs which frequently occur in hospitals and long-term care centers are classified as complicated and have increased risks of pyelonephritis and urosepsis [1,2]. Typical guidelines for UTI diagnosis include a minimum of 105 colony-forming units (cfus) per ml urine and symptoms such as high urgency or frequency of urination and abdominal or pelvic discomfort [2]. The most frequent cause of UTI is uropathogenic Escherichia coli (UPEC), which accounts for 74% of the cases among ambulatory care patients [2]. Pathogens have adapted to form biofilms on urethral catheters. Many, such as Klebsiella pneumoniae, Pseudomonas aeruginosa, Proteus mirabilis and Enterococcus species, are resistant to killing by normal sanitary measures and cause UTIs in the nosocomial environment [2,3]. The high rates of antibiotic prescription to treat UTIs and widespread acquisition of antibiotic resistance genes by the aforementioned pathogens have added to the concern of emerging bacterial strains that are no longer susceptible to any major class of antibiotic drugs [2].
Most UTIs are ascending infections that require the pathogen’s adherence to urothelial cells. Various types of fimbriae are critically important to the adherence of UPEC and P. mirabilis cells to urothelial surface receptors. These include CD11, CD44, uroplakins, and distinct glycosphingolipids, the latter of which activate urothelial cells via Toll-like receptor 4 (TLR4) [4,5]. When lipopolysaccharide (LPS)-binding proteins recognize bacterial LPS surface molecules, they also engage TLR4 and initiate signaling pathways that activate the transcription factor NF-kB. The result is the eventual secretion of chemokines and antimicrobial peptides [4,5]. Infected urothelial cells also express CXCR1, a surface receptor binding the chemokine CXCL8. CXCL8 and the interleukins IL-6 and IL-8 are considered important chemo-attractants for neutrophils and facilitate migration of the latter into the urothelial tissue [4,5]. Macrophages with distinct functionalities also recognize pathogens colonizing the urinary tract and promote the recruitment of neutrophils to the site of invasion [6]. Neutrophils are crucial effector cells for the innate immune response and the main determinants of the pathophysiology of pyuria [3,4]. There is some evidence that UPEC forms bacterial communities inside human urothelial cells, allowing evasion from neutrophil phagocytosis, a process that is potentially linked to recurrence of UTI [7].
Neutrophils are important as the first line of defense against invading pathogens. As recently reviewed [8], their antimicrobial strategies include phagocytosis, degranulation, and the formation of extracellular neutrophil traps (NETs). In the order listed, the strategies require increased amounts of time to respond to the microbial insults. Neutrophil granules have important functional roles in all three processes and are classified into azurophilic, specific, and gelatinase granules, each of which has distinct protein contents. The granules form sequentially during cell maturation, and the time-dependent release of their contents into phagosomes and the extracellular environment influences the inflammatory response overall, including the recruitment of other immune cells [9,10]. During the formation of NETs chromatin scaffolds are released as neutrophil nuclear and plasma membranes disintegrate [11]. The scaffolds are capable of containing the invading microbial cells while minimizing host cell damage. Immunofluorescence microscopy (IF) experiments have revealed that the DNA scaffold is dotted with histones, granular proteins such as myeloperoxidase (MPO) and elastase (NE), and the cytosolic calprotectin proteins S100-A8 and S100-A9 [12–14]. The proteins have direct antimicrobial and pro-inflammatory activities. The generation of reactive oxygen species (ROS) produced by NADPH oxidase is also implicated in the NET formation process and thus synergizes in the killing of the trapped pathogens. [12]. NE is important in early and intermediate stages of NET formation. At an early stage, H2O2 release triggers the dissociation of a complex in azurophilic granule membranes consisting of several effector proteins followed by the translocation of NE into the cytoplasm. Here, the enzyme binds to, and degrades, the F-actin cytoskeleton [15]. NE is freed to translocate into the nucleus where it degrades several core histones and triggers the decondensation of chromatin [13].
Pathogenic Streptococci and Staphylococcus aureus have evolved mechanisms to escape from NETs, in particular via the secretion of nucleases capable of degrading the extracellular DNA scaffold [16–18]. The formation of NETs has not been previously determined in the context of UTIs. Recently, we surveyed patient immune responses in more than 100 cases of UTI and asymptomatic bacteriuria via proteomics and found that urine sediment proteomes were rich in neutrophil contents [19,20]. Here, we examined such urine sediment samples further to assess the occurrence of necrosis and formation of NETs. Newly gained knowledge may lead to insights into the prevention of pathogen escape from neutrophil-associated defense strategies.
Leukocytes, in particular neutrophils, infiltrate the infected human urinary tract to eliminate invading pathogens. Proteomic analyses suggest that neutrophil proteins derived from urine sediments often exceed 50% of the total protein mass in UTI cases [19]. Cell-free urine fractions are also enriched in neutrophil-specific proteins in such cases [21]. This data indicates that phagocytosis is not the only strategy by which neutrophils combat uropathogens. Processes such as neutrophil degranulation, necrosis, and extracellular trap formation may play a role in the host defense. To examine whether any of these mechanisms are relevant in the context of UTIs, we examined 21 urinary pellet (UP) samples with large sediments, high leukocyte counts, and high neutrophil protein contents. Many of these UP samples revealed cellular aggregation prior to or after centrifugation at 1,500 x g and resuspension in PBS. From here on, we use the term aggregated UP (AUP) samples, most of which also had a viscous appearance. We refer to non-aggregated UP samples, which typically had lower leukocyte counts based on microscopy data, as easily dispersed UP (DUP) samples. Hematuria was not a factor in the extent to which the cellular aggregation phenotype was observed.
AUP lysates analyzed in SDS-PAGE gels showed a decreased abundance of uromodulin (UMOD), an 80 kDa protein highly abundant in the uninflamed UP proteome, and intense protein bands in the Mr range lower than 20 kDa (Fig 1). This is in contrast to reports on protein profiles of resting and activated neutrophil lysates where broad distributions of Mr values were observed [22,23] and DUP sample profiles. In the latter, UMOD was frequently detected as a dominant SDS-PAGE band. Some AUP protein profiles featured a high Mr band, in the 55–60 kDa range, that was identified as the heavy chain of MPO by LC-MS/MS (Fig 1), thus supporting the occurrence of urinary tract infiltration by neutrophils. While several proteins highly abundant in activated neutrophils, e.g. calprotectin, which consists of the proteins S100-A8 and S100-A9, cathelicidin, α-defensin 1 and lysozyme (LYZ), naturally have low Mr values, the enrichment of low Mr bands in SDS-PAGE gels for AUP samples was indicative of intense proteolysis. Proteolytic processes occur during necrosis of neutrophils and formation of NETs [15,24]. As shown in Fig 1, analysis by LC-MS/MS and/or urine culture confirmed that AUP samples and some DUP samples contained proteins expressed by pathogens. Detailed data are provided in S1 Data.
To assess neutrophil viability in freshly collected urine samples (stored for maximally six hours at 4°C), their sediments were stained with Trypan Blue and inspected microscopically. AUP samples featured many dead cells and a high level of cellular disintegration and aggregation (Fig 2). DUP sample #142 showed some cell debris, but more viable neutrophils. While the image resolution was modest, microscopic data for samples #151 and #157 were consistent with the morphological changes reported for neutrophils undergoing NETosis [24]. Further experiments at the molecular level were necessary to determine whether UTI-associated neutrophil fates followed a path of necrosis or NETosis. Given that clinical samples were limited in availability and quantity and many experiments were time-sensitive, distinct experiments were performed on subsets of samples in this study. The table presented in S2 Data provides a list of experiments performed with 21 AUP and DUP samples.
Similar to a method previously applied to characterize in vitro formation of NETs [14], we used incubations with DNase to determine whether extracellular DNA was present that could be degraded and afford the solubilization of proteins in AUP samples. In a previous report, neutrophil cultures stimulated to form NETs in vitro were shown to release abundant histones and azurophilic granule proteins, apparently bound to the DNA based on their alkaline pI values [14]. We expected the DNA digestion experiments, performed for 20 AUP and DUP samples, to discern NET-like structures from necrotic neutrophils. Proteins and DNA fragments were separated and visualized in gels, as shown in Fig 3 and S1 Data, following a sequential extraction process including PBS, PBS supplemented with 50 mM DTT, incubation with DNase, and detergent-mediated extraction and sonication to disintegrate residual cells, cytoskeletal and lipid membrane structures. In this order, the fractions UPsol1 to UPsol5 were generated. There was no measurable release of proteins after the DNase incubation from the DUP samples. In contrast, AUP samples released proteins into the fraction UPsol3. There was variation in the relative quantities of proteins in fraction UPsol3 compared to prior extraction steps, as shown for #112 and #122 in Fig 3. The release of DNA fragments following the incubation with DNase was a gradual process and stopped upon addition of the inhibitor Na-EDTA. While there was no visible release of DNA fragments prior to that incubation step in some cases (#122, Fig 3), other cases suggested that the AUP structures were fragile. Large DNA fragments were released from AUP sample #134 prior to the addition of DNase. Technical rather than biological reasons may account for this observation because several AUP samples were freeze-thawed prior to extraction (e.g., samples #20, #118, and #134). MPO and LTF, neutrophil granule effectors with high Mr values, were abundant in UPsol3 extracts in some cases (#112, Fig 3). The volume of insoluble matter retained after DNase incubation was markedly reduced for AUP samples, but not for DUP samples. Microscopic analysis of DNase-treated samples showed evidence of extensive cell degradation, but also of remaining intact neutrophils, thus supporting the notion that the DNase incubation itself did not lyse the cells.
Subsequent extraction steps (fractions UPsol4 and UPsol5) solubilized additional proteins. Calprotectin subunits (S100-A8, S100-A9) and α-defensin 1 were enriched in the low Mr range as shown in Fig 3. Why α-defensin 1 was so abundant in these extracts is unclear; the antimicrobial peptide is, however, known to bind to lipid membranes, which would explain the partitioning into a fraction with less soluble proteins. In summary, the extraction data provided preliminary evidence for structures consistent with the formation of NETs in AUP samples. Contributions from necrotic neutrophils to the release of DNA and proteins could not be ruled out based on the evidence that membranes in necrotic cells are likely permeable for enzymes such as DNase.
The degradation of DNA scaffolds from in vitro-generated NETs was previously shown to solubilize neutrophil effector proteins including histones [12–14]. We characterized UPsol3 fractions, in comparison with UPsol1 and UPsol4/5 fractions and in vitro-formed NETs, via shotgun proteomics. Ten AUP and five DUP samples were analyzed; data are shown for a subset of the analyses in Fig 4. All proteins found to be abundant in in vitro NETs [14] were identified, and often highly enriched, in the proteome of UPsol3 fractions derived from AUP samples. For the samples #94, #134, #33, #151, and #157 (Fig 4), the proteomic similarity to in vitro NETs was striking, especially when considering the fact that AUP samples also contain proteins soluble in urine and epithelial cell debris. Four histones, marked by the black line to the right of the stacked bar of in vitro (exp) NETs, and nine granule proteins located at the bottom in the stacked bar columns (Fig 4) contributed significantly to the overall UPsol3 proteomes. The data were consistent with an enzymatic process triggering protein release from extracellular chromatin. To demonstrate that the aggregation phenotype was not related to bacterial biofilms encased in a sticky extracellular matrix, the DUP sample #64 which was extracted from a urethral catheter biofilm was included here (Fig 4). Total protein release into the UPsol3 fraction was minimal. For sample #64, the twenty proteins dominating the proteomes of AUP samples and in vitro-formed NETs amounted to less than 5% to the total proteome of UPsol1 or UPsol4/5 fractions.
Proteomic differences comparing the UPsol3 and UPsol1 fractions (for a given AUP sample) support the presence of an extracellular DNA scaffold that often remains intact until the digestion step with DNase and that mostly binds histones, particularly H4A, H3A, H2A and H2B, the cytosolic proteins S100-A8 and S100-A9, and the granule proteins MPO, LTF, NE, cathepsin G (CTSG), and azurocidin (AZU). In cases such as AUP sample #20 where DNA is released prior to DNase treatment (Fig 5) and freeze-thaw cycles appeared to have degraded the DNA scaffold, the aforementioned proteins were present in higher quantities in the fractions with disintegrating DNA. Fractions UPsol4 and UPsol5 revealed higher abundances for neutrophil plasma and granule membrane proteins such as stomatin, subunits of cytochrome b-245 which contribute to the NADPH oxidase complex, and subunits integrin αM and β2 which form neutrophil surface receptors stimulating chemotaxis. Nucleophosmin was fairly abundant in some UPsol3 fractions, suggesting that nuclear proteins other than histones remain bound to DNA during the process of NETosis. A specific biomarker of NETs is citrullinated histone H3 [25]. We identified histone H3 peptide V24ArKSAPATGGVK36 with a deamidated R26 residue that is a known citrullinated site [26] and three citrullinated histone H1 peptides (Fig 6). One peptide, E55rSGVSLAALK65, contained a deamidated R56 residue that was reported to regulate chromatin binding [27]. Interestingly, citrullinated peptides were also identified for two histone-binding phosphoproteins, nucleophosmin and acidic leucine-rich nuclear phosphoprotein 32 family member A, and several cytoskeletal proteins. The peptides were often identified in AUP samples (#112, #118, and #157) with additional lines of evidence for early-phase NET formation. A total of 32 mass spectra were consistent with peptides that had citrullination sites (S4 Data). A frequently identified arginine deiminase in AUP samples was PADI2 suggesting that this enzyme catalyzed the citrullination reactions. To summarize, the results pertaining to the specimen susceptibility to digestion with DNase, the release of protein effectors and evidence of citrullinated histones encouraged the notion that NETs were formed in AUP samples.
To support the concept that proteins enriched in DNA-containing fractions bound to host genomic DNA in AUP extracts, we used non-denaturing PAGE analysis and evaluated gel regions simultaneously stained for DNA and protein. As shown in Fig 5, there was evidence of partially degraded DNA co-migrating with proteins, including fragile UP aggregates (UPsol1 fraction, sample #20) and more stable UP aggregates (UPsol3 fractions, samples #94, #151 and #157). The complexes migrated over a wide Mr range with the most intense DNA and protein staining in the 0.6–1.2 mDa range. Native PAGE often preserves macromolecular assemblies. For samples #151 and #157, more than 99% of the total protein extracted from the gel regions shown in Fig 5 consisted of antibacterial NET effector proteins. A protein unexpectedly identified by LC-MS/MS from the gel extracts of samples #20 and #94 was bikunin (BIK). BIK is a protease inhibitor with anti-inflammatory activity secreted into urine by tubular cells of the kidneys. The protein-DNA co-migration pattern was not observed for samples with low DNA content (UPsol1 fraction, #94; Fig 5). Well-resolved protein bands supported the notion that the observed complexes of protein and DNA did not result from unspecific protein retention in a high Mr range of the gels.
Intense staining for low Mr proteins in many AUP samples suggested proteolytic degradation. Indeed, the main neutrophil granule proteases, ELANE, CTSG, and proteinase 3 (PRTN3), were among the 30 most abundant proteins in UPsol3 fractions. To detect evidence of proteolysis in fractions of AUP samples occurred, we performed western blots for MPO, LTF, histone H4A, and ELANE (NE). The protein least affected by proteolysis was MPO (Fig 7A), which is consistent with data presented in Fig 4. Heavy and light chains of MPO were detected with the predicted Mr values in western blots. High peroxidase activities in the UPsol3 fractions derived from the samples #94 and #112 and the UPsol1 fraction derived from sample #20 using 3,3′,5,5′-tetramethylbenzidine as the substrate in colorimetric assays confirmed that MPO was enzymatically active. LTF and NE were visualized with bands corresponding to full-length and truncated versions. Western blot bands for histone H4A were only detected for sample #94 with a Mr value of approximately 9 kDa (Fig 7A) suggesting partial degradation of histones, as reported for in vitro-NETs [13]. The S. aureus protein A (Spa) was detected as a full-length protein in immunoblots (sample #112, red arrowheads in Fig 7A). Spa binds to the heavy chain Fc region of IgG at the bacterial surface and interferes with the opsonization and phagocytosis of S. aureus cells. The activity of Spa is apparently retained despite the use of denaturing methods and a host environment with high protease activities. In the experiments shown, Spa bound to the anti-rabbit IgG antibody conjugate. Spa was identified in the same UPsol fractions by LC-MS/MS. Immunoblots using purified Spa (0.1 μg) and lysates from two S. aureus strains (S1 Data) revealed bands that matched the Mr variants in the 40–50 kDa range. We hypothesize that, in infections with S. aureus where the immune response implicates NETosis, Spa binds to IgG and shields the bacterial surface from the binding of neutrophil effectors.
The activities of NE, PRTN3, and CTSG derived from in vitro-NETs were recently characterized by peptide profiling using synthetic peptide libraries [28]. With this information and the proteolytic events observed in western blots, we assessed LC-MS/MS data for the proteolysis in sequences of 41 proteins in AUP and DUP samples by NE or PRTN3. The range of peptide IDs in AUP samples, but not DUP samples, showed evidence of widespread proteolytic cleavage. The peptide site maps are provided in S5 Data. Large proteins such as MPO (Fig 7B), F-actin, actinin α-1 and gelsolin featured more NE/PRNT3 cleavage sites. Small proteins such as H4A (Fig 7B), H2B and profilin had fewer cleavage sites. Since the experimental approach included digestion with trypsin, an enzyme sharing the cleavage site C-terminal to K and R residues with CTSG, we could not detect CTSG-mediated proteolysis. All three proteases had peptide IDs consistent with their own cleavage by NE and PRTN3, potentially involving autolysis. Cytoskeletal proteins had a high number of cleavage sites. To verify that the probabilistic approaches used for peptide identification did not result in incorrect conclusions, we checked for peptide cleavage sites in precursor protein segments that are rapidly degraded to allow activation or targeted localization in the cell. Indeed, no peptides were detected for precursor regions including the N-terminal 163 amino acids of MPO. We identified the known PRNT3-specific cleavage site for cathelicidin (A136/L137) that results in the generation of the antimicrobial peptide LL-37 [29]. We identified the core sequence of LTF with antibacterial properties (F171-A201), termed kaolicin-1, although the peptide boundaries did not rule out trypsin cleavage. We conclude that protein degradation in AUP samples implicates major neutrophil proteases and is a widespread phenomenon.
We modified the technical approach to examine whether activities of NE, PRTN3, and CTSG are limited to the truncation of proteins to larger fragments rather than degradation to the level of oligopeptides. Analyzing peptide-enriched filtrates with a Mr cutoff of 10 kDa from six AUP samples without tryptic digestion, we found that up to 1,050 unique peptides with a length of 6 to 32 amino acids, assigned to 60 to 220 proteins, were identified. The frequency of small peptides derived from cytoskeletal proteins and histones, hemoglobin, likely derived from hematuria, and cytokeratin variants, likely derived from exfoliated urothelial cells, were particularly high in all AUP samples. The representation of neutrophil granule effector peptides was lower in peptidome datasets (S6 Data). Protein sequence maps with locations for the peptide IDs pertaining to the protein S100-A9, the cytoskeleton-associated protein profilin-1 and histone H2B are displayed in Fig 8A. The peptides clustered in distinct parts of the protein around a core peptide segment flanked by peptide overhangs varying in length. The data suggest that specific sites in the target protein are more susceptible to enzymatic cleavage than others. Greater than 98% of the peptide termini corresponded to the preferred P1 sites of CTSG (R, K, Y, and F) and of NE/PRTN3, again implicating these enzymes in such proteolytic events. The data are consistent with previous reports linking histone and actin degradation to NETosis [13,15]. We compared the peptide IDs derived from shotgun proteome and peptidome datasets for specific function- or localization-related protein groups as shown in Fig 8B. The analysis supported the notion that neutrophil granule proteins were less susceptible to degradation by CTSG, NE, and PRTN3 than other protein groups.
Peptidome data derived from DUP samples were also examined. Only a few peptides were identified from two DUP samples, while numerous peptides were identified from two other samples (#28 and #36). While the latter samples had peptide IDs consistent with lower neutrophil representation, histone and cytoskeletal protein-derived peptides were frequently observed. The findings suggest that the proteolysis in UP samples is not limited to those that have evidence of necrotic neutrophils or NETs. Furthermore, it highlights the possibility that a larger set of peptides is bio-activated in processes similar to those known for α-defensin and LL-37.
To obtain cellular evidence for the presence of necrotic neutrophils and/or NETs in AUP samples, immunofluorescence experiments were performed. The flattened, rounded morphology of enlarged neutrophils present in freshly collected AUP samples was presented (Fig 2). DAPI staining consistent with decondensed chromatin and the loss of nuclear lobules is visible in samples #146, #151, and #157 (open white arrows, Fig 9E, 9I and 9M). MPO staining reveals a decreased number and distribution of granules in contrast to those of intact neutrophils in sample #142 (closed white arrows, Fig 9C). The nuclear membrane border is less distinct with chromatin released into the cytoplasmic space (samples #151 and 157) and visibly expelled chromatin in the extracellular space (#157). MPO co-localizes with chromatin (open yellow arrows, Fig 9I, 9K, 9M and 9O). We did not observe stretches of laminar DNA associated with bona fide NETs. Together with the biochemical experiments, IF data for samples where cell fixation was performed right after collection are consistent with early-phase NET structures. In Fig 9M and 9N, S. aureus cells are visualized as small blue circles (chromatin, DAPI stain) and bright circles of the same size (phase contrast), respectively, suggesting extracellular entrapment.
Differentiating among neutrophil-based immune defense strategies that include degranulation, phagocytosis and NETs is difficult because the strategies largely rely on the same molecules that eventually kill invading pathogens or arrest their growth arrest. Neutrophil granule proteins and reactive oxygen species produced by MPO and NADPH oxidase play key roles. For several AUP samples we associate here with early-phase NET formation, we recovered few microbial colonies upon LB or BHI agar culture (#20, P. mirabilis; #33, E. faecalis; #122, C. albicans; #118 and #134, K. pneumoniae and E. coli). This suggests but does not prove killing by NETs. To assess bacterial viability in microscopy experiments, several samples were stained with a live/dead bacterial cell staining kit or SYTOX-green which stains only nuclei in damaged cells. The images for the samples #122, #146, and #157 (Fig 10) revealed more dead than living microbial cells, and the samples #122 and #157 had evidence of NET-like structures in which dead and living microbial cells were visible. In contrast, K. pneumoniae cells in the image for sample #151 were mostly viable and formed bacterial filaments that have previously been associated with subverting innate immune defenses by E. coli during UTI [30]. This data does not prove, but supports a contributing role of NETs in the defense against the uropathogens.
Analyzing the bacterial proteome from AUP samples (in vivo), compared to their growth to mid-exponential or stationary phase in culture (in vitro), provides information on the adaptation to the hostile host environment associated with the early-phase NETs and activated neutrophils in general. We analyzed the proteome from samples #94 (E. coli) and #112 (S. aureus). For AUP sample #112, the survey included the cell-free fractions (UPsol1/2 and UPsol3) and those that contained cells and cell debris (UPsol4/5). The cell-free fractions were markedly increased for S. aureus cell envelope proteins including the autolysin Atl and the transglycosylase IsaA, which made up more than 17% of the entire proteome (Fig 11). Other cell wall-localized proteins were also identified, including the aforementioned Spa protein, the SsaA antigen, proteins adhering to neutrophil, endothelial, and epithelial cell surfaces (Eap, Efb, IsdA, and IsdB), and transporters for metal ions and siderophores (IsdB, SirA, Cnt, and ZnuA). Two secreted virulence factors, the complement system inhibitor SCIN and hemolysin-γ, were also identified. Neutrophil protease activities and cell envelope damage may have directly triggered the release of these proteins from the bacterial cell. The autolysin Atl is involved in cell division, suggesting that its removal from the bacterial cell may have resulted in the arrest of cell division and growth. In fraction UPsol4/5, the contribution of cell envelope proteins to the proteome was lower, but low ribosome and translation-associated protein contents suggested markedly decreased ribosomal protein synthesis activity compared to in vitro grown S. aureus cells (Fig 11). This observation is consistent with the need to repair and regenerate cellular structures in bacteria facing neutrophil attack and the arrest of cell growth. Similarly decreased ribosomal protein synthesis was observed for E. coli in the sample #94.
Our analysis of AUP samples resulted in molecular and cellular support for the formation of early-phase NETs in the context of UTIs. Using specimens subjected to freeze-thaw cycles prior to experiments characterizing neutrophil defense strategies, it was evident that the structures in most AUP samples lost their aggregated consistency and were susceptible to disintegration. The fragility of NETs in the context of various pathologies was described [12,31,32]. Phase contrast microscopy for a set of freshly collected samples revealed the presence of aggregates of mostly non-viable and lysed neutrophils consistent with necrosis or NETosis. Visualization by IF using DAPI staining showed that some but not all neutrophils had a flattened, rounded morphology. Chromatin filled the entire cellular space with no distinction between euchromatin and heterochromatin. These feature are consistent with necrotic neutrophils [24] and early-phase NETs [12,32]. Visualization by IF with an MPO antibody supported the notion that NET-like structures had formed. In clusters of disintegrating neutrophils, extracellular chromatin streaks that co-localized with MPO were visible, especially for sample #157 (Fig 9). Together, the morphological and DNA/protein co-localization traits are more in line with early-phase NETs than necrotic neutrophils [12,24]. The MPO staining pattern was different in samples with many viable neutrophils (#146, well distributed granules) compared to those dominated by dead and lysing neutrophils (#146, #151, #157; Fig 9). These samples revealed remnants of granules in the cell periphery and co-localized staining with chromatin. The caveat of the interpretations is that the images were not confocal, thus not permitting visualization in confocal slices. Unambiguous co-localization would require confocal IF microscopy. The data justify the term NET-like structures. More mature stages of NETs (bona fide NETs) implicate the spread of chromatin from the cell of origin with elongated spikes protruding into the extracellular space and bridge formation to other neutrophils [12,13,32]. These were not detected for freshly prepared AUP samples.
A set of molecular analyses, with emphasis on proteomics, was performed to corroborate the evidence of NETs as a defense strategy in UTIs. To do this, we benefitted from publications on the protein composition of NETs generated in vitro [14] and on mechanisms of NET formation [13,15]: NETosis requires cytoskeletal degradation and nuclear decondensation, enabled by proteolytic pathways that result in the disappearance of cellular membranes and eventual expulsion of chromatin and proteins into the extracellular milieu. First, we observed that DNase effectively homogenized AUP samples. Second, DNase treatment released proteins consistent with the existence of a DNA-protein network not protected by intact cellular membranes. Third, comparative analysis of the proteome of DNase-digested fractions from AUP samples with that of in vitro-NETs [14] revealed strikingly similar profiles. Such profiles were different in other fractions and in DUP samples. Forth, 0.6 to 1.2 mDa DNA-protein complexes isolated from partially DNase-digested fractions via native PAGE had a composition highly enriched in proteins that are attributed to have antibacterial effector functions in NETs (histones, MPO, NE, AZU and LTF). This data supported the physical binding of such proteins to DNA expelled from lysing neutrophils.
In further support of early-phase NETs, we identified citrullinated peptides for histones H1 and H3. The site in the histone H3 peptide (R26) matched one of the four known sites of citrullination sites for this protein [26]. Citrullinated peptides were also detected in other nuclear and several cytoskeleton-associated proteins raising the question as to whether this modification impacts protein function and degradation in NETs. A study on rheumatoid arthritis determined that NETs are a source of various citrullinated auto-antigens that stimulate inflammatory responses [33]. Considering the abundance of the enzyme PADI2 in AUP samples, apparently responsible for the deamidation, and several highly active proteases, we hypothesize that citrullination modulates protein binding to DNA (loss of a positive charge) and alters the susceptibility to cleavage by CTSG. A cleavage site of CTSG is the C-terminus of arginine.
Although massive proteolysis was not unique to AUP samples analyzed here and is known to occur in necrotic neutrophils, the observation of extensive protein degradation, particularly for known substrates in NETs [13,15], supported the presence of NETs in urine sediments associated with UTI. The role of the azurosome including NE in the degradation of cytoskeletal proteins, specifically F-actin, was reported [15]. While the proteolysis of cytoskeletal proteins was extensive in our analyzed AUP samples, neutrophil granule proteins were less frequently degraded to small peptides. We hypothesize that, like citrullination, proteolysis contributes to NETosis and the formation of new antimicrobial peptides, perhaps not limited to those emerging from histones, cathelicidin and α-defensin-1. Consistent with published data [28], our data support the notion that this proteolysis is driven by three neutrophil proteases, NE, PRTN3 and CTSG. Their autolytic activities may constitute another level of control over pro- and anti-inflammatory activities in the process of UTI resolution. Likewise, proteolysis of chemokines by enzymes released into NETs was previously linked to a decrease in inflammation in gout [34]. Another interesting observation was the high quantity of insoluble α-defensin-1 retained after DNase digestion of the AUP samples (UPsol4/5, Fig 4). It suggests that most α-defensin-1, unlike other neutrophil granule effectors, is not bound to chromatin but to lipid structures also present in early-phase NETs. Previously, α-defensins were reported to be released from necrotic and apoptotic neutrophils and to have a potent anti-inflammatory role [35]. It is of interest to elucidate the role of α-defensin-1 in inflammatory signaling pertaining to early-phase NETs associated with UTI.
To our knowledge, we characterized proteome-wide pathogen responses to potential entrapment in NETs for the first time. Analyzing 13 AUP samples, we identified seven different pathogens: E. faecalis, Serratia marcescens, P. mirabilis, C. albicans, K. pneumoniae, E. coli, and S. aureus. In some instances, a sufficient number of bacterial proteins were surveyed in UP extracts (in vivo) to allow comparative analysis with the proteome from lab cultures of the isolates (in vitro). Analyses performed for E. coli (sample #94) and S. aureus (sample #112) revealed strong decreases in proteins with functions in ribosomal translation and protein biosynthesis for in vivo bacteria while proteins with structural and functional roles in the cell enveloped were increased in vivo. These adaptations are consistent with defensive responses of the pathogens to perturbed integrity of their cell envelopes, which are under attack by a diversity of neutrophil effectors. A low level of ribosomal protein synthesis has been linked to the bacterial persistence state [36], thus offering a link of persister formation to survival in NETs. Ribosomal protein synthesis is a process consuming a lot of energy. Cellular energy sources such as ATP may instead be available to repair and regenerate damaged cell envelopes. Both E. coli and S. aureus cells produced ATP synthase in vivo and in vitro. The S. aureus proteome showed evidence of increased production of adhesins (e.g., Eap, Efb, and Spa), iron/siderophore transporters (e.g., IsdA and IsdB), proteins perturbing the function of the immune system (e.g., Spa and SCIN), and human cell toxins such as γ-hemolysin. The functional roles of these virulence factors are known [37] [38,39]. Finally, we observed two highly abundant proteins linked to cell wall autolysis (Atl and IsaA) [40,41] in the S. aureus proteome derived from the DNase-extracted AUP fraction, suggesting that their release modulates cell division and cell wall properties. The process may allow the bacteria to become more responsive to the starvation of nutrients (e.g., iron/siderophores) and improve the likelihood of immune evasion. Fluorescence-based viability assays indicated that some bacterial cells exposed to the structures we attribute to early-phase NETs survived while others were dead. We did not identify nucleases in the bacterial proteome of sample #112. Endonucleases such as EndA, produced by S. pneumoniae [16], Sda1, produced by S. pyogenes [17], and Nuc, produced by S. aureus [18], were shown to degrade chromatin and trigger bacterial escape from the entrapment in NETs. In the future, we intend to characterize the host-pathogen interactions in early-phase NETs associated with UTI in more depth. Among the most intriguing questions are proteolytic pathways that may generate novel bioactive peptides analogous to LL-37 and the strategies pathogens use to survive entrapment in NET structures.
The human urine specimens analyzed in this study were either 1) de-identified samples collected for diagnostic purposes that otherwise would have been discarded, thus exempt from IRB review or 2) de-identified samples that were retained from an earlier IRB approved study and for which consent for future research purposes was given. Diagnostic specimens were obtained from the Pathology and Clinical Microbiology Laboratory of the Shady Grove Adventist Hospital (SGAH) in Rockville, Maryland.
The specimens were kept at 4°C for up to six hours after collection and centrifuged at 800 × g for 15 min to recover the cellular sediment. PBS was added in a 10- to 20-fold volume to the sediment and gently resuspended followed by another centrifugation step in a 1.5 ml tube for 5 min. Aliquots of this urinary pellet (UP) sample, resuspended in TBS (pH 7.5), were incubated with the Live/Dead-BacLight bacterial viability staining kit (#L7007; ThermoFisher Scientific) at SYTO9 and propidium iodide final concentrations of 5 and 55 μM, respectively or with the fluorescent dye SYTOX-Green (#L7020; ThermoFisher Scientific) for 15 min at 20°C in the dark to prepare the samples for fluorescence microscopy. Unbound dyes were removed by a 3 min centrifugation step, and pellets were gently suspended in TBS. These samples, as well as untreated UP samples suspended in PBS, were subjected to fixation adding paraformaldehyde in a 4% final concentration on a glass slide and incubating for 15 min at 20°C in the dark. Slides were rinsed with water, air-dried and stored at 4°C in the dark until used for IF staining and microscopy.
Slides were viewed with an inverted Zeiss Axiovision microscope to ensure that cells and extracellular materials were immobilized. Blocking solution (2.2% BSA and 0.1% Tween-20 in PBS, 0.01% sodium azide) was added, incubating at 20°C for 75 min. The solution was replaced with a 1:25 dilution (~ 4 μg/ml) of the anti-MPO polyclonal antibody (#16128-R; Sta Cruz Biotech) in Ab diluent (PBS containing 1% BSA and 0.1% Tween-20) and incubated for 2 h at 20°C. The slide was washed five times with PBS containing 1% BSA and 0.1% Tween-20 for 3–5 min, tilting the slide occasionally to enhance exposure of fixed cells to the wash solution. Incubation with a 1:100 dilution (~ 1 μg/ml) of the goat anti-rabbit IgG antibody conjugated to the fluorescent dye CFL555 (#362271; Sta Cruz Biotech) in Ab diluent at 20°C for 75 min in the dark followed. The wash steps were repeated. Adding a drop of Vectashield Antifade Mounting Medium with DAPI (# H-1200; Vector Laboratories), the slide was incubated at 20°C for 2 h in the dark, sealed with a coverslip to stain nuclei and preserve fixed cells, and stored at 4°C in the dark until oil immersion IF microscopy was performed.
Aliquots of PBS-washed UP samples, some of which underwent free-thaw cycles prior to extraction, were spun at 4000 × g for 8 min to recover the UPsol1 fraction. Eight UP samples were available for extractions without prior storage in the freezer at -80°C. The centrifugation conditions were also used to recover solubilized materials from subsequent extraction steps. The pellet was gently re-agitated in a 5-fold volume of PBS or TBS containing 50 mM DTT and left at 20°C for 10 min. The UPsol2 fraction was recovered. In some cases, this extraction step was repeated. The residual pellet was incubated with a 5- to 10-fold volume of PBS or TBS containing 2 μL bovine DNAse I (#D-4527; Sigma-Aldrich) in the presence of 5 mM MgCl2. One μl DNAse I corresponded to an activity of 10 Kunitz units. The suspension was incubated at 20°C for 45 min, gently agitating the tube occasionally, and in a shaker with 880 rpm at 37°C for 30 min. Upon centrifugation, the supernatant was termed the fraction UPsol3. The final extraction steps were performed to lyse and extract microbial cells and host cell debris. Samples containing S. aureus cells were incubated with 20 μg/ml lysostaphin in TBS (pH 8) using a shaker with 880 rpm at 37°C for 20 min. Other samples were incubated with a combination of 50 μg/ml mutanolysin and 100 μg/ml lysosome accordingly. Following addition of 5 mM EDTA and 0.4% m/V CHAPS (final concentrations), such samples were vortexed, left at 20°C for 10 min, and sonicated at the amplitude 6 in ten 30 sec on/off cycles using a Misonex 3000 sonicator ice bath. The supernatant (fraction UPsol4) was isolated. To the residual pellet, the SED solution containing 1% SDS, 0.3% Tween-20, 10 mM Na-EDTA, and 25 mM DTT was added. The homogenate was vortexed, left at 20°C for 10 min, and heat-denatured at 95°C for 3 min. This was followed by the aforementioned sonication step and centrifugation at 16,100 × g for 10 min to isolate the fraction UPsol5. The fractions were used for various molecular analysis procedures. Aliquots of unprocessed UP samples were also incubated with the SED solution at 20°C for 10 min and heat-denatured at 95°C for 3 min. Sonication and centrifugal isolation steps to recover the supernatants of these UP lysates were done as described for the fractions UPsol4 and UPsol5.
Lysates and fractions derived from UP samples were subjected to SDS-PAGE in 4–12% acrylamide gradient gels and stained with the protein dye Coomassie Brilliant Blue G-250 (CBB). The gel staining intensity was used to estimate the overall protein amount in a given sample for the processing steps of filter-aided sample preparation (FASP), comparing the staining intensity with 2 μg BSA. For FASP analysis, where the intent is to separate and concentrate a fraction of denatured proteins larger than approximately 5 kDa from low Mr molecules including peptides, a membrane concentrator device (Sartorius, Germany) with a 10 kDa Mr cut-off was used. Using trypsin as the enzyme to digest a UP lysate or UPsol fraction, we applied a published experimental approach [42,43]. For in-gel digestion of proteins, SDS-PAGE gel bands were excised and digested with trypsin as also previously reported [44]. Digestion products derived from FASP were desalted using the spinnable StageTip protocol [45]. Eluates containing enriched peptide mixtures were applied to LC-MS/MS. For the analysis of peptides contained in the small molecule fraction of UP samples, the combined fractions of UPsol1-UPsol3 were applied to the 10 kDa concentrator device recovering the filtrate by spinning at 16,100 × g. The filtrate was also subjected to the StageTip protocol to enrich for peptides [45].
Peptide mixtures were analyzed using an LC-MS/MS platform consisting of the Ultimate 3000-nano LC coupled via a FLEX nano-electrospray ion source to a Q-Exactive mass spectrometer (Thermo Scientific, USA). We have described the LC-MS/MS experimental and data acquisition methods in detail previously [19,43]. Parallel to an LC solvent gradient elution time of 110 min, peptide ions were analyzed in a MS1 data-dependent mode to select ions for the MS2 scans using the XCalibur software v2.2 (Thermo Scientific). Survey scans were acquired at a resolution of 70,000 (m/Δm) over a mass range of m/z 250–1,800. In each cycle, the ten most intense ions were subjected to fragmentation applying normalized collision energy of 27%. The MS2 scans were performed at a resolution of 17,500. Ions that were unassigned or had a charge of +1 were rejected from further analysis. Two or three replicate LC-MS/MS experiments were run, and raw MS files were combined for database searches. The database was either a version of the human UniProtKB database (release 2013_6) reduced in protein redundancy (27,151 entries) or the former complemented by protein sequence entries of genomes representing the microbial species identified in the investigated UP samples. The microbial database identifiers were reported previously [19]. This set of methods was applied to analyze gel extract, proteome, and peptidome samples. The software tool Proteome Discoverer v1.4 (Thermo Scientific) was used to interpret data qualitatively and confirm mass spectral qualities of individual peptides, using search parameters for mass tolerance, proteolytic cleavage sites for trypsin, and amino acid modifications as described [43]. To evaluate proteolytic cleavages by the enzymes NE, PRTN3, and CTSG, the parameters were modified selecting ‘elastase’ (C-terminal to A, V, I, L, M, S, C, T as reported in [28]) in addition to ‘trypsin’. For peptidome analyses, cleavage at any position in the amino acid sequence was allowed. The false discovery rate (FDR) was set at 1% for protein IDs.
Raw MS raw files were imported into the MaxQuant software suite (v1.4.2) [46] accepting the default settings for quantification via MS1 peak integration and normalization of proteomic data comparing multiple samples (S2 Data). We used the intensity-based absolute quantification (iBAQ) function enabled in MaxQuant to estimate protein abundances in all of the analyzed samples [47]. Processing data in Excel, the iBAQ quantity of a protein in a given sample was divided by the sum of all iBAQ quantities. This method of estimating relative iBAQ values was applied to human and microbial proteome datasets. One analysis pertained to the comparison of the human proteome present in UPsol fractions and in vitro-NETs). Another analysis pertained to the S. aureus proteome in UPsol fractions versus lysates from in vitro cultures. Information on protein functional role categories, their subcellular localizations and post-translational modifications reflected information provided for protein entries in the UniProt database [48].
Native-PAGE gels (Life Technologies, Carlsbad, CA) with a 3% acrylamide concentration were used to examine the co-migration of proteins and DNA molecules. We used the experimental procedures recommended by the product manufacturer. Gels were stained with CBB and ethidium bromide to visualize proteins and DNA, respectively. SDS-PAGE in 4–12% acrylamide gradient gels was performed loading 5 to 20 μg total protein from several UPsol fractions. The gels were blotted onto PVDF membranes and incubated with antibodies, washed, and developed using chemiluminescence, as previously described [49]. The following polyclonal antibodies raised against peptide segments (Santa Cruz Biotechnology, Dallas, Texas) were used: anti-MPO heavy chain C-16 (sc-16128-R), anti-LTF (H-65, sc-25622), anti-ELANE (H-57, sc-25621), anti-histone H4 (H-97, sc-10810), and a horseradish peroxidase conjugate of goat anti-rabbit IgG-HRP (sc-2004). Primary antibodies and secondary antibody conjugates were diluted 1:1,000 and 1:10,000, respectively, for 90 min incubation steps. The peroxidase activity of MPO was measured with the substrate 3,3’,5,5’-Tetramethylbenzidine (KPL, Gaithersburg, MD; #54-11-50) in a colorimetric endpoint assay over 4 min measuring the absorption of the product at 650 nm.
Two isolates, a S. aureus strain from UP sample #112 and an E. coli strain from UP sample #134, were recovered from LB agar plates grown for 12–24 h aerobically at 37°C. After confirming species identities by Gram staining and LC-MS/MS analysis and an overnight pre-culture, 20–25 ml fresh LB media were inoculated at an OD600 of ca. 0.05 for suspension cultures in a shaker at 880 rpm set at 37°C. The cultures were stopped when the OD600 values reached 0.35–0.5 (mid-exponential phase; ~4–6 h) and 0.9–1.2 (stationary phase; ~14 h). Bacterial cells were lyzed by addition of the SED solution, letting the suspension sit for 10 min vortexing a few times, applying heat at 95°C for 3 min, and sonication as described for the UPsol5 extraction step. For S. aureus cell lysis, the initial incubation step was cell wall digestion with lysostaphin for 20 min followed by the cell lysis protocol. The S. aureus strain HIP5827 used for experiments to detect the cell surface protein Spa was described earlier [50].
|
10.1371/journal.pmed.1002026 | Risks and Population Burden of Cardiovascular Diseases Associated with Diabetes in China: A Prospective Study of 0.5 Million Adults | In China, diabetes prevalence is rising rapidly, but little is known about the associated risks and population burden of cardiovascular diseases. We assess associations of diabetes with major cardiovascular diseases and the relevance of diabetes duration and other modifiable risk factors to these associations.
A nationwide prospective study recruited 512,891 men and women aged 30–79 y between 25 June 2004 and 15 July 2008 from ten diverse localities across China. During ~7 y of follow-up, 7,353 cardiovascular deaths and 25,451 non-fatal major cardiovascular events were recorded among 488,760 participants without prior cardiovascular disease at baseline. Cox regression yielded adjusted hazard ratios (HRs) comparing disease risks in individuals with diabetes to those without. Overall, 5.4% (n = 26,335) of participants had self-reported (2.7%) or screen-detected (2.7%) diabetes. Individuals with self-reported diabetes had an adjusted HR of 2.07 (95% CI 1.90–2.26) for cardiovascular mortality. There were significant excess risks of major coronary event (2.44, 95% CI 2.18–2.73), ischaemic stroke (1.68, 95% CI 1.60–1.77), and intracerebral haemorrhage (1.24, 95% CI 1.07–1.44). Screen-detected diabetes was also associated with significant, though more modest, excess cardiovascular risks, with corresponding HRs of 1.66 (95% CI 1.51–1.83), 1.62 (95% CI 1.40–1.86), 1.48 (95% CI 1.40–1.57), and 1.17 (95% CI 1.01–1.36), respectively. Misclassification of screen-detected diabetes may have caused these risk estimates to be underestimated, whilst lack of data on lipids may have resulted in residual confounding of diabetes-associated cardiovascular disease risks. Among individuals with diabetes, cardiovascular risk increased progressively with duration of diabetes and number of other presenting modifiable cardiovascular risk factors. Assuming a causal association, diabetes now accounts for ~0.5 million (489,676, 95% CI 335,777–681,202) cardiovascular deaths annually in China.
Among Chinese adults, diabetes is associated with significantly increased risks of major cardiovascular diseases. The increasing prevalence and younger age of onset of diabetes foreshadow greater diabetes-attributable disease burden in China.
| In China, approximately one in ten adults now has diabetes, following large and rapid increases in diabetes prevalence over recent decades.
In Western populations, individuals with diabetes are at higher risk of developing cardiovascular diseases, but relevant evidence is lacking in China, where patterns of cardiovascular disease (e.g., high stroke rates) differ from those in Western populations.
Even in Western populations, questions remain as to whether diabetes increases the risk of haemorrhagic stroke and how other modifiable risk factors (e.g., smoking, hypertension, overweight or obese) may act together with diabetes to further increase cardiovascular disease risk.
We recruited 0.5 million Chinese adults to our study in 2004–2008; participants completed a detailed questionnaire interview and underwent physical measurements and blood tests, and we then tracked their health for seven years.
Individuals with diabetes (previously or newly diagnosed) at the initial baseline assessment had 1.5- to 2.5-fold higher risks of developing ischaemic heart disease and stroke (both ischaemic and haemorrhagic).
Among individuals with diabetes, cardiovascular disease risks increased progressively with the number of other modifiable cardiovascular disease risk factors present (e.g., smoking, overweight or obese, hypertension, and low physical activity) and with a longer duration of diabetes diagnosis.
These findings show that Chinese adults with diabetes have a substantially higher risk of cardiovascular diseases than those without diabetes.
Based on these findings, we estimate that diabetes now accounts for almost 0.5 million cardiovascular disease deaths each year in China; this disease burden will likely increase as diabetes is expected to become more prevalent in China over coming decades, especially among young adults.
To help to reduce the risk of cardiovascular diseases, individuals with diabetes should not only take necessary medications, but also improve their lifestyles, e.g., through smoking cessation, weight loss, and increased physical activity.
| Worldwide over 400 million people have diabetes, and the prevalence is increasing rapidly in both developed and developing countries [1]. Previous studies of mostly Western populations have shown that diabetes is typically associated with a 2-fold increased risk of ischaemic heart disease (IHD) [2,3]. Uncertainty remains, however, about whether similar excess risk applies to other populations, and about the strength of the association of diabetes with stroke and, particularly, stroke subtypes [2–4]. Appropriate understanding of these issues is of considerable relevance to China, where stroke rates are high [5].
Since the 1980s, there has been a rapid and substantial increase in the prevalence of diabetes in China [1,6,7], which now affects ~10% of adults [8]. Compared with those in Western populations, individuals with diabetes in China have tended to be leaner and to have worse pancreatic beta-cell function, which may result in greater susceptibility to microvascular complications and cancer than to macrovascular complications [9]. Despite the growing diabetes epidemic, there is limited evidence about the association of diabetes with cardiovascular disease in China [10–14], with previous studies limited by potentially outdated risk estimates [12], highly selected study populations [11], relatively small size [10–12], and lack of proper investigation of the relevance of diabetes duration and other modifiable factors, such as smoking, adiposity, blood pressure, and physical activity, which frequently differ between China and Western populations.
We report relevant findings from a large prospective study, the China Kadoorie Biobank (CKB), of 0.5 million adults, recruited between 25 June 2004 and 15 July 2008 across ten diverse areas of China. The aim of the present study is to examine the association of diabetes—both self-reported and screen-detected—with the risks of major cardiovascular diseases, including IHD and stroke subtypes. By applying population-attributable fractions for cardiovascular mortality from the present study to national cause-specific mortality and representative diabetes prevalence data, we also estimate the number of cardiovascular deaths attributable to diabetes in mainland China.
All participants provided written consent prior to participation, including permission for follow-up. Ethics approval was obtained from Oxford University, the Chinese Centre for Disease Control and Prevention (CDC), and the ten study areas’ local CDCs.
Details of the CKB design, methods, and population have been reported previously [15,16]. Briefly, the 2004–2008 baseline survey took place in ten localities across China (five urban and five rural; S1 Fig) chosen to provide diversity in exposure and disease patterns, additionally taking account of logistical considerations, including population stability, death and disease registry quality, and local capacity. All residents aged 35–74 y from 100–150 administrative units (rural villages or urban residential committees) in each area were invited to attend study assessment clinics. Overall, ~30% responded [15], and 512,891 individuals were enrolled (including a few slightly outside the 35–74 y range).
At study assessment clinics, trained health workers administered laptop-based questionnaires that collected data on socio-demographic status, smoking, alcohol consumption, diet, physical activity (related to leisure, household work, occupation, and commuting [17]), medical history, and—among participants reporting a history of diabetes, IHD, stroke/transient ischaemic attack, and hypertension—current use of medications. Information on age at diagnosis was collected from participants reporting a history of diabetes. Health workers measured height, weight, waist and hip circumference, and blood pressure and took non-fasting venous blood samples (with the time since last food recorded) for storage and immediate on-site testing of plasma glucose level using a SureStep Plus meter (LifeScan). Participants without self-reported diabetes with a plasma glucose level of 7.8–11.0 mmol/l were invited to undergo fasting glucose testing (using the same measurement technique) the following day. Every 4–5 y, a 5%–6% random sample of surviving participants was resurveyed, collating the same data as at baseline, with certain additions.
Participants answering “yes” to the question, “Has a doctor ever told you that you had diabetes?” at baseline were defined as having self-reported diabetes; among them, information about age at diagnosis and current medication use was collected. Screen-detected diabetes was defined as not having self-reported diabetes but having a random plasma glucose (RPG) level ≥ 7.0 mmol/l if time since last food ≥ 8 h, RPG level ≥ 11.1 mmol/l if time since last food < 8 h, or fasting plasma glucose level ≥ 7.0 mmol/l on subsequent testing [18]. Participants with screen-detected diabetes were provided with a referral letter and advised to seek formal medical consultation.
The vital status of each participant was obtained periodically from China CDC’s Disease Surveillance Points, checked annually against local residential and health insurance records and by active confirmation through street committee or village administrators. Causes of death from official death certificates were ICD-10 coded by trained Disease Surveillance Point staff blinded to baseline information. Information on non-fatal outcomes was collected through linkage with established disease surveillance systems (for cancer, IHD, stroke, and diabetes) and, via unique national ID, with the national health insurance system, which records details of ICD-10-coded hospitalisations in all study areas [16].
By 1 January 2014, 25,488 (5.0%) participants had died, and 2,411 (0.5%) were lost to follow-up. For the present study, the primary endpoints were cardiovascular death (ICD-10 I00–I25, I27–I88, I95–I99), myocardial infarction (MI) (ICD-10 I21–I23), major coronary event (MCE) (non-fatal MI or fatal IHD; ICD-10 I20–I25), ischaemic stroke (IS) (ICD-10 I63), intracerebral haemorrhage (ICH) (ICD-10 I61), total stroke (ICD-10 I60, I61, I63, I64), and major occlusive vascular disease (MOVD) (IS or MCE).
The main analyses excluded participants reporting prior doctor-diagnosed IHD (n = 15,472, 3.0%) or stroke/transient ischaemic attack (n = 8,884, 1.7%) at baseline. A further 1,081 (0.2%) participants with missing, implausible, or extreme values for blood pressure, height, waist circumference, hip circumference, waist-to-hip ratio, or body mass index (BMI) were excluded, leaving 488,760 individuals (199,896 men, 288,864 women) for the present analyses.
Analyses were done separately for self-reported and screen-detected diabetes, with a common reference group consisting of individuals without self-reported or screen-detected diabetes. The mean values and prevalence of certain variables were calculated by diabetes status, standardised by 5-y age group, sex, and study area. Direct standardisation was used to calculate age-, sex-, and study-area-adjusted disease incidence rates, using the total study population as the standard. Cox regression yielded hazard ratios (HRs) for diabetes versus not, stratified by age at risk (5-y age groups), study area, and, where appropriate, sex, and adjusted for education (no formal education, primary school, middle school, high school, college/university), smoking (never, occasional, ex-regular, current regular), alcohol consumption (never, occasional, ex-regular, reduced, weekly), systolic blood pressure (SBP) (nine groups), and physical activity (five groups). Sensitivity analyses were performed adjusting for age, SBP, and physical activity as continuous variables.
Comparison of HRs for the first four and subsequent years of follow-up revealed no evidence of departure from the proportional hazards assumption. Adjusted HRs were calculated across strata of other cardiovascular risk factors and duration of diabetes at baseline, and chi-square tests for trend and heterogeneity (i.e., effect modification or statistical interaction) were applied to the log HRs and their standard errors [19]. In separate analyses, risk estimates were also adjusted for adiposity. Finally, the HR for MOVD associated with the number of other presenting baseline cardiovascular risk factors (hypertension, overweight or obese, ever regular smoking, physical inactivity) among individuals with diabetes was assessed using the floating absolute risk method [20], which provides estimates of variance across all exposure categories.
Population-attributable risk is P(HR − 1)/(1 + P[HR − 1]), where P is the prevalence of diabetes. By applying age-specific HRs (which can be assumed to approximate the relative risk [21]) to nationally representative, age-specific prevalence of diabetes [8] and national cause-specific mortality data [22], we estimated the number of cardiovascular deaths attributable to diabetes. All analyses used SAS version 9.3.
Among 488,760 participants without prior cardiovascular disease, the mean baseline age was 51 (standard deviation 11) y, 2.7% (n = 13,284) reported a history of doctor-diagnosed diabetes (self-reported diabetes), and a further 2.7% (n = 13,051) had screen-detected diabetes, of whom 8,554 (65.5%) were identified through RPG measurement (RPG ≥ 7.0 mmol/l and time since last food ≥ 8 h: n = 3,986; RPG ≥ 11.1 mmol/l and time since last food < 8 h: n = 4,568), 1,447 (11.1%) through fasting plasma glucose measurement, and 3,050 (23.4%) through both. For both self-reported and screen-detected diabetes, the prevalence was slightly higher in women than in men (2.9% versus 2.5% for self-reported diabetes and 2.8% versus 2.6% for screen-detected diabetes), mainly at ages > 50 y (S2 Fig), and the prevalence was higher in urban than in rural areas (3.9% versus 1.8% for self-reported diabetes and 3.4% versus 2.1% for screen-detected diabetes). Participants reporting a history of diabetes were older and less likely to be current smokers or alcohol drinkers, but more likely to be ex-smokers or ex-drinkers and to be less physically active than those without diabetes, and had an approximately 4-fold greater prevalence of a family history of diabetes (Table 1). Those with screen-detected diabetes were older than and about twice as likely to have a family history of diabetes as those without diabetes, but had comparable smoking and alcohol consumption patterns. The prevalence of overweight or obesity (BMI ≥ 25.0 kg/m2) and hypertension (SBP ≥ 140 mm Hg) were higher in participants with diabetes, particularly screen-detected diabetes. Among participants with self-reported diabetes, the median age at diagnosis was 53 (range 4 to 77) y, and the median duration of diabetes at baseline was 4 (range 0 to 56) y, with 76% (n = 8,501) reporting use of insulin or oral hypoglycaemic agents.
During ~3.4 million person-years (mean 7 y per person) of follow-up, 7,353 (1.5%) participants died from cardiovascular disease. Individuals with self-reported diabetes had a 2-fold increased hazard of cardiovascular mortality (HR 2.07, 95% CI 1.90–2.26) (Table 2), and the HR for cardiovascular mortality was greater in women than in men (2.37, 95% CI 2.11–2.65, versus 1.73, 95% CI 1.50–2.00) (S3 Fig). Likewise, there were significantly increased risks of MOVD (1.77, 1.69–1.86), MCE (2.44, 2.18–2.73), and IS (1.68, 1.60–1.77) among those with self-reported diabetes (Table 2). For IHD and IS, the HRs were somewhat greater for fatal than non-fatal events. For IS and MOVD (S4 Fig), the associations were stronger at younger ages (both p for trend < 0.001) and in rural areas (p for heterogeneity = 0.003 and < 0.001, respectively). The risk of MCE varied little across population subgroups.
Self-reported diabetes was also associated with significant, but more modest, excess risk of ICH, with an adjusted HR of 1.24 (95% CI 1.07–1.44), more extreme for fatal (1.59, 95% CI 1.30–1.96) than for non-fatal (1.00, 95% CI 0.81–1.23) events (Table 2). The risk varied little across participant subgroups (S5 Fig). For total stroke, self-reported diabetes was associated with an adjusted HR of 1.61 (95% CI 1.54–1.69; Table 2).
Additional adjustment for waist-to-hip ratio modestly, but non-significantly, attenuated the HRs for ischaemic cardiovascular diseases, but not for ICH or cardiovascular mortality (S1 Table). In sensitivity analyses excluding participants who developed incident diabetes during follow-up (n = 8,896), HRs for all cardiovascular diseases remained largely unchanged (S2 Table). Sensitivity analyses adjusting for age, SBP, and physical activity as continuous variables also produced similar findings.
Among participants with screen-detected diabetes, the excess risks were also highly significant, though somewhat less extreme compared with self-reported diabetes. For cardiovascular mortality, the adjusted HR was 1.66 (95% CI 1.51–1.83), while for MOVD, MCE, and IS the adjusted HRs were 1.50 (95% CI 1.41–1.58), 1.62 (95% CI 1.40–1.86), and 1.48 (95% CI 1.40–1.57), respectively. Similarly, for ICH the excess risk was more modest (1.17, 95% CI 1.01–1.36) (Table 2). There was little evidence of heterogeneity in the observed HRs for MCE, ICH (S6 Fig), or cardiovascular mortality (S7 Fig) across population subgroups. For IS and MOVD (S8 Fig), however, the risk was somewhat greater at younger ages (both p for trend < 0.001), in rural residents (p for heterogeneity < 0.001 and 0.003, respectively), and in more physically active individuals (p for trend 0.008 and 0.006, respectively).
Among individuals with diabetes, the risk of major ischaemic cardiovascular events increased progressively with longer duration of diabetes (p for trend < 0.001; Fig 1), but no such trend was seen for ICH (p = 0.3). Similarly, among individuals with self-reported diabetes, the risk of MOVD increased progressively with the number of other presenting cardiovascular risk factors, with adjusted HRs of 1.68 (95% CI 1.53–1.85), 2.08 (95% CI 1.93–2.23), and 2.82 (95% CI 2.59–3.06) in those with one, two, and three or more risk factors, respectively, compared with participants with self-reported diabetes who had no other cardiovascular risk factors (p for trend < 0.001; Fig 2). A similar gradient was seen for screen-detected diabetes (p for trend < 0.001; Fig 2).
By applying the age-specific population-attributable fractions for cardiovascular mortality in the present study to diabetes prevalence derived from a recent nationally representative survey [8] and the number of sex- and age-specific cardiovascular deaths in mainland China [22], it was estimated that 489,676 (95% CI 335,777–681,202) cardiovascular deaths could be attributable to diabetes in 2010.
Our study provides, to our knowledge, the first large-scale prospective evidence of the cardiovascular consequences of diabetes among adults in China. It showed that self-reported doctor-diagnosed diabetes was associated with 1.5- to 2.5-fold increased risks of cardiovascular mortality and incident IHD and IS. Moreover, it provided strong prospective evidence of a significant, though more modest, adverse association of diabetes with ICH risk. Individuals with screen-detected diabetes also had significantly increased cardiovascular disease risks. If these associations are causal, then almost 0.5 million cardiovascular deaths a year in China can now be attributed to diabetes.
Previous studies of predominantly Western populations have shown approximate doubling of IHD risk with diabetes [2,3,10,11,23]. The comparability of our risk estimates with previous studies suggests that differences between East Asian and Western populations in the relative importance of insulin resistance and beta-cell dysfunction in the aetiology of diabetes [9] have little, if any, impact on associated IHD risk. The method of ascertainment of IHD in the CKB may misclassify a proportion of individuals with “silent” IHD, which is more prevalent among individuals with diabetes [24], leading to underestimation of diabetes-associated IHD risk. The more modest effects of screen-detected—compared to self-reported—diabetes may reflect the shorter duration of, or less severe glycaemic aberration in, screen-detected diabetes, and greater potential for misclassification.
The CKB provides reliable evidence about the association of diabetes with stroke and, in particular, stroke subtypes. The association of screen-detected diabetes with IS in our study is largely consistent with data from studies of mostly Western populations [3]; the association with self-reported diabetes was moderately weaker than in those studies [2,3] and also weaker than in a small prospective Chinese study of 30,000 men and women [10] and a study—based on routine data and adjusted only for age—in Taiwan [25]. In mainland China, computed tomography or MRI is widely used, often leading to detection of a relatively high proportion of lacunar infarcts without major, or any, apparent focal neurological deficit [26]. The relatively low IS case fatality rate in our study might explain, at least in part, the more modest association with diabetes.
Existing evidence relating ICH to diabetes is more limited [2,3,10,25,27], partly due to lower ICH rates in Western populations [2,3] and the lack of widespread imaging in earlier studies. Where available, findings have been inconsistent [2–4,10,25]. Two of the largest studies in Western populations, including 1,000 [3] and 2,500 [4] ICH events, found modest elevated risks associated with known diabetes (HR 1.56 [3] and 1.28 [4]). Our study included almost 5,000 well-characterised ICH cases and provides the most reliable evidence to date of a positive, but modest, association between self-reported diabetes and ICH, and the first opportunity, to our knowledge, to investigate effect modification beyond age and sex. The large number of well-characterised stroke events (~90% of validated stroke events were confirmed with computed tomography/MRI) is a strength of the CKB. Outcome adjudication, through review of medical records for all IHD and stroke cases, is currently underway; findings to date have shown high positive predictive values for IHD events (~85%) and for stroke (~90%). The apparent difference in excess risk between fatal and non-fatal cardiovascular disease events could reflect poorer survival following such events in individuals with diabetes [28,29] or more severe disease in fatal cases.
The large urban—rural differences in excess risk associated with self-reported diabetes in the CKB, especially for IS, may reflect the higher proportion of undiagnosed diabetes cases in rural areas, such that self-reported diabetes in rural areas may represent more severe disease. More frequent use of diabetes medications in rural areas supports this hypothesis. Several studies have previously reported greater diabetes-associated IHD risk [3,23] and, less consistently, stroke risk [3,4,30] in women. We found a sex difference for cardiovascular mortality only, and this may have contributed to the higher diabetes-associated risks in never or occasional smokers and never or occasional alcohol drinkers. Diabetes, particularly type 2 diabetes, frequently coexists with other cardiovascular risk factors [31]. In our study population, the cardiovascular disease risk among those with diabetes was also dependent on the presence of other common and modifiable cardiovascular risk factors, which highlights the need for multi-factorial approaches to managing cardiovascular disease risk, even after development of diabetes. The small proportion of participants with self-reported diabetes who reported use of medications for lowering cardiovascular disease risk suggests that there is considerable scope for improvement in this regard in China. The impact of diabetes diagnosis duration on disease risk is largely consistent with the almost 2-fold greater odds of prevalent cardiovascular disease in early-onset (<40 y of age), compared to late-onset (≥40 y), diabetes in China, with the majority of the association explained by diabetes duration [32].
The CKB was not designed to be nationally representative, but given the size and diversity of the CKB study population and the CKB’s minimal loss to follow-up, this would not be expected to bias risk estimates or reduce their generalisability to the Chinese adult population [33]. Furthermore, the prevalence of diabetes in the CKB was similar to that reported in a reasonably contemporaneous representative Chinese survey in 2000–2001 [7]. More recent surveys have reported a higher rate of diabetes in China [7], with, for example, a prevalence rate of 11.6% in the 2010 survey [8]. As well as reflecting secular trends, the lower prevalence of diabetes in the CKB may reflect different study settings, sampling methods, and approaches used to identify undiagnosed diabetes. The prevalence of self-reported diabetes was, however, similar (3.5% in the 2010 survey versus 2.7% in the present CKB population and 3.2% in the CKB population including individuals with prior cardiovascular disease). Although self-reported diabetes is widely used [33], it may be subject to misclassification. However, amongst almost 30,000 participants resurveyed in the CKB, over 90% of those who reported a history of diabetes at baseline gave the same answer at resurvey. Furthermore, the mean plasma glucose level among those with self-reported diabetes was consistent with the diagnosis. We did not collect specific information on diabetes type, but, based on age at diagnosis and medication use, <1% of diabetes in the CKB would likely be type 1.
Separate examination of screen-detected diabetes should reduce bias from lifestyle changes and treatment following diagnosis, although screen-detected diabetes may be subject to greater misclassification resulting from use of RPG [34] measurement on a glucometer [35] for diagnosis. Such misclassification likely contributed to the lower screen-detected diabetes prevalence in the CKB compared to the most recent national survey [8] (2.7% in CKB versus 8.1% in the 2010 survey), which used multiple glycaemic indicators (oral glucose tolerance testing, fasting plasma glucose, and HbA1c), and would result in underestimates of diabetes-associated risks. However, excluding participants who developed new diabetes during follow-up (identified from mortality, disease surveillance, and national health insurance data) did not materially alter risk estimates. The high proportion of undiagnosed diabetes [8] and the elevated cardiovascular risk associated with it lend support to population screening for diabetes to enable more effective primary prevention of cardiovascular diseases. Data on lipids are not currently available; this may have resulted in residual confounding, although progressive adjustment for confounders in one published study suggests that this confounding would be minimal [3]. Similarly, renal function data are not currently available in the CKB; this precludes investigation of effect modification of the association of diabetes with cardiovascular diseases but would not bias our risk estimates. Additional adjustment of presented risk estimates for other potential confounding variables, including dietary intake, had minimal effect.
In China, about 10% of the adult population is estimated to have diabetes [8]. The present nationwide prospective study provides large-scale evidence from mainland China that individuals with diabetes are at significantly increased risk of major cardiovascular diseases, similar in magnitude to that observed in Western populations [2,3]. Much of the diabetes-associated excess risk is likely to be causal, accounting for almost 0.5 million cardiovascular deaths annually in China. With further adverse changes in lifestyle, the prevalence of diabetes will likely increase further in China, especially among young adults [1,7], foreshadowing an even greater diabetes-attributable burden of cardiovascular (and other) diseases.
|
10.1371/journal.pgen.1001342 | A Molecular Phylogeny of Living Primates | Comparative genomic analyses of primates offer considerable potential to define and understand the processes that mold, shape, and transform the human genome. However, primate taxonomy is both complex and controversial, with marginal unifying consensus of the evolutionary hierarchy of extant primate species. Here we provide new genomic sequence (∼8 Mb) from 186 primates representing 61 (∼90%) of the described genera, and we include outgroup species from Dermoptera, Scandentia, and Lagomorpha. The resultant phylogeny is exceptionally robust and illuminates events in primate evolution from ancient to recent, clarifying numerous taxonomic controversies and providing new data on human evolution. Ongoing speciation, reticulate evolution, ancient relic lineages, unequal rates of evolution, and disparate distributions of insertions/deletions among the reconstructed primate lineages are uncovered. Our resolution of the primate phylogeny provides an essential evolutionary framework with far-reaching applications including: human selection and adaptation, global emergence of zoonotic diseases, mammalian comparative genomics, primate taxonomy, and conservation of endangered species.
| Advances in human biomedicine, including those focused on changes in genes triggered or disrupted in development, resistance/susceptibility to infectious disease, cancers, mechanisms of recombination, and genome plasticity, cannot be adequately interpreted in the absence of a precise evolutionary context or hierarchy. However, little is known about the genomes of other primate species, a situation exacerbated by a paucity of nuclear molecular sequence data necessary to resolve the complexities of primate divergence over time. We overcome this deficiency by sequencing 54 nuclear gene regions from DNA samples representing ∼90% of the diversity present in living primates. We conduct a phylogenetic analysis to determine the origin, evolution, patterns of speciation, and unique features in genome divergence among primate lineages. The resultant phylogenetic tree is remarkably robust and unambiguously resolves many long-standing issues in primate taxonomy. Our data provide a strong foundation for illuminating those genomic differences that are uniquely human and provide new insights on the breadth and richness of gene evolution across all primate lineages.
| The human genome project has revolutionized such fields as genomics, proteomics and medicine. Markedly absent from these many advances however, is a formal evolutionary context to interpret these findings, as the phylogenetic hierarchy of primate species has only modest local (family and genus level) molecular resolution with little consensus on overall primate radiations. The exact number of primate genera is controversial and species counts range from 261–377 [1]-[3]. Although whole genome sequencing of 12 primate species are now completed, or nearly so, broader genome representation of man's closest relatives is necessary to interpret human evolution, adaptation and genome structure to assist in biomedical advances.
Primate taxonomy has undergone considerable revision but current views [1]-[3] concur that 67–69 primate genera originated from a common ancestor during the Cretaceous/Paleocene boundary roughly 80–90 MYA. An Eocene expansion formed the major extant lineages of 1) Strepsirrhini, which is composed of Lorisiformes (galagos, pottos, lorises), Chiromyiformes (Malagasy aye-aye) and Lemuriformes (Malagasy lemurs); 2) Tarsiiformes (tarsiers) and 3) Simiiformes composed of Platyrrhini (New World monkeys) and Catarrhini, which includes Cercopithecoidea (Old World monkeys) and Hominoidea (human, great apes, gibbons) (see Figure 1). Primate taxonomy, initially imputed from morphological, adaptive, bio-geographical, reproductive and behavioral traits, with inferences from the fossil record [1]-[3] is complex. Recent application of molecular genetic data to resolve primate systematics has been informative, but limited in scope and constrained to just specific subsets of taxa. Efforts to overcome this deficiency using a supermatrix approach [4], [5] with published sequences culled from these prior studies are inherently flawed by a prohibitively large proportion of missing data for each taxon (e.g. 59–85% see [5]).
Here we employ large-scale sequencing and extensive taxon sampling to provide a highly resolved phylogeny that affirms, reforms and extends previous depictions of primate speciation. In turn, the clarity of the primate phylogeny forms a solid framework for a novel depiction of diverse patterns of genome evolution among primate lineages. Such insights are essential in ongoing and future comparative genomic investigation of adaptation and selection in humans and across primates.
A comprehensive molecular phylogeny based on 34,927 bp (after correction for ambiguous sites from the original dataset of 43,493 bp per operational taxonomic unit, OTU) amplified from 54 nuclear genes in 191 taxa including 186 primates representing 61 genera is presented (Figure 1, Figure 2, Figure S1, Table S1, and Table S2). The phylogeny is highly resolved, with bootstrap values of 90–100% and Bayesian posterior probabilities of 0.9–1.0 at 166 of the 189 nodes (88%)(Table 1, Table 2, Table 3). Further, only 3 of 189 nodes (nodes 28, 38, 158) are polytomies in the bootstrap analyses (Table 1 and Table 3; Figure 2, Figure S1). (Note: nodes listed hereafter refer to Figure 2, Figure S1, Table 1, Table 2, Table 3). Roughly equal amounts of coding (14742 bp) and non-coding (17185 bp) genomic regions were sampled from X chromosome (4870 bp), Y chromosome (2630 bp) and autosomes (27427 bp) (Table 4) using newly developed PCR primers derived from a bioinformatics approach specific to primates in addition to primers from previous large-scale phylogenetic analyses (Materials and Methods, Tables S2, S3, S4).
Separate phylogenetic analyses of these data partitions are generally concordant. The greatest proportion of phylogenetically informative sites occurred in Y-linked genes (56%) compared with regions sequenced from the X-chromosome (40%) and autosomes (42%) (Table 4, Table S4), a finding also observed in carnivores [6], [7]. However, greater frequency of phylogenetic inconsistencies or unresolved nodes occur in these subset trees (Figures S2, S3, S4, S5, S6, S7, S8, S9, S10, S11, S12, S13), compared with the entire concatenated data set (Figure 2, Figure S1). Thus, these findings illustrate the need for both genome-wide datasets and maximum representation of species to resolve differences among previous studies that used only single genes, the uniparentally inherited mtDNA molecular marker and smaller numbers of primate taxa.
The relative placement of suborder Strepsirrhini and infraorder Tarsiiformes at an early stage of primate evolution has been difficult to resolve [8]-[11]. Presently distributed in the islands of Borneo, Sumatra, Sulawesi and the Philippines, Tarsiiformes had a broad Holarctic distribution during the Eocene [10]. Phylogenetic placement of tarsiers has alternatively been defined as 1) sister taxa to Strepsirrhini to form Prosimii [2], [8], [12], 2) allied with Simiiformes (Anthropoidea) to form Haplorrhini [1], [13], [14] and 3) a separate relict lineage with an independent origin [15]. Here we provide strong evidence that strepsirrhines split with suborder Haplorrhini approximately 87 MYA (node 185). The ancient lineage is monophyletic and defined by a long branch and eight shared insertions/deletions (indels) (node 144). Rooted by Lagomorpha, the phylogeny affirms Dermoptera as the closest mammalian order relative to Primates, followed by Scandentia [16], [17].
A long continuous Tarsiiformes branch (node 142), marked by 25 synapomorphic indels, is consistent with a relict lineage of ancient origin. The sequence phylogeny unambiguously supports tarsiers as a sister lineage (albeit distant) to Simiiformes (BS = 85 MP; 98 ML; 0.99 PP) to form Haplorrhini (node 143). A few indels (Table S5) define alternate evolutionary topologies, such as tarsiers aligned with Strepsirrhini (1 indel, ZFX) or Scandentia (1 indel, DCTN2), compared with those that support an ancestral grouping of Tarsiiformes +Strepsirrhini +Dermoptera +Scandentia (2 indels, PLCB4, POLA1). These incongruent alternatives suggest further investigation of more complex rare genomic changes as cladistic markers of ancient speciation is needed [17], [18].
Aided by samples of rare taxa, the phylogeny expands upon recent findings [19]-[21] to better resolve long-standing questions on the evolution of Lorisiformes and the two endemic Madagascar infraorders of Chiromyiformes and Lemuriformes. Our data affirm the ancient split of Strepsirrhini, approximately 68.7 MYA (node 144), into the progenitors of Lemuriformes/Chiromyiformes (origin 58.6 MYA, node 174) and Lorisiformes (origin 40.3 MYA, node 184).
Lorisiformes evolution includes the radiation of Lorisidae (pottos and lorises, 37 MYA, node 179) and Galagidae (19.9 MYA, node 183) species. Within Lorisidae, the four extant genera split into the African subfamily Perodicticinae (Arctocebus, Perodictus) and the Asian subfamily Lorisinae (Nycticebus, Loris) and are the most divergent within all of primates. For example, mean nucleotide divergence between Lorisidae species is 4–5 times that observed in family Hominidae (Figure 3) and significantly (p<0.05) exceed the average genetic divergence across all of Strepsirrhini (nodes 176–178, Table S7, Figure 3). Galagidae are found only in Africa and currently are divided into four genera. However, the Otolemur lineage (node 180) is placed as part of a paraphyletic grouping (node 182) along with two other extant Galago lineages (nodes 181, 183), suggesting that further taxonomic investigation of Galago is warranted.
Common ancestors of Chiromyiformes and Lemuriformes likely colonized the island of Madagascar prior to 58.6 MYA (node 174). Noted for extensive adaptive evolution, the relative hierarchical branching patterns of the four Lemuriformes families (Indriidae, Lepilemuridae, Lemuridae, Cheirogaleidae) recognized by taxonomists, has proven difficult to resolve conclusively. Inferences on species versus subspecies classification are controversial with as many as 97 Malagasy lemurs [22] under taxonomic review. Chiromyiformes diverged from a common ancestor with Lemuriformes shortly after colonisation of Madagascar [14], [19] and today consists of a single relict genus Daubentonia defined by a long branch with high indel frequency (N = 14) (Figure 2, Figure S1, Table S7). The evolution of the four Lemuriformes families began 38.6 MYA (node 173) with the emergence of Lemuridae, followed by Indriidae and a monophyletic lineage that split 32.9 MYA (node 152) to form the sister lineages of Lepilemuridae and Cheirogaleidae. This branching pattern among families agrees with earlier nuclear gene segment findings [20] that differ from studies using mtDNA sequence and Alu insertion variation which were unable to resolve these hierarchical associations [19]. Further, relatively weak nodal support here collapses Lemuriformes into an unresolved trichotomy of Lemuridae, Indriidae, and the Lepilemuridae + Cheirogaleidae lineage (node 158). Optimal resolution of this node is observed with exon sequences (Figures S8 and S9), indicating that intron sites may be saturated, while more conserved coding regions remain informative and reflect the ancient rapid radiation of Lemuriformes families.
The phylogeny clarifies formerly unresolved questions concerning New World primate evolution including branching order among families, relative divergence of genera within families, and phylogenetic placement of Aotus, and provides genetic support for examples of adaptive evolution that led to nocturnalism, “phyletic dwarfism” and species diversification within the Amazonian rainforest. Here, Platyrrhini clearly diverged from a common ancestor with Catarrhini (node 141) roughly 43.5 MYA during the Eocene. Although questions remain about the route and nature of primate colonization of the New World [23], [24] and the impact of historic global climate change in neotropical regions [25], [26], the phylogeny unambiguously resolves the relative divergence pattern among families from a common ancestor 24.8 MYA (node 78).
The common ancestor to Pitheciidae (uakaris, titis and sakis) originated 20.2 MYA (node 140) and the majority of these species currently are distributed in the neotropical Amazonian basin extending from the Andean slopes to the Atlantic. Next to radiate are the Atelidae (node 126), with the most basal lineage leading to Alouatta (howler monkeys), currently widely distributed from Mexico to northern Argentina, followed by the divergence of Ateles (spider monkeys) from South American lineage comprised of sister genera (node 121) of Lagothrix (woolly monkeys) and Brachyteles (muriquis).
The Cebidae radiation initiated with the emergence of sister taxa Cebus (Cebinae) and Saimiri (Saimirinae) approximately 20 MYA (node 113), in agreement with other molecular studies [27]-[30]. Subsequently, during a relatively brief interval (∼700,000 years) a lineage arose (node 112) that split to form the Callitrichinae (marmosets and tamarins) and Aotus (night monkeys). The Aotus lineage (node 98) radiated with unusually high numbers of synapomorphic indels (N = 15), the most observed in Simiiformes (Table 2 and Table 3), to form a complex species group of controversial taxonomic designation as subfamily or family and uncertainty over its exact placement relative to other Cebidae lineages. Here, Aotus is the sister lineage to Callitrichinae (marmosets, tamarins) as originally hypothesized by Goodman (1998) [1], [28]. Aotus species divide into sister lineages, with the “grey-necked” species (A. trivirgatus + A. lemurinus griseimembra) distributed north of the Amazon River, and “red-necked” species A. nancymaae, A. azarae species and associated subspecies located most to the south (nodes 98, 101, 102). The unusual depth of divergence (i.e. sizeable nucleotide substitutions/site; high indel frequency) may exemplify adaptive speciation as Aotus are the only nocturnal Simiiformes [31], and thereby may have reduced competition with diurnal small-bodied platyrrhines inhabiting the same neotropical environments.
Another case of adaptation termed “phyletic dwarfism,” defined as a gradient in morphological size partially correlated with evolutionary time [32], is supported in Cebidae. Aotus, Cebus and Saimiri species are larger than the more derived and smaller squirrel-sized Callitrichinae of Saguinus, Leontopithecus, Callimico, Mico, Cebuella and Callithrix. In Callitrichinae, Saguinus is the first to diverge with S. fuscicollis currently distributed south of the Amazon River. Subsequently, the genus diversified into northern (S. bicolor, S. midas, S. martinsi, S. geoffroyi, S. oedipus) and south Amazonian species (S. imperator, S. mystax, S. labiatus); a trend generally similar to findings based on mtDNA [33] and single nuclear genes [34]. The hierarchical branching order among the remaining Callitrichinae of Leontopithecus, Callimico, Callithrix and Mico mirrors decreasing body size and culminates with the smallest platyrrhine species, Cebuella pygmaea, as most derived. This phylogenetic depiction of Callitrichinae is concordant with several other morphological and reproductive traits [32], [35] related to dwarfism and perhaps reflects adaptive evolution selected by fluctuating resource availability within the Amazon and Atlantic coast rainforests [36].
Cercopithecoidea (family Cercopithecidae) speciation patterns are confounded by symplesiomorphic traits in morphology, behavior and reproduction, and are further confused by hybridization between sympatric species, subspecies and populations (summarized in [2]). Cercopithecidae includes two subfamilies, Colobinae and Cercopithecinae, which diverged 18 MYA (node 62), but classification schemes [2] are marked by inconsistencies between morphological [37], [38] and genetic data, as well as differences among genetic data studies [4], [27], [39]-[44].
Colobinae radiation started approximately 12 MYA (node 42) with species adapted to an arboreal, leaf-eating existence. Asian (tribe Presbytini) and African (tribe Colobini) genera are monophyletic (nodes 53 and 61, respectively), supporting earlier genetic findings [4], [40] over morphology-based taxonomy [2], [45]. Whilst African genera Piliocolobus and Colobus are commonly recognized, the taxonomic schemes for the critically endangered Asian langur and leaf monkeys, all sharing digestive adaptations for an arboreal folivorous diet, have ranged from a single genus Presbytis to three distinct genera (Trachypithecus, Semnopithecus, Presbytis). Here, the Presbytis lineage, distinguished by 3 indel events (node 56), diverged first within Asian Colobinae, followed by the odd-nosed group (Rhinopithecus, Nasalis, Pygathrix), Trachypithecus and Semnopithecus. As odd-nosed species are not exclusively arboreal and folivorous, the results indicate either 1) morphological convergence between Presbytis with Trachypithecus and Semnopithecus, 2) adaptation for an expanded diet in the odd-nosed group, or 3) that a folivorous diet is a symplesiomorphic trait within Asian colobines.
Trachypithecus and Semnopithecus genera consist of closely related, often sympatric species (node 51), distributed in the Indian subcontinent and SE Asia, with inconsistent phylogenetic resolution among species [4], [27], [40], [44], [46]. Nonetheless, all genetic studies, including the present, place Trachypithecus vetulus (monticola) nested within the Semnopithecus clade (node 50), suggesting the need for taxonomic revision. Further, previously ambiguous associations between Trachypithecus and Semnopithecus (nodes 43–51) are clarified. Inter-specific genetic differences are roughly half those observed among other colobine genera (Figure 2, Figure S1, Table 3, Table S9) and may indicate that recent speciation, taxonomic over-splitting, reticulate evolution, or a combination thereof, (e. g. see [40], [44], [46]) are common within the Asian Colobinae radiation.
The remainder of Old World monkeys (tribes Papionini and Cercopithecini) [2] arose from a common ancestor approximately 11.5 MYA (node 41). Considerable interest in Cercopithecinae speciation is motivated not only by primate conservation, but increased biomedical surveillance for novel zoonotic agents and comparative research of host-pathogen adaptation relevant to the study of deadly human viral pandemics such as HIV/SIV.
Cercopithecini (guenons, patas monkey, talapoin, green monkeys) include lineages rooted by divergent monotypic genera followed by more recent speciation, characterized by transition from an arboreal to a terrestrial lifestyle. Generally arboreal, Miopithecus and Allenopithecus are early offshoots with respect to the two Cercopithecini subclades formed approximately 7 MYA. The Cercopithecus lineage (node 34) radiated after Miopithecus and retained an arboreal lifestyle. The second, rooted by Allenopithecus, forms a terrestrial clade of Erythrocebus patas and Chlorocebus species, with Cercopithecus l'hoesti separated the other Cercopithecus. This paraphyly, also reported in earlier genetic studies [39], [47], [48] and counter to initial morphological classifications [2], suggests taxonomic revision of Cercopithecus. Further, resolution of Allenopithecus (node 40) and Miopithecus (node 35) speciation herein suggests a single evoluiontary transition from an arboreal to a terrestrial lifestyle in E. patas, C. l'hoesti, and Chlorocebus species.
Papionini (macaques, mandrills, drills, baboons, geladas, mangabeys) is a taxonomically complex tribe [2]. One of the more familiar genera within Cercopithecoidea, Macaca (macaques) diverged 5.1 MYA and today is represented by an African lineage comprised of a single species M. sylvanus, and an Asian lineage consisting of well-defined species groups (fascicularis, sinica, mulatta, nemestrina, Sulawesi) inhabiting India and Asia, SE Asia and Sundaland [49]. With the exception of the fascicularis group, which is split in this study whereby M. arctoides [fascicularis] is more closely aligned with M. thibetana [sinica] rather than M. fascicularis as expected, our data otherwise strongly support these macaque species groups (nodes 6, 11). Moreover, the phylogeny affirms Groves [2] proposal that Lophocebus and Theropithecus are distinct clades apart from Papio (nodes 18, 19), although the average nucleotide divergence among these three genera are generally less than between other recognized Papionin genera (Macaca, Mandrillus, Cercocebus) (Figure 2, Figure S1, Table 3, Table S9). Lastly, sequence divergence between tribes is unequal with Cercopithecini nearly twice that of Papionini (mean branch length = 13.1, 7.43, respectively, p<0.005) and there are numerous instances of discordance between the present phylogeny with previous mtDNA studies [4], [5] suggesting that continued resolution of Cercopithecinae speciation and of Papionini in particular, will likely include evidence of reticulate evolution represented by ongoing and historic episodes of hybridization (e.g. see [39], [48]).
Once contentiously debated, the closest human relative of chimpanzee (Pan) within subfamily Homininae (Gorilla, Pan, Homo) is now generally undisputed. The branch forming the Homo and Pan lineage apart from Gorilla is relatively short (node 73, 27 steps MP, 0 indels) compared with that of the Pan genus (node 72, 91 steps MP, 2 indels) and suggests rapid speciation into the 3 genera occurred early in Homininae evolution. Based on 54 gene regions, Homo-Pan genetic distance range from 6.92 to 7.90×10−3 substitutions/site (P. paniscus and P. troglodytes, respectively), which is less than previous estimates based on large scale sequencing of specific regions such as chromosome 7 [50]. The highly endangered orangutan forms the single genus Pongo in subfamily Ponginae (nodes 75–76), the sister lineage to Homininae. Currently restricted to the islands of Borneo and Sumatra, orangutans once inhabited all of Southeast Asia during the Pleistocene [51]. Differences in behavior, morphology, karyology, and genetic data between the two island populations [2] support the taxonomic designation as two separate species of Bornean (P. pygmaeus) and Sumatran orangutans (P. abelii), and these designations are upheld by the data presented here.
Hylobatidae (siamang, gibbons, hoolock) are noted for exceptional rates of chromosome re-arrangement [52], [53], 10–20 times faster than in most mammals [54]. Classification schemes of the 12 species range from two genera (Hylobates and Symphalangus) to four subgenera and/or genera (Hylobates, Nomascus, Symphalangus, Hoolock), defined by unique numbers of chromosomes [54], [55]. The eight species included in this study form three clades that coincide with genus designation (absent is Hoolock; nodes 64–69) that diverged rapidly 8.9 MYA. Moreover, Nomascus species appear more recent than Symphalangus and Hylobates, with node divergence dates estimated at less than 1 MY (Table 3, Table S9, Figure 2). Thus, Hylobatidae exhibits episodes of rapid divergence perhaps related to excessive genome re-organization and warrants additional investigation.
The clarity of the primate phylogeny here can be used to assess nucleotide divergence patterns, rates of substitution and accumulation of synapomorphic and autapomorphic indels. Genome divergence varies across primate lineages, with the least inter-specific differences observed in Cercopithecidae lineages and the most in Lorisidae, reflecting recent speciation in the former and the more ancient origins of the latter (Figure 3, Table 1, Table 3, Tables S7 and S9).
The global rate of nucleotide substitution across the entire primate phylogeny is 6.163×10−4 substitutions/ site/ MY, but exhibits significant heterogeneity across lineages (Figure 3) and among branches (Table 1, Table 2, Table 3; Tables S6, S7, S8). For example, the “hominoid slow-down” hypothesized to have occurred in human evolution, is confounded by the reduced rates observed in all Catarrhini (not just Homininae) compared with Platyrrhini and Strepsirrhini (Figure 3, Table S10). By contrast, the “phyletic dwarfism” of the Callitrichinae (nodes 97, 85) and the evolution of nocturnalism in Aotinae are correlated with increased rates along specific branches (see nodes 99, 100) rather than an being a function of an average rate among all branches within the lineage (Figure 3), suggesting that an adaptive “speed-up” occurred in the common ancestors of these extant species.
The genome accumulates indels over evolutionary time, altering the degree of sequence homology between taxa. Further, large-scale genome sequence analysis demonstrate that indel formation is an indicator of genome plasticity, positively correlated with adjacent nucleotide substitution rate [56], [57], gene segmental duplication, chromosomal position, hybridization between species and speciation, and is enhanced by molecular mechanisms of recombination among repetitive elements [58]-[60]. Here, the distribution of indels is ubiquitous in both coding and noncoding segments (Tables S4, S5, S6), but is markedly disjunct among primate lineages (Figure 3). Excluding the infraorders Tarsiiformes (25 indels) and Chiromyiformes (14 indels) due to statistically inadequate sampling, the indel frequency per branch varies by a factor of 20 (Table 1, Table 2, Table 3; Tables S7, S8, S9) with the greatest accumulation within Lorisidae (particularly Arctocebus calabarensis) and the least in Cercopithecidae (Figure 3). The major correlate of indel frequency is not substitution rate, but overall genome divergence represented by branch length (R-square = 0.659 Lorisiformes; 0.610 Lemuriformes; 0.3286 Simiiformes; P<0.05).
The molecular genetic resolution of the primate phylogeny provides a robust comparative genomic resource to affirm, alter, and extend previous taxonomic inferences. Approximately half of the 261–377 species and 90% of the genera are included facilitating resolution of long-standing phylogenetic ambiguities. Early events within primate evolution are resolved such as: Dermoptera is the closest mammalian order to Primates; Tarsiiformes are sister taxa with Simiiformes to form Haplorrhini; Chiromyiformes (Daubentoniidae) and Lemuriformes are monophyletic indicating a common ancestral lineage colonized the island of Madagascar once; and the hierarchical divergence pattern among New World families Pitheciidae, Atelidae, and Cebidae is clarified.
Additional insights are possible because the relative branching patterns among infraorders, parvorders, superfamilies, families, subfamilies, genera and species are resolved with high measures of support for all but three nodes. For example, Old World monkeys (Cercopithecoidea) display remarkably low levels of divergence, particularly within Papionini, consistent with reticulate evolution, recent speciation and possibly augmented by taxonomic over-splitting. By contrast, the Lorisidae are marked by extraordinary divergence relative to other primate lineages. In the New World, the phylogenetic placement of the unique, nocturnal Aotinae is unambiguously resolved, diverging rapidly after the sister lineage of Cebinae+Saimirinae and prior to the Callitrichinae within the family Cebidae. Further, the pattern of divergence of Callitrichinae is correlated with a gradation in species size, supporting “phyletic dwarfism” [32], [35]. In the context of human evolution, the large amount of sequence available here for each well-recognized species in Hominidae provides a baseline estimate of average genetic divergence per taxonomic level in primates. However, deviations from these values observed across diverse lineages illustrate the remarkable biodiversity and species richness within the Primate order.
One of the more intriguing unresolved questions is the origin of primates. Generally concordant, most molecular data suggest extant primates arose approximately 85 MYA from a common ancestor. However, the debate continues over the geographic locale most consistent with the existing fossil record [9], [10], [12], [16], [23], [26], [61]-[63]. A parsimonious interpretation of the present data would suggest an Asian origin as the ancient Asian Tarsiiformes and the strepsirrhine Lorisinae are most basal and the closest relatives of primates, Dermoptera and Scandentia, are also exclusive to Asia.
Primate genomes harbor remarkable differences in patterns of speciation, genome diversity, rates of evolution and frequency of insertion/deletion events that are fascinating in their own right, but also provide needed insight into human evolution. Advances in human biomedicine including those focused on changes in genes triggered or disrupted in development, resistance/susceptibility to infectious disease, cancers, mechanisms of recombination and genome plasticity, cannot be adequately interpreted in the absence of a precise evolutionary context or hierarchy. Resolution of the primate species phylogeny here provides a validated framework essential in the development, interpretation and discovery of the genetic underpinnings of human adaptation and disease.
Primate DNA samples were obtained following the guidelines of Institutional Animal Care and Use Committee policies of respective research institutions (see Table S1). All tissue samples for the Laboratory of Genomic Diversity were collected in full compliance with specific Federal Fish and Wildlife permits from the Conservation of International Trade in Endangered Species of Wild flora and Fauna: Endangered and Threatened Species, Captive Bred issued to the National Cancer Institute (NCI)-National Institutes of Health (NIH) (S.J.O. principal officer) by the U.S. Fish and Wildlife Services of the Department of the Interior. Duke University Lemur samples (J.E.H.) were collected under research project BS-4-06-1 and Institutional Animal Care and Use Committee (IACUC) project A094-06-03, and this paper is Duke Lemur Center publication #1192.
A complete list of individual and source DNA are presented in Table S1. DNA was extracted from whole blood, buffy coat, hair or buccal swab samples using DNeasy Blood & Tissue Kit (Qiagen) following manufacture's protocol. DNA from different tissues (muscle, kidney etc) or cell culture pellets was extracted using standard phenol∶chloroform extraction methods. Proteinase K digestion in lysis buffer (100 mM NaCl, 10 mM Tris-HCl pH 8.0, 25 mM EDTA pH 8.0, 0.6% SDS, 100 µg/ml RNAse A) at 56 °C for 3–12 hours rotating was followed by 30 minute phenol, phenol∶chloroform 70∶30, and chloroform extractions using phase-lock gel tubes (Eppendorf) followed by ethanol precipitation and 70% ethanol wash. Dried DNA was reconstituted in TE pH 7.4 buffer and stored at 4 °C. DNA was quantified using Nanodrop (Thermo Scientific) and its quality was assessed using 0.7% agarose gel electrophoresis.
DNA of limited quantity was used for whole-genome amplification using REPLI-g Midi Kit (Qiagen). 50–100 ng of genomic DNA (depending on its quality) was used per one 50 µl reaction according to the manufacturer's protocol. A negative control (no template) was included in every WGA and was verified by downstream PCR and sequencing. Some strepsirrhine DNA was extracted and/or whole genome amplified as previously described [21].
A complete list of 54 primer sets used in this study is presented in Table S2. This list includes primers from earlier studies [12], [16], [21], [64]-[68], as well as those designed specifically for this study using a unique bioinformatics approach (Pontius, unpublished data). A panel of species representing the breadth of primate diversity was used in the testing and optimization of PCR primers and included: Gorilla gorilla, Pan paniscus, Nomascus leucogenys, Symphalanges syndactylus, Erythrocebus patas, Macaca fuscata, Macaca tonkeana, Chiropotes satanas, Saimiri boliviensis, Saimiri sciureus, Callithrix jacchus, Ateles fusciceps, Saguinus fuscicollis, Cheirogaleus medius, Lemur catta and Tupaia minor.
All nuclear gene regions in all the samples were amplified with the following conditions. Either 30 ng of genomic DNA or 1 µl of WGA product was diluted 1∶10 with 0.1XTE per PCR reaction. DNA quantity was increased for poor quality DNA. Genomic and WGA DNA was aliquoted into plates, dried at room temperature and stored at 4 °C. Each 15 µl PCR reaction contained 2 mM MgCl2, 250 µM of each dNTP, 150 µM of each forward and reverse primer, 0.8 units of AmpliTaq Gold polymerase (ABI) with 1X GeneAmp 10X PCR Gold Buffer. PCR was performed in PE ABI GeneAmp 9700 and Biometra T1 thermal cyclers. PCRs were carried out using a touchdown program with the following parameters: initial denaturation for 10 min at 95 °C; followed by 10 cycles of 95 °C for 15 s, 60–52 °C (2 cycles for each of the five down gradient annealing temperature steps: 60 °C, 58 °C, 56 °C, 54 °C and 52 °C) for 30 s, and 72 °C for 1 min; and followed by 25 cycles of 95 °C for 15 s, 50 °C for 30 s, and 72 °C for 1 min; and a final extension at 72 °C for 30 min. PCR products were analyzed on 2% agarose gels. Only PCR products that produced single bands were sequenced. PCR products were purified using AMPure kit (Agencourt) or Mag-Bind EZ Pure (OMEGA). PCR products were sequenced directly in two reactions with forward and reverse primers. The sequencing reactions were carried out with the BigDye Terminator v1.1 cycle sequencing kit (Applied Biosystems, Inc.). For 10 µl sequencing reactions we used 0.25 µl of BigDye, 2 µl of 5X Sequencing buffer, 0.32 µM primer and 2.5 µl of PCR product (we diluted PCR product if bands on the gel were too bright). Sequencing reactions were performed as following: 25 cycles of 96 °C for 10 s, 50 °C for 5 s, 60 °C for 4 minutes. Sequencing products were purified using paramagnetic sequencing clean-up CleanSEQ (Agencourt) or Mag-bind SE DTR (OMEGA). PCR and sequencing cleanups were performed on Beckman Coulter Biomek FX laboratory automation workstation. The sequencing products were analyzed with an ABI PRISM 3730 XL 96-well capillary sequencer. Some of the prosimian PCR products and sequences were obtained following earlier published methods [21]. Consensus sequences for each individual were generated from sequences in forward and reverse directions using Sequencher 4.9 program (Gene Codes Corporation). All sequences were deposited in GenBank under accession numbers presented in Table S11.
Multiple sequence files for each gene segment amplified were aligned by MAFFT version 6 [69], [70], imported into Se-Al ver 2.0a11 [71] and verified by eye. Regions of sequence ambiguity within the alignment were identified by GBLOCK version 0.91b [72], and removed from subsequent phylogenetic analyses. A FilemakerPro database was created to manage all sequence records for each individual DNA specimen and the concatenated dataset was exported. The final, post-GBLOCK, edited, annotated PAUP* nexus alignment of the 54 concatenated genes used for this study is publically available at the following website:
http://lgdfm3.ncifcrf.gov/190Taxa_Rabbit_PAUP.zip
The file is a compressed zip file that can be viewed in either a generic text editor, PAUP*, or alignment programs that read large nexus format files.
Gene partitions were analyzed separately, as well as combined, for genome comparison and phylogenetic reconstruction. Six gene partitions were created, corresponding to X-chromosome, Y-chromosome, autosome, intron, exon and UTR segments. A separate phylogenetic analysis was conducted for each of the six data partitions to compare the concordance among tree topologies derived from each partition. It should be noted that the Y-chromosome tree is not directly comparable to the topologies of the other data partitions because the number of males (N = 127) was a subset of the total (N = 191). In the concatenated data set of all 54 genes, females were coded as “missing” for the Y-chromosome gene sequence. Aligned multiple sequence files of either combined data or gene partitions were imported into ModelTest ver 3.7 [73] and the optimal model of nucleotide substitution was selected using the AIC criterion. Models are listed in Table S12.
Phylogenetic trees based on nucleotide data were obtained using a heuristic search with different optimality criteria of maximum likelihood (ML) and maximum parsimony (MP) as implemented in PAUP* ver 4.0a109 [74] for Macintosh (X86) and additional runs of ML as implemented in GARLI ver 0.96 [75]. In PAUP*, conditions for the ML analysis included starting trees obtained by stepwise addition, and branch swapping using the tree-bisection-reconnection (TBR) algorithm. The MP analyses used step-wise addition of taxa, TBR branch swapping and excluded indels. Support for nodes within the phylogeny used bootstrap analysis with identical settings established for each method of phylogenetic reconstruction and values greater than 50% were retained. The number of bootstrap iterations consisted of 1000 for MP methods and 100 for ML. Detailed control files used for GARLI ML analyses are available from corresponding author.
We estimated the phylogeny and divergence time splits simultaneously using a Bayesian approach as implemented in the program BEAST ver 1.5.3 [76], [77]. Due to computational constraints, analyses were performed with 5 different sets of species: 1) genus-level data set including 61 Primate genera, two Dermoptera genera and one Scandentia genus rooted by Lagomorpha, 2) Catarrhini species with outgroups, 3) Platyrrhini species with outgroups, 4) Strepsirrhini species with outgroups and 5) genus-level analysis with a partitioned data set allowing for rate heterogeneity and different substitution models for autosome, X-chromosome, and Y-chromosome sequences.
By using the uncorrelated lognormal relaxed-clock model, rates were allowed to vary among branches without the a priori assumption of autocorrelation between adjacent branches. This model allows sampling of the coefficient of variation of rates, which reflects the degree of departure from a global clock. Based on the results of ModelTest, we assumed a GTR+I+G model of DNA substitution with four rate categories. Uniform priors were employed for GTR substitution parameters (0, 100), gamma shape parameter (0, 100) and proportion of invariant sites parameter (0, 1). The uncorrelated lognormal relaxed molecular clock model was used to estimate substitution rates for all nodes in the tree, with uniform priors on the mean (0, 100) and standard deviation (0, 10) of this clock model. We employed the Yule process of speciation as the tree prior and a Unweighted Pair Group Method with Arithmetic Mean (UPGMA) tree to construct a starting tree, with the ingroup assumed to be monophyletic with respect to the outgroup. To obtain the posterior distribution of the estimated divergence times, nine calibration points were applied as normal priors to constrain the age of the following nodes (labeled A-H in Figure 1 of main text): A) mean = 40.0 MYA, standard deviation (stdev) = 3.0 for time to most recent common ancestor (TMRCA) of galagids and lorisids [78], B) mean = 43.0 MYA, stdev = 4.5 for TMRCA of Simiiformes [79], [80], C) mean = 29.0 MYA, stdev = 6.0 for TMRCA of Catarrhini [80], D) mean = 23.5 MYA, stdev = 3.0 for TMRCA of Platyrrhini [26], [81], E) mean = 7 MYA, stdev = 1.0 for TMRCA of Papionini [82], F) mean = 4.0 MYA, stdev = 0.4 for TMRCA of Theropithecus clade [40], [83], G) mean = 15.5 MYA, stdev = 2.5 for TMRCA of Hominidae [14] and H) mean = 6.5 MYA, stdev = 0.8 for TMRCA of Homo-Pan [84]. A normal prior for the mean root height of 90.0 MYA with stdev = 6.0 was used based on molecular estimates of MRCA of all Primates [14], [82], [85]. The calibration points selected are based on fossil dates that have undergone extensive review in previous publications and are supported by a consensus of paleoanthropologists. Rather than re-iterate the considerable amount of information forming the basis for each calibration point, we list the respective citations with the most detailed overview and attendant references.
Four to seven independent Markov chain Monte Carlo (MCMC) runs for each analysis were run for 20–100 million generations to ensure sampling of estimated sample size (ESS) values. The Auto Optimize Operators function was enabled to maximize efficiency of MCMC runs. Trees were saved every 1000 generations. Log files from each run were imported into Tracer ver 1.4.1, and trees sampled from the first 1 million generations were discarded. Mixing of trees was assessed in Tracer by examination of ESS values. Analysis of these parameters in Tracer suggested that the number of MCMC steps was more than adequate, with ESS of all parameters often exceeding 200, and Tracer plots showing strong equilibrium after discarding burn-in. Tree files from the individual runs were combined using LogCombiner ver 1.5.3 after removing 1000 trees from each sample. The maximum-clade-credibility tree topology and mean node heights were calculated from the posterior distribution of the trees. Final summary trees were produced in TreeAnnotator ver 1.5.3 and viewed in FigTree ver 1.3.1.
Heterogeneity in nucleotide substitution rates among primate taxa was assessed by a Bayesian approach, allowing for unequal rates of nucleotide substitution among lineages as implemented in BEAST. Rate estimates provided for each branch within the primate phylogeny were analyzed by ANOVA as implemented in SAS (SAS Institute Inc., SAS 9.1.3). Significant differences among means used the Duncan multiple means test.
Indels were assessed as possible indicators of genome plasticity among primate lineages. An a priori approach was developed that used the derived primate phylogenetic tree (Figure 2) as a guide for identification of synapomorphic and autapomorphic indels. First, all indels were identified using FASTGAP on GBLOCKED alignments and verified by eye. Second, only indels that correctly conformed to the species associations of the primate phylogeny (Figure 2) were used and identified as a subset of synapomorphic events (Table 1, Table 2, Table 3; Tables S5, S6). Third, another subset of autapomorphic indels were identified and assessed as potential signatures of genome plasticity for a given species (Tables S7, S8, S9). Infrequently, some indels included in the analysis were positioned in regions that did not amplify across all species. In these cases, indels were identified as synapomorphic for a lineage providing ∼70% of the relevant species were successfully PCR amplified, and that species with missing sequence for the indel did not all occur on the same node within the lineage. The hypothesis that patterns of nucleotide substitution are influenced by indel frequency was tested by regression of ln-transformed branch length against ln-transformed indels per branch. Tests of the association between genome rates of evolution and indel frequency were conducted by regression of the rate of nucleotide substitution (substitution/site/MY) versus ln-transformed indel frequency per branch. Statistical software used was SAS (SAS Institute Inc., SAS 9.1.3).
|
10.1371/journal.pgen.1006930 | Fruit weight is controlled by Cell Size Regulator encoding a novel protein that is expressed in maturing tomato fruits | Increases in fruit weight of cultivated vegetables and fruits accompanied the domestication of these crops. Here we report on the positional cloning of a quantitative trait locus (QTL) controlling fruit weight in tomato. The derived allele of Cell Size Regulator (CSR-D) increases fruit weight predominantly through enlargement of the pericarp areas. The expanded pericarp tissues result from increased mesocarp cell size and not from increased number of cell layers. The effect of CSR on fruit weight and cell size is found across different genetic backgrounds implying a consistent impact of the locus on the trait. In fruits, CSR expression is undetectable early in development from floral meristems to the rapid cell proliferation stage after anthesis. Expression is low but detectable in growing fruit tissues and in or around vascular bundles coinciding with the cell enlargement stage of the fruit maturation process. CSR encodes an uncharacterized protein whose clade has expanded in the Solanaceae family. The mutant allele is predicted to encode a shorter protein due to a 1.4 kb deletion resulting in a 194 amino-acid truncation. Co-expression analyses and GO term enrichment analyses suggest association of CSR with cell differentiation in fruit tissues and vascular bundles. The derived allele arose in Solanum lycopersicum var cerasiforme and appears completely fixed in many cultivated tomato’s market classes. This finding suggests that the selection of this allele was critical to the full domestication of tomato from its intermediate ancestors.
| Starting about 10,000 years ago, during the Neolithic period, human societies began the transformation from a hunting and gathering-dependent lifestyle to an agrarian lifestyle. This transformation was accompanied by plant and animal domestication. Tomato shows a huge increase in fruit weight that has arisen as a consequence of its domestication. We identified a gene that encodes a poorly characterized protein that controls fruit weight in tomato. The mutation that led to the increase in fruit weight arose early during the cultivation of tomato and is now incorporated in all large tomato varieties. The gene regulates cell size in the fruit and is called Cell Size Regulator. The increases in cell size are proposed to relate to cellular maturation that accompanies fruit growth.
| Rapid morphological diversification among closely related organisms often arise in response to strong selection pressures such as those imposed by domestication. Starting approximately 10,000 years ago, during the Neolithic period, human societies began the transformation from a hunting and gathering-dependent lifestyle to an agrarian lifestyle which was accompanied by plant and animal domestication [1, 2]. The process of domestication is associated with the taming of wild relatives and the selections of types that benefit human use in terms of food production and clothing as well as shelter and companionship. Specifically for vegetable and fruit crops, the domesticated plants feature much larger produce, reduced plant branching and often a determinate growth habit [1].
The fully wild ancestor of tomato (Solanum pimpinellifolium) is indigenous largely to the coastal regions of Ecuador and Peru [3, 4]. Selections to the intermediate type, S. lycopersicum var cerasiforme, took place in the Andean mountain region of Ecuador and Northern Peru and further selections to the earliest domesticate type, S. lycopersicum var lycopersicum, took place in Central America [4]. Continued selections after the initial domestication events led to a huge diversity of size and shape with fruit weight increasing as much as a 1,000 fold [5].
Even though the fruit is a terminal structure, the parameters that determine its final shape and weight are rooted throughout the plant’s lifespan. The development of the fruit starts as early as the formation of the inflorescence and floral meristem 3 weeks after sowing [6, 7]. In tomato, floral development from inflorescence meristem to anthesis-stage flowers takes place in approximately 3 weeks whereas fruit development from opened flower to ripe fruit lasts between 4.5 and 7 weeks depending on the genotype [7, 8]. Final fruit morphology is the result of a combination of processes that regulate meristem organization, overall cell division rates and duration, cell shape and cell expansion. Meristem organization processes that lead to larger meristems support the formation of a bulkier fruit. For example, natural mutations in the tomato orthologs of the meristem architecture genes WUSCHEL (WUS) and CLAVATA3 (CLV3) lead to fruit with more locules associated with large and flat fruits [9]. Induced mutations in genes in the same WUS-CLV3 pathway also support the notion that enlarging meristems lead to bulkier fruit [10]. Distinct from meristem organization are the cell division, cell shape and cell expansion processes which can be controlled at multiple time points throughout the ontogeny of the ovary and fruit. These cellular processes are typically defined to take place before or after anthesis [11]. Whereas details of cell division, cell shape and cell expansion are not well described for tomato ovary development, processes following pollination and fertilization of the ovules are much better understood. The initial stages of fruit development following successful pollination and fertilization are marked by a rapid increase in cell proliferation through reinitiating cell divisions [7, 8]. In most fruit tissues such as the pericarp, cell division ceases 5 to 10 days after anthesis with the exception of the epidermal cell layer. Following this rapid cell division stage, fruit growth continues by extensive cell enlargements that last for three to five weeks until the fruit ripening stage [7, 8]. Thus far, two natural mutations in tomato impact cell number by increasing the number of cell layers in the pericarp of the fruit. This increase in cell layers is achieved by changing cell division rates and/or duration in the developing ovary or fruit [12, 13]. CNR/FW2.2 changes cell layers in the ovary walls and these organs are already larger at anthesis [13]. SlKLUH/FW3.2 also impacts cell layers mainly in the pericarp but of the developing fruit instead of the ovary [12]. In addition to fruit size, the mutation in SlKLUH is associated with delayed ripening suggesting that the duration of cell division is extended [12]. Derived alleles for CNR/FW2.2 and SlKLUH/FW3.2 are resulting in expression level changes as gene expression is different in the NILs. Moreover, association mapping led to the most significant polymorphisms in the promoter of these two fruit weight genes [12, 14]. Thus far, natural mutations in tomato genes that impact cell size have not yet been identified. However, increases in cell size are associated with tomato domestication and the emergence of larger fruit [15].
The increase in size from a 1 to 2-mm wide ovary to a 5 to 10-cm wide fruit is predominantly the result of the dramatic increase in cell enlargement in the pericarp that follows the cell proliferation stage [7, 8]. In the developing fruit, the transition and maintenance of cell enlargement is not well understood. On the other hand, the transition from proliferation to enlargement is better described in root development [16, 17]. Cells differentiate from the root meristem zone to the transition zone which is controlled by the antagonistic action of auxin and cytokinin [16, 18]. Certain cytokinin signaling regulators, the B-type Arabidopsis Response Regulators (ARRs), appear to play crucial roles in synchronizing the entry into the endoreduplication cycle in roots of Arabidopsis [16, 17, 19]. In roots, the transition zone also marks the initiation of endoreduplication, a process marked by DNA replication without nuclear division. This process is known to occur in maturing cells and results in higher nuclear ploidy levels by bypassing chromatid segregation and cytokinesis [20, 21]. Endoreduplication is proposed to occur in many multicellular organisms as a potential mechanism for cell enlargement and cellular differentiation. In particular for tomato, the proper development of the tomato fruit is intimately associated with endoreduplication [15, 22].
We here report on the cloning and characterization of a novel tomato fruit weight gene Cell Size Regulator (CSR). This quantitative trait locus had been identified in a prior study as a minor QTL explaining only 8% of the phenotypic variance [23]. Despite its small effect on the trait in a segregating F2 population, the heritability in progeny studies was high, facilitating the fine mapping and eventual cloning of the gene. The derived allele CSR-D increases fruit mass through cell enlargement which is accompanied by small increases in nuclear ploidy level of pericarp cells. Expression and co-expression analyses with CSR suggests a role in cell differentiation during the later stages of fruit development, including vascular development. We propose that CSR impacts cellular differentiation leading to a potentially indirect effect on endoreduplication during tomato fruit maturation. The CSR-D mutation likely arose in S. lycopersicum var cerasiforme and became entirely fixed in the large fruited tomato germplasm suggesting that this improvement was an important step towards the domestication of tomato from the cherry type to the cultivated S. lycopersicum var lycopersicum.
The fw11.3 locus was previously mapped to a 149 kb region on chromosome 11 [24] and further fine-mapped to a 13kb region between markers EP2030 and EP2032 (Fig 1A, S1 Table). This region included two full open reading frames, Solyc11g071940 and Solyc11g071950, and the partial Solyc11g071960. The common nucleotide polymorphisms among the fw11.3 segregating populations were a large deletion at the 3’ end of Solyc11g071940 and four SNPs in the non-coding region surrounding the same gene (Fig 1B; S1 Table). We next conducted an association study to determine whether the partial gene deletion was correlated with fruit weight variation in the tomato germplasm. Using a population that was developed to conduct association studies for fruit weight in tomato [25], the 3’ deletion of Solyc11g071940 was shown to be significantly associated with fruit enlargement and over three years of fruit weight evaluations (Fig 1C). The association of the large deletion in Solyc11g071940 with fruit weight implies that this gene underlies fw11.3.
Detailed sequence analyses showed that the derived allele carried a 1,406 bp deletion and 22 bp insertion resulting in a 194 amino-acid truncation of the predicted wild type protein. Interestingly, the truncation might not result in a loss-of-function of fw11.3, as the derived allele was partially dominant over the wild type allele [24]. To further corroborate that Solyc11g071940 was the likely candidate FW11.3 gene and that the derived allele constituted a gain-of-function mutation, we constructed transgenic complementation lines by transferring the partially dominant allele of Solyc11g071940 to the lines in the wild type fw11.3 background. Fifteen independent transgenic complementation lines (T0) were obtained of which four progeny families (T1) were evaluated at least once. Fruit weight was significantly increased in nearly all transgenic progeny lines compared to their non-transgenic sibs (Fig 1D, S2 Table). Therefore, the results from the transgenic complementation lines demonstrate that Solyc11g071940 controls tomato fruit size and that the deletion found in the derived allele is indeed acting in a dominant manner. This finding supports the notion that fw11.3-D functions as a gain-of-function mutation.
As expected, the fruit weight, fruit area and fruit perimeter were significantly larger in the fw11.3-D nearly isogenic lines (NIL) in the cultivated background (Table 1). To evaluate the effect of the QTL in the wild species background, we developed NILs where fw11.3-D was introgressed in the mostly LA1589 background, an accession of S. pimpinellifolium that is the closest fully wild ancestor species of cultivated tomato. These NILs were either differing for just fw11.3 or were fixed for the derived alleles at fw2.2 and fw3.2 while differing at the fw11.3 locus. Similar to the NILs in the cultivated background, fruit weight, fruit area and fruit perimeter were significantly larger in the fw11.3-D in the wild species background (Table 1). To evaluate which parts of the fruit were enlarged by the derived allele, we analyzed the pericarp, columella and placenta areas in the NILs (Fig 2). The pericarp area was significantly enlarged in the NILs that carry the derived allele in both cultivated and wild tomato species backgrounds. By contrast, the columella and placenta area were not consistently different in the NILs (Table 1). Because the effect of fw11.3 was most pronounced in the pericarp in all the NILs, we evaluated whether increased cell layers (number) in the mesocarp or increased cell size contributed to the larger pericarp area values. Our data showed that the number of cell layers was mostly unaltered in NILs in both cultivated and wild species backgrounds whereas mesocarp cell size was increased in all the lines carrying the derived allele of fw11.3 (Fig 2, Table 2). Moreover, transgenic lines that complemented the large fruit phenotype also showed increased cell sizes compared to non-transgenic control. Thus, the role of FW11.3 is to increase cell size and therefore we named Solyc11g071940 to Cell Size Regulator (CSR).
The predicted CSR protein featured a domain that is recognized as the FAF domain (formely known as DUF3049, Pfam accession PF11250). The tomato genome harbored 13 predicted proteins that represented FAF domain family. In Arabidopsis, the FAF domain was found in 10 proteins which included four FANTASTIC FOUR (FAF) and a single FAF-like protein. In tomato, three genes were recognized as CSR paralogs because their proteins were closely related to one another and shared similar motif patterns. We named these three paralogs CSR-like1, CSR-like2 and CSR-like3 (Fig 3). At the protein level, the MEME motifs 2 and 3 represented the FAF domain whereas motif 6 defined the CSR family proteins. Motifs 4 and 5 were also often found in members of the CSR clade. The single copy Arabidopsis FAF-like protein was the most similar to the tomato CSR clade, whereas the four FAFs formed a separate clade (Fig 3A). Likely orthologs of the four Arabidopsis FAFs were identified in tomato, including three SlFAF1/2 and two SlFAF3/4. The FAF proteins have been recognized as controlling shoot meristem size [27].
Only one CSR member was found in Arabidopsis whereas four CSR paralogs were found in tomato. To further explore the apparent expansion of the CSR family, we searched databases for protein members that were closely related to CSR and included motifs 4 and/or 6. The most likely orthologs of tomato CSR and CSR-like 1 to 3 were found in other Solanaceous species namely potato, eggplant and chile pepper (Fig 3B). When comparing species with sequenced genomes outside the Solanaceae family but within the Asterids clade, only one copy of CSR was found in coffee (Gentianales), sesame and mimulus (Lamiales) (Fig 3B). However, based on motif structure, this single paralog appeared more similar to the CSR-like1 to 3 encoded proteins than to CSR. For species in the Rosids clade, most of them carried only one copy of a CSR-like except watermelon, cucumber and poplar. These results suggested a single common ancestor of CSR that evolved independently in Rosids and Asterids clades, and only expanded to four in the Solanaceace family.
To further understand the role in tomato fruit growth, we evaluated the temporal and spatial expression patterns of CSR. Despite its effect only on fruit weight, CSR-WT expression was not detected in whole developing fruit (Fig 4A). Low expression however, was detected in mature leaf tissues (Fig 4A, S5 Table). Moreover, analyses of other RNA-seq datasets showed that CSR expression was undetectable in tomato meristems (http://tomatolab.cshl.edu/efp/cgi-bin/efpWeb.cgi), as well as ovary and 4 days old fruit tissues collected by laser captured microscopy (http://ted.bti.cornell.edu/cgi-bin/TFGD/digital/experiment.cgi?ID=D009). These results demonstrated that CSR expression was undetectable from meristems, young floral buds, anthesis to the cell proliferation stage of fruit development. Although CSR expression was undetectable when evaluating the whole fruit, we expected gene expression in certain tissues within this organ at later developmental stages. Indeed when sampling the pericarp, columella and seed&placenta tissues separately from maturing fruits, the expression of CSR was found to be low but detectable. CSR was expressed the highest in the columella where its expression increased between 7 to 33 days post anthesis (DPA), and decreased again at the turning stage (Fig 4B). CSR was also expressed in pericarp and seed/placenta tissues albeit at low or very low levels (Fig 4B). Interestingly, CSR-D showed higher expression compared to CSR-WT but with a similar profile in both columella and pericarp. Recently, higher resolution expression information in developing tomato fruits (from J.K.C Rose, Cornell University) also supported the notion that CSR-D was highest expressed in the columella. Moreover, the expression was particularly high in the vascular bundles of the developing tomato pericarp (S2 Fig).
Of all the genes that encode the FAF motif, CSR-like1 was the highest expressed in nearly all plant tissues examined including seedling tissues (cotyledons, hypocotyl, shoot apex and roots) and in the columella of developing fruit (S5 Table). In developing fruit, CSR-like1 featured the most similar expression dynamic as CSR, even though CSR-like1 was often 15 fold or higher expressed than CSR-D (S5 Table). Expression of CSR-like2 and CSR-like3 was specifically detected in hypocotyl, cotyledon and shoot apex (Fig 4A, S5 Table). In all, CSR was expressed dynamically in developing fruit and CSR-D allele was higher expressed than CSR-WT allele. On the other hand, tomato FAF1/2a showed the highest expression in floral meristem tissues and early developing floral buds; FAF1/2c was particularly well expressed in emerging leaves; and FAF3/4a was well expressed in seedling tissues as well as floral meristem and early floral buds. In sum, CSR and CSR-like genes show distinct expression in maturing tissues whereas the FAFs are highest expressed in meristems and young tisues.
Fruit weight was an important selection criterion that drove the evolution of tomato from a wild relative bearing small fruit to the large tomatoes found in grocery stores today. To determine whether selections for the derived allele of CSR might have taken place, we evaluated the origin and distribution of CSR-D in wild, semi-domesticated, early landraces as well as modern breeding germplasm (Fig 5). The mutation was found in low frequency in S. lycopersicum var cerasiforme from Peru and Mesoamerica. However, the distribution of the derived allele frequency greatly increased in the Mesoamerican domesticated landraces and is now completely fixed in the large fruited contemporary germplasm. The contemporary germplasm included processing and fresh market tomatoes, the latter which were comprised of globe/large used for slicing and plum tomatoes used for soups and stews [4]. These results imply that CSR-D arose late but became rapidly fixed during the selections by early farmers.
In this study, we demonstrated by fine mapping, association mapping and plant transformation experiments that the tomato fw11.3 QTL is controlled by Solyc11g071940 encoding a largely uncharacterized protein. The 1.4 kb deletion caused a 194 amino acid truncation at the C-terminus of the encoded protein creating a partial dominant allele. The heavier fruit was predominantly caused by increased pericarp area resulting from larger cells in this tissue, hence the name CSR which is short for Cell Size Regulator. Increases in cell size in the pericarp were correlated to increases in nuclear ploidy levels in this same tissue albeit that CSR’s effect on this process appears to be small. Fixation of the derived allele in cultivated tomato supported the notion that increased cell size mediated by CSR was indeed critical for the recent evolution of the crop.
In roots, cellular differentiation occurs when cells progress from the meristematic (dividing) zone through the transition zone into the elongation and differentiation zone. Development of the tomato pericarp may follow a similar trajectory as described for the root: immediately following pollination and fertilization and after a short period of cell division, pericarp cells transition to the cell expansion stage. In small fruited tomatoes, the cell layers in the walls of the tomato increase from 10 to 20 in less than five days following pollination [7]. This period is followed by approximately 3.5 weeks of cell enlargement prior to ripening. Expression of CSR coincides with the cell expansion stage and therefore, this gene might be associated with cellular differentiation. A conserved domain FAF with unknown function has been identified in CSR and CSR-like proteins. This domain is named after the Arabidopsis FANTASTIC FOUR (FAF) proteins regulating shoot meristem size [27]. Whereas FAFs are expressed strongly in meristematic cells and young growing tissues, CSR expression was undetectable in these tissues and instead only found in maturing plant organs such as the fruits prior to ripening (S5 Table). Thus, the function of CSR and FAFs is not likely the same at the tissue level even though biochemically they might be involved in similar processes within the cell. Our research efforts focused on the effect of CSR in the pericarp because this tissue contributes greatly to overall fruit weight, and the area was significantly expanded through cell size increases in the fw11.3-D NILs. However, expression of CSR was the highest in the columella and not the pericarp (Fig 4). Recently, we obtained higher resolution expression of CSR-D in developing tomato fruit which showed that expression of this gene is high in and around vascular bundles in the pericarp. Thus, the high expression of CSR in the columella could be due to the relatively high vasculature density in columella tissues. We attempted to evaluate cell size in the columella which ranged from very small (in the center) to very large (towards the periphery). However, coumella structure varied a lot from fruit to fruit and therefore, it was not feasible to evaluate cell size similarly in the columella sections taken from different fruits. Since the entire vascular bundle was captured by laser captioned microdissection, it is not known where in or around the bundles CSR is expressed. To further understand the function of CSR, future studies should be directed to evaluate its role in vascular development and in the columella.
Even though CSR encodes a protein of unknown function, co-expression analyses may have revealed the cellular processes that provide insights into the function of CSR. The antagonistic roles of auxin and cytokinin might possibly be associated with CSR function since several genes associated with these pathways were found in the co-expression clusters. For example in Arabidopsis root development, the B-type response regulators ARR2 and ARR12 play important roles in the mitotic exit by upregulating the expression of CCS52A1, an activator of the APC/C complex, and SHY2/IAA3, leading to inhibition of auxin signaling [17, 19]. Putative orthologs of ARR2 (Solyc01g0655540) and ARR12 (Solyc07g005140) were found in the CSR co-expressed dataset. Additional cytokinin signaling and biosynthetic proteins were identified including LOG3 (Solyc11g069570) of unknown function but thought to play a role in the biosynthesis of cytokinins, a cytokinin synthase protein IPT5 (Solyc01g080150) [29], a cytokinin receptor CRE/AHK4 (Solyc04g008110) [30], and a cytokinin sensitivity protein PRL1 (Solyc01g094480) [31] were found in the CSR gene clusters in the fw11.3 NILs. Auxin signaling and response genes were found in these clusters as well, namely genes encoding auxin transport protein PILS5 (Solyc03g032080) and ABCB15 (Solyc02g087410) [32, 33], an auxin signaling protein ARF11 (Solyc05g0560400), a pleckstrin domain-containing protein (Solyc08g066860) that is involved in vascular patterning and auxin canalization [34], a HB-2 transcription factor (Solyc08g078300) involved in cell expansion and response to auxin [35, 36], IBR5 (Solyc12g005990) a dual specificity phosphatase that promotes auxin responses and acts as a regulator of organ size in Arabidopsis [37], and a KNOTTED-like protein 1 (Solyc04g077210) whose expression is repressed by auxin resulting in the promotion of leaf fate [38, 39]. GO term enrichment for genes in the CSR expression clusters also implied that plant vascular development is one of the processes that may be affected by CSR and CSR-like1 which was consistent with its expression in this tissue. Indeed, deeper searches in the entire list of co-regulated genes led to the identification of additional genes that may be critical in vascular development. These include genes that encode putative orthologs of receptor-like kinases XIP1 (Solyc04g077010) involved in the maintenance of cell files or cell morphology in conductive elements [40] and BRI1-like 2 (Solyc04g008430) associated with the development of provascular/procambium cells [41–43], as well as several phloem expressed lectin genes (Solyc02g069020, Solyc02g069030, Solyc02g069060, Solyc02g031740). In addition, putative orthologs of genes involved in vascular development [44] were found in the dataset including APL (Solyc12g017370), CNA (Solyc12g044410) and BRX-like 4 (Solyc12g044410). With CSR expression particularly high in vascular bundles and the columella (which is enriched for xylem and phloem tissues), these co-regulated genes suggest that CSR and perhaps CSR-like1 play a role in cellular maturation in the vascular bundle leading indirectly to increases in pericarp cell size and nuclear ploidy. Therefore, CSR might be involved in the antagonistic effects of auxin and cytokinin as a mechanism for cellular differentiation and enlargement in different tissue types.
Cell enlargement is also associated with enhanced endoreduplication and larger fruit weights [15]. Ubiquitously found in both higher plants and animals, the function of endoreduplication is not well understood other than its association with increased cell size. The transition from cell proliferation to enlargement in roots coincides with the initiation of endoreduplication [16, 17, 19, 20]. Other studies have suggested a role for endoreduplication in enhanced metabolism [45] or to sustain growth under adverse conditions including pathogen attack [46, 47]. Even though the function is not well understood, the mechanisms regulating the core entry and progression of endoreduplication has been reasonably well established. Distinct stages of cell division are regulated by CYCLINS (CYC), Cyclin-Dependent Kinases (CDK) and CDK inhibitors (CKI). The onset and progression of endoreduplication is mediated by these same core cell cycle proteins through transcription factors regulating gene expression as well as regulators that control the ubiquitination of the proteins which then targets them for proteolytic degradation. Specifically, CYCLINS and CDKs that regulate the M stage are suppressed when endoreduplication is promoted [20, 48]. This suppression is mediated by activation of a specific E3 ubiquitin ligase named the Anaphase Promoting Complex/Cyclosome (APC/C) that leads to ubiquitination of the mitotic Arabidopsis proteins CDKB1;1 and CYCA2;3 which is then followed by proteasome-mediated protein degradation [49]. Activators of APC/C are for example Cell Cycle Switch52 A1 (CCS52A1) encoding a WD-repeat protein and mutations or transcriptional downregulation leads to termination of cell expansion and reduced endoreduplication. It is unlikely, however that CSR plays an important role in endoreduplication because none of the known aforementioned endoreduplication genes such as those encoding CYCLINS, CDK and CKI or any related to APC/C proteasome complex and most of their transcriptional and translational regulators, were present in the CSR gene expression clusters (S1 Dataset) [26]. Moreover, the impact of CSR-D on enhancing endoreduplication was very small.
How did the CSR clade including the FAF-like genes evolve in plants? In our study, we observed that CSR expanded in the Solanaceae family resulting in four paralogs: CSR-like1, CSR-like2, CSR-like3 and CSR. Among them, CSR-like2 and CSR-like3 might have been the result of a recent tandem duplication event because they shared high sequence similarity to one another and were located next to each other on chromosome 1. When compared with selected species in the Solanaceae family, most of the four paralogs corresponded to an ortholog in pepper, eggplant and potato. The clustered CSR-like2 and 3 shared two orthologs in potato and pepper, but only one in eggplant. The eggplant genome sequence is not yet available and thus, it was possible that eggplant also carried two CSR-like2/3 gene copies. As for other species in the Asterids clade, coffee, sesame and mimulus carried only one paralog of the CSR family genes. This result suggest that the duplication resulting in CSR, CSR-like1 and CSR-like2/3 occurred after the Solanales diverged from the Gentianales and Lamiales orders, but before the Solanaceae family split to tomato, potato, pepper and eggplant. A single FAF-like gene is found in monocot and dicot species and is proposed to be the ancestor to FAF [27] as well as the CSR clade based on our findings (Fig 3A). In fact, the predicted protein motif structure of FAF-like in Arabidopsis is more similar to CSR than to FAF. As CSR and CSR-like genes only expanded in the Solanaceae family, they may have evolved specific functions that are specific to the family. Regardless and as mentioned above, the CSR subclade of the FAF proteins may be involved in cellular differentiation and enlargement resulting from the antagonistic action of auxin and cytokinin. And this might be the unifying role of CSR/CSR-like/FAF-like genes in land plants.
Tomato seeds or DNA were obtained from the TGRC (http://tgrc.ucdavis.edu/; S. pimpinellifolium LA1589); Tomato Growers Supply Co (Howard German, Rio Grande, Yellow Pear); Dr. Mathilde Causse (Institut National de la Recherche Agronomique-Avignon, France; Tomato core collection for association mapping [25]); Drs. Maria José Díez and Jose Blanca (Universitat Politècnica de València, Spain; Instituto de Conservación y Mejora de la Agrodiversidad Valenciana collection [4]); Dr. David Francis (The Ohio State University, United States; SolCAP collection [50]). Plants were grown in field (for fruit weight and plant evaluations) and greenhouse (for population development and evaluation of additional plant phenotypes) at Ohio Agricultural Research and Development Center (Wooster, OH USA) or at the University of Georgia (Athens, GA USA) under standard conditions. For the greenhouse, the plants were grown in 2-gallon pots using 15-9-12 Osmocote slow release supplemented by 20-20-20 fertilization in Fafard 3B growing media with supplemental lighting.
To fine map fw11.3, a population derived from a cross between S. lycopersicum c.v. Howard German (HG) and the wild species S. pimpinellifolium accession LA1589 was used [51] (S4 Fig). New markers were developed for the fine mapping and were based on the tomato genome sequence and known marker sequences (S6 Table). Eight recombinants were obtained through screening of 1906 seedlings, and two additional recombinants were obtained through screening 732 seedlings. Fruit weight was compared between fw11.3-WT and fw11.3-D plants within each recombinant family using Student’s t-test (S1 Table). Additional mapping was conducted in populations derived from crosses between S. lycopersicum c.v. Rio Grande (RG) × LA1589 and Gold Ball Livingston (GBL) × Yellow Pear (YP). In RG, three recombinants were progeny tested (S1B Table) whereas another three recombinants were progeny tested from an GBL × YP F2 population [52] (S1C Table).
NILs in the cultivated background were developed from the fine mapping populations in the HG and RG populations and only differed at the locus of interest while all other parts of the genome were fixed. Resulting from initial backcrosses to the cultivated parent, 75% and 87.5% of the loci in these NILs were estimated to be fixed for the HG or RG parent, respectively. The HG NILs originated from BC1F7 11S62-2 plant (S4 Fig), with the introgression region of around 36kb. The RG NILs originated from BC2F5 12S114-5 plant in RG mapping population (S4B Fig), with the introgression region around 532kb. A second set of RG NILs were derived from the BC2F6 heterozygous plant 12S255-11, with the introgressed region of around 131kb.
The fw11.3 NILs in LA1589 background were developed by marker-assisted selection after several generations of backcrossing and selfing. Breeding scheme for NIL development is schematically shown in S4C Fig. The entire introgressed region at fw11.3 locus (73 kb) contained 11 annotated genes (ITAG2.4 tomato genome annotation release), including CSR. In all cases, the NILs were grown in a randomized plot design and therefore cultivated under the same conditions.
The accessions used for the association mapping were a Core Collection that included 93 S. pimpinellifolium, S. lycopersicum var cerasiforme and S. lycopersicum var lycopersicum accessions [25] (S7 Table). Association mapping was performed using three InDel markers, HP61, HP32 and HP31 (S6 Table) found in the 13 kb region spanning the locus in a population described by [25]. Association analysis was performed using MLM model of TASSEL2.1 software [53]. Population structure matrix Q and kinship matrix K were generated with STRUCTURE 2.2 [54] and SPAGeDi [55], respectively. Twenty EST-SSR markers distributed throughout the genome were used to generate Q and K matrix [25].
Fosmid SL_FOS0119H09 (provided by Dr. J.J. Giovannoni, USDA-ARS, Ithaca NY) from tomato cultivar Heinz1706 spanned the fw11.3-D locus. To release the insert for transformation, the clone was digested with AvrII and SgrAI corresponding to the Solyc11g071940 coding region, 7,101-bp upstream region and 3,466-bp downstream region. The resulting 11,314 bp insert was cloned into the Agrobacterium transformation vector pHaoN, modified from pCAMBIA1300 (by adding the following selectable marker: Pnos-KAN-Tnos), linearized by digestion with XbaI and XmaI. The enzymers AvrII and XbaI, SgrAI and XmaI are two pairs of isocaudomers that generate compatible ends after digestion. The complementation construct pHORF2 was transformed into VIR347, carrying the wild type allele of fw11.3 similar to LA1589; and also in the HG NIL containing the LA1589 wild type allele (11S167 carrying fw11.3-WT, S4A Fig). Transformation was conducted at the Plant Transformation Core Research Facility at the University of Nebraska-Lincoln (Dr. T. Clemente) and independent transgenic lines were identified after Southern blot hybridizations of EcoRI and EcoRV-digested genomic DNA following standard procedures. For each transgenic family, 10 to 13 transgenic complementation plants and non-transgenic sib plants were identified from the T1 generation using marker assisted selection, transplanted in a random plot design, and evaluated under the same conditions. A total of four independent transgenic events (HF3, HF4, 9F, CF7) were evaluated in 2013 and 2014, and HF3 and 9F were replicated in both years. For the T1 transgenic lines from HF3-1 and HF4 grown in 2014, only a few fruit had matured at the time of harvest and 8 to 12 fruit per plant were weighted individually. Due to large variations among plants within same genotype, plants that carried the largest and smallest fruit of the same genotype were removed prior to statistical evaluations [26] (S2 Table). Average fruit weight per plant was calculated by divding the fruit number by total weight.
Cell layer and mesocarp cell area were measured in pericarp of breaker-stage fruit of the fw11.3 NILs, and the VIR347 transgenic and non-transgenic sib plants. Two to three slices per fruit and two representative fruit per plant were evaluated. Transverse sections of the pericarp (approximately 1 mm thick and 1 cm long) cut from the equatorial region were stained by adding a drop of 0.5% Toluidine Blue in 0.1% Na2CO3 solution for 1 to 2 seconds. The samples were rinsed with water to prevent staining of the internal cell layers. The stained sections were photographed using the attached digital camera (SPOT RT KE, Diagnostic Instruments) on the Leica MZFLIII dissecting microscope (Leica Microsystems, Switzerland). The cell layers were counted four times in each section from the exocarp to endocarp avoiding the vascular bundles. Largest cell size was measured by tracing the six largests cells with ImageJ. The average cell size was measured by counting the cell number in an equal sized rectangle and divided by the rectangular area. Student t-test were used to compare the cell number and cell size difference between the two genotypes.
Representative mature green fruit were used for ploidy analyses. Five slices of fresh pericarp tissue (1 mm thick slice of 0.5–1 cm2 area, avoiding septum tissue) from each fruit were chopped finely under 1.2 ml nuclei extraction buffer (100 mM Tris-Cl pH 7, 85 mM NaCl, 5 mM MgCl2, and 0.1% Triton X100), with a razor blade A 100 μm nylon mesh filter (Sysmex Partec GmbH, Görlitz, Germany) was used to filter the nuclei from the chopped tissue suspension (600 μL). Three μL of DAPI solution (0.2 mg/mL) was added to each sample prior to loading onto the BD LSRII flow cytometer (Biomedical Research Tower Facility, Columbus, OH) or the CyAn ADP flow cytometer (Beckman Coulter, Cytometry Shared Resource Laboratory, Athens GA) for fluorescence-activated cell sorting (FACS) analysis. 10,000 nuclei were counted using 405 nm laser excitation and blue emission filter 450/50, Only 3,000 nuclei were counted if gating became clogged (possibly due to debris interference). A 7 DPA fruit and mature leaves were used as internal control to calibrate nuclei content C-values for the RG NILs and LA1589 NILs, respectively. After manually adjusting the gating to exclude background noise, the histograms of different nuclei level (C-values) events were generated (S1 Fig). The percentage of each ploidy level from all nuclei counts was calculated. The lower C value nuclei were not evaluated as those peaks were not discernable above background noise. The percentage of each ploidy level was compared between fw11.3-D and fw11.3-WT. To calculate the Endoreduplication Index (EI), we modified the established formula by removing 2C and 4C ploidy levels and calculated EI as EI’ = [4C*1+ 8C*2+ 16C*3+ 32C*4+ 64C*5+ 128C*6+ 256C*7+ 512C*8] / [total counts from 4C to 512C] [56].
Ovaries at anthesis were collected for size measurement. The middle part of ovaries were infltrated and fixed with 3% glutaraldehyde, 2% paraformaldehyde in 0.1 M potassium phosphate buffer pH7.4 overnight, dehydrated with a graded ethanol series from 25% to 100%. Ovaries were cut transversely in the middle with sharp blade after critical point drying and before platinum coating. Samples were scanned and imaged recorded with Hitachi S-3500N scanning electron microscope (Hitachi Ltd, Japan) under high vacuum. Ovary sizes were measured with the images using ImageJ software. Total yield of ripe and green fruit were recorded separately by the weight and number of the fruit according to previously established protocols [57]. Fruit ripening was recorded with hand pollinated first two flowers of each inflorescence. Six to ten fruits per plant that set well were evaluated. The dates were recorded as each fruit turned to orange (30%-60% surface color change to orange) and red (>90% surface color change to red). Fruit quality was measured as the total soluble solid content (degree of Brix). A quarter of each 14 representative ripe fruits were blended together per plant. The homogenized juice was filtered with Kimwipes and Brix was measured by a pocket refractometer (ATAGO CO., LTD, Tokyo, Japan). Seeds were extracted from fully ripe fruit and soaked in 12.5% HCl, rinsed and air dried in the laboratory for 6 days on mesh screens with paper towels, weighted and counted. Inflorescence number and flower number per inflorescence per plant were counted with greenhouse grown plants under 20-20-20 with Calcium supplement or 14-7-14 fertilizer condition. Developing inflorescences and aborted flowers were also included. Plant architecture was measure as plant height, node number, side shoot number and total side shoot length at 55 and 66 days after sowing (DAS) in the greenhouse and 87 and 108 DAS or 42 and 63 days after transplanting (DAT) in the field. Leaf structure were measured with the mature leaves at 8th, 9th, and 10th nodes counting from cotyledon. Leaf weight, rachis length, petiole length, intercalary leaflet number, secondary leaflet number, tertiary leaflet number and terminal leaflet size (width and length) were measured. To test source-sink relationship, two fruits per inflorescence of a total of 7 inflorescences were kept and fruit weight from these plants was compared with control plants with no fruit removal. All phenotypic evaluations were performed with fw11.3 NILs under RG background with two replications except source-sink relationship experiment, each with 6 to 13 plants (S3 Table). Student t-tests were performed to compare each trait between fw11.3-D and fw11.3-WT NIL.
To identify conserved domains in the CSR protein, the Conserved Domain Database (CDD, https://www.ncbi.nlm.nih.gov/cdd/) [58] was used. The FAF domain (Pfam accession PF11250) was identified in 14 proteins in tomato genome protein sequence ITAG2.4 release (SGN, http://solgenomics.net/tools/blast/; Solyc01g009260.1.1, Solyc01g009270.1.1, Solyc01g079740.2.1, Solyc01g098570.2.1, Solyc06g008990.1.1, Solyc06g054310.1.1, Solyc06g073940.2.1, Solyc06g074270.1.1, Solyc06g084280.1.1, Solyc09g065140.1.1, Solyc10g018270.1.1, Solyc11g068530.1.1, and Solyc11g071940.1.1 (the latter corresponds to CSR). Thirteen of the tomato proteins were used in this study, except Solyc04g072650.1.1, which appeared to be an outlier in the protein phylogeny. To ensure the accuracy of protein sequences, all encoded proteins were confirmed using the predicted mRNA sequences with the ExPASy translation tool [59]. Additional motif searches were conducted using the full length protein sequence in MEME 4.10.0 [60] using the settings of 6 different motifs with 6–50 motif width. Three predicted proteins (Solyc06g073940.2.1, Solyc01g009260.1.1, and Solyc01g009270.1.1) featured the most similar motif patterns and highest similarity with CSR (E-value ≤ 1e-80), and therefore these were defined as paralogs. The FAF domain was found in 10 Arabidopsis proteins (TAIR, http://www.arabidopsis.org/wublast/index2.jsp) and seven of them were used in this study. Of these, only one was considered the closest paralogs of CSR because of high MEME motif similarities. Four FAF domain-containing Arabidopsis proteins corresponded to the FANTASTIC FOUR (FAF) clade.
Multiple sequence alignments of FAF by CLUSTALW2 [61] with default settings were performed and the results were exported to MEGA6 [62] for phylogenetic analysis. Neighbor-joining tree was constructed for 13 tomato and seven Arabidopsis FAF domain sequences with 1000 replicates for bootstrap validation. A FAF domain from Selaginella moellendorffii (Phytozome ID: 418746) was used as outgroup. Two proteins (Solyc01g079740.2.1 and Solyc06g054310.1.1) were closely related to FAF3 and FAF4 and were renamed SlFAF3/4a and SlFAF3/4b. Three proteins (Solyc06g084280.1.1, Solyc06g008990.1.1 and Solyc09g065140.1.1) were also closely related and presented the FAF1 and FAF2 subclade, and were renamed to SlFAF1/2a, SlFAF1/2b, and SlFAF1/2c, respectively.
To identify potential orthologs in other crop species, the full length protein sequence of CSR-WT and its three paralogs were used as a query using the SGN database (http://solgenomics.net/tools/blast/), Cucurbit Genomics Database (http://www.icugi.org/cgi-bin/ICuGI/tool/blast.cgi), Phytozome version 9.1 (http://www.phytozome.net/) and Genome Database for Rosaceae (GDR; http://www.rosaceae.org/) and paralogs were identified in potato, eggplant, pepper, cocoa, sesame, mimulus, watermelon, cucumber, grape, poplar, peach, and strawberry (E-value ≤ 1e-80). MEME analysis was conducted to identify the most likely orthologs in each species by selecting those with the most similar motif patterning proteins. Multiple sequence alignment and phylogeny tree construction were performed as the same method described above.
Three to thirty fruit from 10 fw11.3-WT and 10 fw11.3-D NIL plants each were collected between 1:00 pm to 3:00 pm and dissected tissues were immediately frozen in liquid nitrogen. The three tissues dissected were pericarp, columella, and developing seeds with placenta at the following developmental stages: 4, 7, 10, 15, 25, 33 DPA and turning stage fruit. For 4 DPA fruit, the columella, placenta and developing seeds were collected together. Most samples consisted of four replicates, except 7 DPA and 25 DPA (three replicates), 33 DPA (one replicate) and turning stage (two replicates). The low number of replicates for the latter tissues was due to severe incidence of blossom-end rot in the greenhouse.
Hot borate RNA extraction method [63] was used for total RNA extraction from 25, 33 DPA and turning stage fruit. The TRIzol (Invitrogen, Carlsbad, CA) RNA extraction method was used for 4, 7, 10 and 15 DPA fruit following the manufacturer’s specifications. RNA quantity and quality were assessed using a Qubit 2.0 fluorometer RNA Assay Kit (Invitrogen Inc. USA) and an Agilent 2100 Bioanalyzer RNA 6000 Nano kit (Agilent, USA). Approximately 5μg total RNA was used to prepare strand-specific libraries of approximately 250bp fragments [64, 65]. Libraries were barcoded and 8 samples were pooled per lane on the flowcell. Fifty-one bp single-end reads were generated on the Illumina HiSeq2000 at Genomics Resources Core Facility at Weill Medical College (New York, NY).
The pre-processing, read alignment and quantification of gene levels were performed using previously established protocols in the lab [64] with minor modifications. Briefly, ribosomal RNA-free reads were mapped to the Solanum lycopersicum reference genome (Build SL2.50) using tomato gene model annotation (ITAG2.4 release) to facilitate mapping reads across exon-exon junctions. The final expression data were shown as reads per kilobase of exon model per million mapped reads (RPKM). For the expression of selected genes at different developmental stages, the average RPKM were used. Summary statistics for each of the RNA-seq libraries are shown in S8 Table.
The correlations among samples were evaluated because the results would address reproducibility among samples (S9 Table). The columella tissue at the turning stage (TCol) showed low correlation between the two replicates (64% for fw11.3-WT and 85% for fw11.3-D, respectively). After ruling out the possibility of fw11.3-D or fw11.3-WT sample switch by evaluating the SNPs at fw11.3, we found that unlike TCol_rep1, TCol_rep2 was more correlated with other turning stage tissue types (pericarp and seed/placenta) than TCol_rep1. This suggested mixed up tissue samples for TCol_rep2. We therefore decided to discard TCol_rep2 and only use TCol_rep1 to represent turning stage columella tissue.
The gene expression data used for co-expression cluster analysis was the gene expression RPKM values of three fruit tissue types at seven developmental stages (six stages for columella tissue). The identification of co-expressed genes with CSR was done separately with CSR-D and CSR-WT alleles. The data was first pre-processed by Mfuzz [66] for normalization. The heatmap.2 function in R was then employed to generate a heatmap based on the normalized RPKM. Twelve clusters for CSR-D and 10 clusters for CSR-WT were identified visually based on the heatmap results. Fuzzy C-means clustering in Mfuzz was applied with cluster number as 12 and 10 (CSR-D and CSR-WT, respectively) and default settings. Soft clustering was chosen in “visualization” to generate clusters. Finally, the clustering results with the probabilities of each gene in each cluster were exported to Excel. Genes with the probabilities below 90% were removed from the clusters.
Arabidopsis ortholog genes were obtained for CSR-D and CSR-WT co-expressed overlapping genes by using BLASTP against TAIR10 amino acid sequence (p-value ≤ 7.00E-06). Cytoscape plug-in ClueGO (Version 2.3.2) [67] was used to perform the Gene Ontology (GO) analysis using the GO biological process available on November 17 2016. The ClueGO networks were set to ‘medium’ and their connectivity was based on a kappa score of 0.4. GO Term grouping was selected with an initial group size of 1 and group merging set at 50%. Two-sided hypergeometric tests were applied and p-value correction was carried out using the Bonferroni step-down method. GO terms with adjusted p ≤ 0.05 were considered as significant.
Sequence data from this article can be found in the EMBL/GenBank data libraries under accession number SRP017242, SRP089936, SRP089970.
The raw FASTQ files for the RNA-seq libraries were deposited at NCBI Sequence Read Archive (SRA) with SRA study accession SRP089936. Gene expression data (RPKM) are available through a Tomato Functional Genomic Database (TFGD; http://ted.bti.cornell.edu/cgi-bin/TFGD/digital/home.cgi).
|
10.1371/journal.pcbi.1004903 | Canonical Cortical Circuit Model Explains Rivalry, Intermittent Rivalry, and Rivalry Memory | It has been shown that the same canonical cortical circuit model with mutual inhibition and a fatigue process can explain perceptual rivalry and other neurophysiological responses to a range of static stimuli. However, it has been proposed that this model cannot explain responses to dynamic inputs such as found in intermittent rivalry and rivalry memory, where maintenance of a percept when the stimulus is absent is required. This challenges the universality of the basic canonical cortical circuit. Here, we show that by including an overlooked realistic small nonspecific background neural activity, the same basic model can reproduce intermittent rivalry and rivalry memory without compromising static rivalry and other cortical phenomena. The background activity induces a mutual-inhibition mechanism for short-term memory, which is robust to noise and where fine-tuning of recurrent excitation or inclusion of sub-threshold currents or synaptic facilitation is unnecessary. We prove existence conditions for the mechanism and show that it can explain experimental results from the quartet apparent motion illusion, which is a prototypical intermittent rivalry stimulus.
| When the brain is presented with an ambiguous stimulus like the Necker cube or what is known as the quartet illusion, the perception will alternate or rival between the possible interpretations. There are neurons in the brain whose activity is correlated with the perception and not the stimulus. Hence, perceptual rivalry provides a unique probe of cortical function and could possibly serve as a diagnostic tool for cognitive disorders such as autism. A mathematical model based on the known biology of the brain has been developed to account for perceptual rivalry when the stimulus is static. The basic model also accounts for other neural responses to stimuli that do not elicit rivalry. However, these models cannot explain illusions where the stimulus is intermittently switched on and off and the same perception returns after an off period because there is no built-in mechanism to hold the memory. Here, we show that the inclusion of experimentally observed low-level background neural activity is sufficient to explain rivalry for static inputs, and rivalry for intermittent inputs. We validate the model with new experiments.
| Perceptual rivalry is the subjective experience of alternations between competing percepts when an individual is presented with an ambiguous stimulus. It has been suggested that rivalry between neurons may be a ubiquitous property and competition between neural representations occurs throughout the brain [1]. Evidence for this includes: 1) neural correlates of visual rivalry found at multiple regions of visual processing (e.g., V1, V4, MT, IT) [2–5], 2) most sensory modalities exhibit rivalry suggesting similar biophysical processes [6], 3) brain regions that do not receive specific sensory inputs, such as the inferior frontal and parietal lobe, exhibit activity that is correlated with perceptual switches [7], and 4) independent rivalry can occur at different spatial locations, with different senses, and under different modalities [6,8–10]. Rivalry provides a unique window into cortical function because it directly accesses an internal computation elicited by but distinct from an external sensory input. Many experiments on rivalry utilize stimuli that are always in view although possibly moving. We refer to these situations as static rivalry. In intermittent rivalry, the stimulus presentation is periodically removed while the perception alternates with a longer period. Rivalry memory refers to the return of the last dominant percept after an extended time duration without stimulus.
Computational and mathematical work has shown that a neurophysiologically-constrained cortical model whose primary features are mutual inhibition between pools of neurons and a fatigue process can reproduce the basic experimental properties of static rivalry, winner-take-all behavior, normalization, and decaying oscillations induced by distractors under different input conditions [11–14]. These findings support the universality of a simple neuronal model that can be used to explain a myriad of basic cortical behaviors and as such is a candidate canonical circuit for simple cognitive functions (e.g. see [15–17] for discussion of such functions). A canonical circuit is universal in the sense that it does not imply a single function but can exhibit multiple operating regimes under a change in parameter values (including just a change in the type of stimulus). The biophysical dynamics can be inferred from behavioral observations such as rivalry, decision-making, working memory, and contrast-sensitivity [1,6,13,18] making them potential clinical endophenotypes (simple biomarkers associated with and possibly underlying a more complex trait of interest). As such, rivalry has been suggested as a diagnostic tool for probing cognitive dysfunction such as in autism, bipolar disorder, and major depressive disorder [19–21].
The canonical circuit and similar models have been validated against a set of nontrivial observations in static rivalry including Levelt’s propositions [11,22–26]. Although mutual inhibition acting as a positive feedback has been proposed as a memory mechanism in other contexts [27], it has been argued [22,28] that a circuit with mutual inhibition and fatigue cannot reproduce intermittent rivalry nor rivalry memory because presumably there is no mutual inhibition when the stimulus is absent and the fatigue process would give the opposite result of what is observed, favoring the suppressed percept (i.e. masking not priming). The question then is whether a single circuit model can account for all of the aforementioned behaviors (including non-rivalry dynamics) or whether independent or additional circuits are required to explain all the phenomena.
Here, we show that the same cortical circuit model can also be applied to time-varying inputs in intermittent rivalry and rivalry memory. We explore various possible mechanisms and show that just the inclusion of a previously neglected phenomenon—namely low, nonspecific, background activity—is sufficient to provide a unified account of static rivalry, intermittent rivalry, rivalry memory and all previously accounted for phenomena. The background activity at arbitrarily low values can provide sufficient drive to the neurons such that the memory can be held by mutual inhibition alone and withstand the effects of fatigue. The memory persists over a wide range of parameter values, is robust to noise, and does not require the inclusion of another process. The memory is topological in the sense that the total number of memory states is invariant to a range of changes to the input and network properties. Importantly this memory preserves all of the previously known behaviors of the basic model since the background activity does not play a role under the static conditions. This is in contrast to recurrent excitation models of short-term memory where neural activity is bistable between an active and inactive state, which is generally antagonistic to rivalry [29], may conflict with local balanced states [30], and relies on fine-tuning [31].
The circuit model can reproduce two main features of rivalry memory and intermittent rivalry that we observed in the ambiguous quartet illusion: 1) a dynamic Levelt’s fourth proposition (dynamic L4), where the dominance duration increases with increasing stimuli presentation intervals and 2) habituation, where the initial percept durations (epochs) decrease upon repeated stimulus presentations until a steady state is reached after several epochs. Dynamic L4 has been shown in prior work with other intermittent paradigms [32]. Habituation has also been reported for rivalry; although, not necessarily intermittent rivalry [33–35]. We provide a theoretical explanation for these behaviors. The theory suggests that for the observed habituation to occur, local fatigue such as spike-frequency adaptation or recurrent synaptic depression is the dominant form of fatigue rather than synaptic depression between pools, and that the interval between pulses is shorter than the fatigue time constant.
We informed our model with experiments on the ambiguous quartet illusion, which is a prototypical example of intermittent rivalry and has the advantage of naturally incorporating a periodic stimulus (see Fig 1). The inputs are time dependent and the perceived motion only occurs when the frame presentation (input) is switched from one parity (e.g. dots in upper-right and lower-left corners (UR-LL)) to the other (e.g. lower-right and upper-left corners (LR-UL)). The motion is ambiguous because the transition has two possible interpretations. For example, in the transition from UR-LL to LR-UL, the dot located at UR could be interpreted as “sliding” vertically down to LR, while the dot at LL simultaneously slides up to UL. The alternative interpretation is that the dot at UR slides horizontally leftwards to UL, while the other dot at LL slides horizontally rightward to LR. If one imagines a bar connecting the dots then the vertical interpretation corresponds to a bar rotating clockwise from 45° to –45° (like a seesaw) while the horizontal interpretation corresponds to the stick rotating counterclockwise from 45° to 135°. Hence, the perceived motion in the quartet illusion is characterized by two degrees of freedom—orientation and direction. Orientation refers to whether the dot motion is aligned along the vertical (V) or horizontal (H) axis and direction refers to whether the motion is clockwise (+) or counterclockwise (-) (see Fig 2a).
We presented the illusion to subjects with different frame presentation intervals (Tframe). We examined dynamic L4 by testing if the perception dominance duration (TD) (inverse of the percept alternation rate) was correlated with Tframe. In our first experiment we tested two subjects across a wide range of Tframe inputs. The subjects showed a dramatic response to the stimulus parameter, which occurred at different ranges for each subject (different sensitivities). In Fig 3a and 3b we plotted the TD distributions to give the qualitative picture. We also observed some evidence of a habituation effect (Fig 3c), where the dominance durations decreased to a steady state value.
To verify the phenomenon we recorded dominance durations on a group of 16 hypothesis-naive subjects. We probed the system over a range of Tframe: 300, 350, 400, and 450 milliseconds (minimum chosen to avoid photosensitive epilepsy). Individuals were told to report on any change of motion and were trained on biased H and V motion stimuli prior to the quartet stimulus (see Methods). From our post-task survey we found that all individuals observed oscillating V motion (seesaw) and H motion (seesaw rotated by 90 degrees). For a majority of subjects, these were the only reported percepts but a subset occasionally observed rotation (H, V, H,…). There were also infrequent reports of the dots disappearing or changing size and the dots moving toward and away from the subject (three-dimensional motion). The stimulus used in the experiments can be seen in S1 Movie.
To test for the presence of dynamic L4, we analyzed the pooled data across subjects and across percept-types (H, V, rotation). We observed a weak effect across subjects. The qualitative results are shown by the TD distributions and their means (see Fig 4a and 4b). We quantified the relationship with a Cox proportional hazards mixed effects model. The statistical model treats the dominance time as the survival time for a percept and evaluates the effects of factors on the survival probability (see Fig 4c). There was a significant but weak positive relationship of TD with Tframe with P = 2 × 10−4 (Wald test using 1,322 samples and 1,266 recorded switch events) and hazard ratio of 0.996 (s.e. 0.001). We used these preliminary findings together with previously reported effects for another rivalrous stimulus [32] as support for the dynamic L4 constraint.
For the hypothesis-naive experiment we discovered a larger habituation effect of the percept durations over percept epochs. The initial percept epochs had longer TD as shown by Fig 5, and we estimated a hazard ratio of 1.024 (s.e. = 0.008) (P = 2 × 10−3; Wald test using 1,322 samples and 1,266 recorded switch events). We used this habituation effect as a second constraint for the model.
We examined several intermittent rivalry and rivalry memory mechanisms in a canonical cortical circuit model and concluded that the inclusion of experimentally observed low-level, nonspecific background neuronal activity that induces a mutual-inhibition memory is the most plausible. The mechanism preserves all the phenomena of static rivalry as shown in previous studies [11,12,22]. Mutual inhibition has been proposed before as a mechanism for short-term memory [27]. However, it was not considered for rivalry since it was assumed that the circuit could not maintain the memory in the absence of a stimulus and the presence of fatigue would favor the suppressed percept. Even if neurons are not completely inactive during the off-states, it is not a priori obvious that an asymmetric state would exist and persist indefinitely in the presence of fatigue. We show that for reasonable conditions, this form of memory is topological (holds for a wide range of parameters) and stable for arbitrarily low input. Topological mutual-inhibition memory can exist as long as inhibition is strong enough and the gain function (i.e., frequency-input curve) has a non-increasing slope, which is experimentally observed [36,49]. Thus, it may not be limited to rivalry and could be the mechanism for other forms of memory in the brain.
The model explains the two intermittent rivalry phenomena of dynamic L4 and habituation in the quartet illusion. Dynamic L4 is similar to what Leopold et al. (2002) found for the rotating sphere (RS) stimulus using manually inserted blank periods. A habituation effect similar to what we found for the quartet has also been noted for other stimuli (e.g., static images, motion) and sensory domains (e.g., visual, auditory) [33–35]. However, binocular rivalry shows the opposite effect where the switch rate decreases over epochs. This has been attributed to contrast adaptation [50] and can be accounted for in the rivalry model by reducing input over time. The dynamics of our model are geometrically similar to the model proposed by Noest et al. 2007[28]. However, instead of requiring a subthreshold current with shunting adaptation, we show that the activity itself acts as the memory with the property of persisting indefinitely. There are many possible mechanisms for the background activity and perhaps the most interesting is that zero-mean noise is sufficient. While either local (e.g., spike-frequency adaptation, recurrent synaptic depression) or nonlocal (e.g., cross-pool synaptic depression) is sufficient to explain intermittent rivalry, local fatigue in particular reproduces habituation.
Prior experimental designs of intermittent rivalry introduced the off-states by either manually removing a static stimulus or by utilizing a more complex protocol such as combining motion-induced blindness with a static stimulus, or by instructing the subject to independently attend to mixed “static, intermittent” epochs[10,32]. It is possible that some of these more complex paradigms involve higher order processing that makes replication difficult. We propose that the quartet is a robust self-contained system for the study of intermittent rivalry since periodicity in the quartet invokes natural motion processing. In retrospect, our experimental design was suboptimal for detecting dynamic L4, but the model provides guidance for future experiments. Further study is necessary to determine whether different fatigue modes are important for different rivalry stimuli. This could be predicted with the model and tested experimentally by behavioral and electrophysiological measurements, perhaps combined with pharmacological manipulation of K+-channels that modulate adaptation effects [51]. Finally, the role of background activity in the theory predicts that perturbing or suppressing the activity bias will extinguish rivalry memory.
Ethics approval for this study was granted by the NIH Combined Neuroscience Institutional Review Board under protocol number 10-M-0027 (ClinicalTrials.gov ID NCT01031407).
The naive-subject study consisted of 16 adult male participants from the National Institutes of Health campus and DC metropolitan region. The data were screened such that at least two percept switches were reported in a test block. One subject did not have adequate data and was removed, leaving 15 subjects. Author data was collected on SV and PT for the quartet.
Naive-subjects were tested on the standard quartet task. The standard quartet animation consists of two alternating still frames where a single frame consists of one set of dots located at opposing corners of an upright square as shown in Fig 1. The task was administered in a dark room with maximum screen and keyboard brightness (approximately 2 lux) on a 15 inch 2010 MacBook Pro with 60 Hz frame rate. Our stimulus consisted of three nested quartets centered on the viewing screen as shown in Fig 1 (also see S1 Movie); chosen to mitigate the Troxler effect and to induce motion perception over a large receptive field. The dots were outlined with the RGB color = 46, 55, 254 and set on a black background. For a typical viewing distance of 24 inches, although this was not rigidly enforced, the square length visual angle (dot diameter visual angle) of each nested quartet were 5.5 (1), 3.5 (0.5), 1.5 (0.25) degrees. There was a fixed orange, central dot with RGB color = 255,127,0. This was the instructed focal point for the subject. These parameters were arbitrarily chosen. During the task, subjects were instructed to maintain fixation on this central dot and report a change in motion by pressing the Return-key. To give feedback that a switch was recorded, the central dot briefly changed color from orange to green (RGB = 0,255,0) to orange when the subject reported a change.
The task consisted of 7 blocks: a tutorial, a task-comprehension block, and 5 frame-period (Tframe) blocks with a two second black screen period between blocks. Participants were told that the task would test their ability to detect changes in motion and were not informed that the motion was ambiguous. They were also shown examples of the types of motions to expect: horizontal or vertical, consisting of unbiased apparent motion animations, where a set of three frames was repeatedly presented such that the dots progressed from corner, to center, to corner, to center. This was done for vertical and horizontal orientations of motion. The task-comprehension block used this same biased motion stimulus and the direction was changed at fixed intervals that were known to the experimenter. The first frame-period block was set to 300 millisecond frame intervals and lasted six minutes (we refer to this as the practice block). This was followed by four blocks with frame periods (and block durations) of 300 (5), 350 (5), 400 (6), and 450 (8) ms (minutes). The order of the last four blocks was shuffled between participants.
Author-data was collected on the quartet task. The quartet task was similar to the above but without the tutorial, task-comprehension, or practice blocks. These sessions consisted of randomized Tframe with 2–3 sampled frame periods and the duration was based on reporting 11 switches (some Tframe were repeated in separate sessions).
Data from the naive-subject group was analyzed using a Cox proportional hazards model (see [52,53] for review). It is a semi-parametric model that is robust to the distribution of the data, which was non-normal for our case. The analysis calculates statistics across all time points and accounts for right-censored data. We chose this approach since it accounts for percept duration that may extend beyond the test block, including those that may span the majority of a block. The frequency of percepts that remained dominant after TD dotted frame presentations were fit to a baseline survival model for each test condition (e.g., Tframe) and normalized (the non-parametric component of the model). The hazard ratio is the exponential adjustment to the baseline survival for the conditioned effects, which is the parametric component. An increased hazard ratio means there was an increased probability of percept extinction over a time period as the parameter was increased. To be considered a valid model fit, there were two statistical tests: 1) test whether the estimated coefficient for the effect is non-zero, and 2) that the distribution of the residuals has zero slope across time (is flat).
We used the coxme mixed effects R function to estimate the hazard ratios. The mixed effects model was used so we could account for subject- or Tframe- specific deviations. A Wald test was used to estimate a P-value for the hazard ratio. The coxme library does not have a test for the residuals. To check this we calculated a noise vector (z) from the coxme fit given by z = ŷ − θx, where ŷ is the linear model prediction, θ is the estimated coefficient, and x is the predictor vector. The noise vector z was used as an offset in the R function coxph and the residuals from this were checked with cox.zph. We used a criterion of no significant deviation of the residual slope from zero (i.e., P>0.05), in order for the proportionality assumption to apply.
Bifurcation analysis was performed in XPPAUT [54]. Rate models were simulated using Python. Conductance-based model simulations were performed with circuit of two mutually inhibiting neuron pools with a calcium activated spike frequency adaptation-like current and synaptic depression. This model was simulated in XPPAUT and the output was analyzed in Python. All code used for simulations and numerical analysis including parameters are attached as S2 Text.
|
10.1371/journal.ppat.1003054 | The Initial Draining Lymph Node Primes the Bulk of the CD8 T Cell Response and Influences Memory T Cell Trafficking after a Systemic Viral Infection | Lymphocytic choriomeningitis virus (LCMV) causes a systemic infection in mice with virus replication occurring in both peripheral tissues and secondary lymphoid organs. Because of the rapid systemic dissemination of the virus, the secondary lymphoid organs responsible for the induction of the LCMV-specific CD8 T cell response are poorly defined. We show that the mediastinal lymph node (MedLN) serves as the primary draining lymph node following LCMV infection. In addition, we demonstrate that the MedLN is responsible for priming the majority of the virus-specific CD8 T cell response. Following resolution of the acute infection, the draining MedLN exhibits characteristics of a reactive lymph node including an increased presence of germinal center B cells and increased cellularity for up to 60 days post-infection. Furthermore, the reactive MedLN harbors an increased frequency of CD62L− effector memory CD8 T cells as compared to the non-draining lymph nodes. The accumulation of LCMV-specific CD62L− memory CD8 T cells in the MedLN is independent of residual antigen and is not a unique feature of the MedLN as footpad infection with LCMV leads to a similar increase of virus-specific CD62L− effector memory CD8 T cells in the draining popliteal lymph node. Our results indicate that CD62L− effector memory CD8 T cells are granted preferential access into the draining lymph nodes for an extended time following resolution of an infection.
| CD8 T cells are required for the elimination of infected host cells following an acute virus infection. In addition, memory CD8 T cells provide immunity to the host against a secondary infection. Much is known about the priming of CD8 T cells towards viruses that induce a localized infection, however the site responsible for priming the majority of CD8 T cells following a systemic viral infection remains unclear. Lymphocytic choriomeningitis virus (LCMV) induces an acute systemic viral infection when inoculated intraperitoneally, eliciting a robust CD8 T cell response. Although intraperitoneal LCMV infection results in rapid systemic viral replication, we demonstrate that the mediastinal lymph node (MedLN) serves as the initial draining lymph node and represents the primary site for the induction of the acute CD8 T cell response. In addition, we observe that CD62L− effector memory CD8 T cells are preferentially recruited into the draining MedLN for up to 60 days following LCMV infection. Collectively, these studies indicate that the draining lymph node remains poised to defend the host against a secondary encounter with a pathogen for a prolonged time following the primary infection.
| Lymph nodes (LN) play a critical role in initiating the adaptive immune response following viral infections. For example, intravenous (i.v.) vesicular stomatitis virus infection of splenectomized (SplnX) mice yields a similar number of virus-specific CD8 T cells as control mice. In contrast, vesicular stomatitis virus infection of lymphotoxin-α-deficient knockout (LT-α-KO) mice that lack LNs results in a significant decrease in the total number of virus-specific CD8 T cells [1]. Similarly, intraperitoneal (i.p.) lymphocytic choriomeningitis virus (LCMV) infection of LT-α-KO mice results in a decrease in the total number of virus-specific CD8 T cells in the spleen [2]. Taken together, these data suggest that virus-specific CD8 T cell responses are initiated in LNs following systemic viral infection. However, it is currently unclear which LNs are primarily responsible for initiating the virus-specific CD8 T cell response following a systemic viral infection. In addition, it is currently unknown how events that occur during induction of the CD8 T cell response affect the distribution of antigen-specific memory CD8 T cells in the draining LN following resolution of the infection.
CD8 T cell entry into LNs is dependent on their differentiation status. Naive CD8 T cells express high cell surface levels of both CD62L and CCR7 [3]. The combined expression of these two molecules facilitates CD8 T cell entry into LNs via binding to peripheral node addressin and CCL21, respectively, in the high endothelial venules [3]. Upon activation, naïve CD8 T cells rapidly proliferate and downregulate expression of CD62L. The loss of CD62L expression combined with the upregulation of new adhesion molecules and chemokine receptors facilitates the trafficking of effector CD8 T cells into peripheral tissues [4]. Following pathogen clearance, CD8 T cells undergo contraction and two major subsets of memory CD8 T cells remain: CD62L− effector memory CD8 T cells and CD62L+ central memory CD8 T cells. Effector memory CD8 T cells resemble effector CD8 T cells as they lack expression of CD62L and traffic primarily to peripheral tissues. In contrast, central memory CD8 T cells regain expression of CD62L and more efficiently enter the LNs as compared to either effector or effector-memory CD8 T cells [3], [5]. Furthermore, the lack of CD62L cell surface expression on memory CD8 T cells results in ∼90% decrease in their capacity to migrate into peripheral LNs, suggesting that CD62L expression is necessary for efficient entry of memory CD8 T cells into LNs [6].
We demonstrate that the mediastinal LN (MedLN) serves as the primary draining LN following an i.p. LCMV infection. Furthermore, we show that the majority of the LCMV-specific CD8 T cell response is primed in the MedLN, despite other LNs and the spleen acquiring viral antigens during the course of systemic viral spread. These data suggest that the initial draining LN plays a critical role in initiating the virus-specific CD8 T cell response. In addition, we demonstrate that the majority of LCMV-specific memory CD8 T cells in the MedLN are CD62L− for up to 60 days post-infection (p.i). This accumulation of LCMV-specific CD62L− memory CD8 T cells in the MedLN positively correlates with the presence of a sustained germinal center response and increased LN cellularity. Importantly, we demonstrate that the presence of CD62L− effector memory CD8 T cells in the draining LN is not due to the presence of residual virus-derived antigens, but instead due to the preferential recruitment of CD62L− effector memory CD8 T cells. Taken together, these data suggest that very early events that occur during a systemic viral infection profoundly alter the long-term trafficking of virus-specific memory CD8 T cells.
LCMV infection of mice leads to systemic viral spread with almost all organs supporting virus replication [7]. Previous studies have shown that LT-α-KO mice that lack peripheral LNs exhibit a ∼5-fold decreased LCMV-specific CD8 T cell response as compared to wild-type (WT) mice, suggesting that the majority of the LCMV-specific CD8 T cell response is primed in the LNs [2]. However, LT-α-KO mice also exhibit alteration in splenic architecture [8], thus not completely ruling out a role for the spleen in priming the LCMV-specific CD8 T cell response. Therefore, to confirm the role of the LNs vs. the spleen in priming the LCMV-specific CD8 T cell response, WT, LT-α-KO as well as SplnX mice were infected i.p. with LCMV. Organs were harvested at day 8 p.i. and the total number of LCMV-specific CD8 T cells was assessed by intracellular cytokine staining (ICS) for IFN-γ. Consistent with previous studies [2], Figure 1 demonstrates that LT-α-KO mice exhibit a >10-fold decrease in the total number of LCMV glycoprotein (GP)33-specific (Figure 1A), nucleoprotein (NP)396-specific (Figure 1B) and GP276-specific (Figure 1C) CD8 T cells in the spleen, lung and liver as compared to their WT counterparts. A similar decrease in the frequency of LCMV-specific CD8 T cells was observed in the peripheral blood (Figure 1D). In contrast, there were similar total numbers of LCMV-specific CD8 T cells in the lung, liver and peripheral blood (Figure 1) in SplnX mice as compared to WT mice. Furthermore, there was a significantly (p<0.05) greater total number of CD8 T cells of all specificities in the mesenteric LN (MesLN) and MedLN in SplnX mice as compared to WT mice. These data indicate that the spleen is not required to mount a LCMV-specific CD8 T cell response whereas the LNs are necessary for the efficient priming of the CD8 T cell response.
It is currently unclear which specific LN is responsible for initiating the virus-specific CD8 T cell response following an i.p. LCMV infection. A previous study demonstrated that soluble antigens, bacteria, or dyes administered i.p. all drained into the MedLN [9]. Therefore, we hypothesized that the MedLN would serve as the draining LN following an i.p. LCMV infection. Mice were infected i.p. with LCMV and viral titers in the spleen, MedLN, inguinal LN (ILN), MesLN and cervical LN (CLN) were assessed by plaque assay. Figure 2 shows that there is significantly (p<0.05) more virus in the MedLN than either the spleen or MesLN at 12 or 24 hours (h) p.i.. Furthermore, there was no virus detected at either 12 or 24 h p.i. in either the ILN or the CLN. However, by 48 h p.i., all tissues examined contained detectable levels of virus. The viral titers in the MedLN peaked at 48 h p.i. and started to decline by 72 h p.i., whereas viral loads in other tissues had either plateaued (i.e. spleen) or continued to increase (i.e. ILN, MesLN and CLN) until 96 h p.i.. Taken together, these data suggest that i.p. infection with LCMV leads to an initial infection of cells within the MedLN.
The above results indicate that following systemic LCMV infection the infectious virus drains first to the MedLN. We next sought to determine if the presence of infectious virus in the MedLN early following an i.p. infection correlated with initial CD8 T cell priming in the MedLN. We adoptively transferred 2×106 carboxyfluorescein succinimidyl ester (CFSE)-labeled LCMV-specific T cell receptor transgenic P14 CD8 T cells into naïve mice prior to i.p. LCMV infection. At various times p.i., P14 CD8 T cells in the spleen and LNs were monitored for increased CD25 expression as well as proliferation via CFSE dilution. Figures 3A and 3B demonstrate that P14 CD8 T cells upregulate CD25 expression in the MedLN as early as 12 h p.i.. In contrast, we did not observe substantial upregulation of CD25 on P14 CD8 T cells in the spleen, ILN or MesLN until 48 h p.i. (Figure 3A, B). Furthermore, CD8 T cell proliferation occurred initially in the MedLN at 48 h p.i., followed by the spleen, ILN and MesLN at 72 h p.i. (Figure 3A, B). By 96 h p.i. all P14 CD8 T cells in each of the organs examined had proliferated (Figure 3A, B). In addition, downregulation of CD62L and upregulation of CD43glyco on P14 CD8 T cells occurred first in the MedLN at 24 h and 48 h p.i., respectively (Figure 3B).
The above results suggest that LCMV-derived antigens are displayed first to naïve CD8 T cells in the MedLN following an i.p. LCMV infection. To examine the role of the MedLN as compared to the non-draining LNs and the spleen in priming the LCMV-specific CD8 T cell response, we adoptively transferred a physiological number (i.e. 2,000) of P14 CD8 T cells into naïve mice one day prior to infection. Starting 24 h p.i., the mice were treated daily with either vehicle (i.e. H2O) or the S1P receptor agonist FTY720 to trap LCMV-specific CD8 T cells in the LNs. Figure 4 demonstrates that at day 5 p.i. there is a decrease in the frequency (Figure 4A) and a significant decrease (p<0.05) in the total number (Figure 4B) of P14 CD8 T cells in the spleen and ILN in FTY720-treated mice as compared to control mice treated with vehicle. There were a similar total number of P14 CD8 T cells in the MesLN in both the control and FTY720-treated mice, suggesting that the MesLN may induce a small proportion of the LCMV-specific CD8 T cell response consistent with the low level of virus in the MesLN early following infection. In contrast to the spleen and ILN, there was a significant increase (p<0.05) in the frequency and total number of P14 CD8 T cells in the MedLN of FTY720-treated mice as compared to vehicle-treated control mice (Figure 4). Furthermore, there was a significantly greater (p<0.05) number of P14 CD8 T cells in the MedLN of FTY720-treated mice as compared to the spleen of FTY720-treated mice. These data suggest that the initiation of the virus-specific CD8 T cell response occurs primarily in the MedLN early following i.p. LCMV infection.
Following localized immunizations, the draining LN can exhibit altered characteristics such as the presence of germinal center (GC) B cells and increased overall cellularity [10]–[12] identifying it as a reactive LN. Given that the majority of the CD8 T cell response is primed in the MedLN following a systemic LCMV infection, we hypothesized that the MedLN would exhibit a “reactive” phenotype. To test this hypothesis, we examined the presence of GC B cells as a measure of LN reactivity following LCMV infection. The draining MedLN exhibited a significantly increased (p<0.05) frequency of GC B cells at days 15 and 34 p.i. as compared to the non-draining LNs (i.e. ILN, CLN and MesLN) and the spleen (Figure 5A, B). However, by day 60 (Figure 5B) and >400 p.i. (data not shown) the frequency of GC B cells was similar between all LNs examined. Furthermore, the frequency of GC B cells was significantly greater (p<0.05) at day 34 p.i. only in the MedLN as compared to the corresponding LN in naïve mice (Figure 5C). Additionally, reactive LNs exhibit a prolonged increase in overall cellularity as compared to naïve LNs [12]. Figure 5D demonstrates that there were a greater (p<0.05) total number of cells in day 34 p.i. MedLN relative to the MedLN from naïve mice. In contrast, there was no difference (p>0.05) in the total cell number between naïve ILN, MesLN or CLN as compared to the same LNs obtained from mice infected 34 days prior with LCMV (Figure 5D). These data demonstrate that following an acute systemic viral infection, the initial draining LN remains “chronically” reactive for an extended period of time.
Previous work has shown that reactive and non-reactive LNs differ in their capacity to attract both memory CD4 and CD8 T cells [12], [13]. Therefore, based on our results demonstrating that the MedLN remains “chronically” reactive following an i.p. LCMV infection, we questioned if the trafficking of memory CD8 T cells into the MedLN would be altered. Given the importance of CD62L for entry of T cells into LNs, we compared the expression of CD62L on P14 cells in the MedLN vs. other LNs. The reactive MedLN exhibited a reduced frequency of CD62L+ P14 CD8 T cells at days 15, 34 and 62 p.i. as compared to the non-reactive LNs (i.e. ILN and CLN) (Figure 6A). However, by day >400, all LNs exhibited a similar frequency of CD62L+ P14 CD8 T cells. The chemokine receptor CCR7 represents another important molecule involved in T cell entry into the LN. In contrast to CD62L, the frequency of CCR7+ P14 CD8 T cells was similar between the MedLN and the other LNs at day 34 p.i. (Figure S1). In addition, the frequency of CD62L+ P14 CD8 T cells in the MesLN, a LN that has been previously shown to utilize α4β7 in addition to CD62L for CD8 T cell entry [14]–[16], was similar to that of the MedLN at all time points examined (Figure 6A, B). Consistent with a role for α4β7 in facilitating entry of T cells into the MesLN, we observed an increased frequency of β7+ P14s in the MesLN as compared to the MedLN, ILN and CLN (Figure S2). Although the majority of P14 CD8 T cells in the MedLN were CD62L− for ∼60 days following infection, the expression levels of two other memory-associated molecules (e.g. CD127hi and KLRG-1lo) were remarkably similar between P14 cells in the MedLN as compared to the P14 cells in the ILN and CLN at virtually all times following LCMV infection (Figure 6C, D). These data argue against retention of effector CD8 T cells in the MedLN long-term following viral infection and rather suggest that virus-specific memory CD8 T cells that enter the MedLN>15 days following an i.p. infection with LCMV do not require CD62L.
One potential explanation for the decreased frequency of CD62L+ P14 CD8 T cells in the MedLN as compared to either the ILN or the CLN is the presence of residual virus-derived antigen. Persistent antigen in the MedLN could cause reactivation of memory P14 cells resulting in either the down-regulation or the cleavage of CD62L. Work by Khanna et al. demonstrates that residual antigen is present within the MedLN for up to 30 days following an acute pulmonary influenza virus infection and that this antigen is capable of activating newly recruited CD8 T cells [17]. The reactive MedLN contained an increased frequency of CD25+ and CD69+ memory P14 cells as compared to the non-reactive LNs 34 days following LCMV infection (Figure S3A, B). This increased frequency of CD25+ and CD69+ memory P14 cells in the MedLN could be the result of continued antigen presentation in this LN due to the presence of persisting antigen. Therefore, to determine if residual antigen persists within the reactive MedLN, naïve P14 CD8 T cells were CFSE-labeled and subsequently transferred into day 34 LCMV-immune mice. MedLN, ILN, MesLN and CLNs were harvested 6 days post-transfer and the activation status of the transferred P14s was analyzed. The transferred P14s did not exhibit upregulation of the activation markers CD25 and CD69 following transfer (Figure 7A, B). Additionally, the transferred cells did not dilute CSFE expression or downregulate CD62L expression (Figure 7) indicating that no residual antigen is present within the MedLN (or within any other LN) of a mouse infected with LCMV 34 days prior. The above experiments were performed with naïve P14 CD8 T cells because previous data has suggested that transferred memory cells may not be able to fully access antigen-bearing dendritic cells (DCs) [17]. However, it is well established that memory CD8 T cells display an increased sensitivity to antigen as compared to naïve CD8 T cells [18], [19]. Therefore, to further test if residual antigen is present within the MedLN using memory cells, we harvested the MedLNs, ILNs and spleens from mice infected 34 days prior with LCMV. Mononuclear cells from these tissues were cultured in vitro with CFSE-labeled memory P14 CD8 T cells that had been isolated from the spleens of LCMV-immune mice. After 3 days in culture we examined the expression of CD62L and the dilution of CFSE on the memory P14 CD8 T cells. Memory P14 CD8 T cells did not divide nor downregulate cell surface expression of CD62L when cultured with cells isolated from the spleen, MedLN, CLN or ILN of either naïve or day 34 LCMV-infected mice (Figure 8 A–C). However, the memory P14 CD8 T cells did both divide and downregulate CD62L when cultured with tissue-derived mononuclear cells pulsed with GP33–41 peptide. As an additional test for the presence of residual antigen, we assessed the levels of the LCMV GP by RT-PCR. LCMV GP was readily detectable by RT-PCR in all LNs examined at day 4 p.i. In contrast, no detectable LCMV GP was present within the MedLN at day 34 p.i. as determined by RT-PCR (Figure 9). Taken together, these data suggest that residual LCMV-derived antigen is not responsible for the decreased frequency of CD62L+ P14 memory CD8 T cells in the MedLN as compared to the ILN or CLN.
Figure 6 demonstrated an increased frequency of CD62L− LCMV-specific memory CD8 T cells in the reactive MedLN as compared to the non-reactive LNs at day 34 following an i.p. LCMV infection. The MedLNs are similar to the MesLNs of the gut in that they both drain tissues that are constantly inflamed. The gut is constantly exposed to foreign antigen and is inhabited by commensal bacteria. The lung is similar in that it is also continuously exposed to foreign antigens. Thus, it is possible that these LNs share a similar CD62L-independent trafficking mechanism. To address this possibility, we infected mice via the footpad with LCMV. This route of infection is commonly used to direct antigens to the popliteal LN (PopLN) [20], [21]. Early after footpad infection, we examined both PopLNs as well as the spleen to ensure preferential infection of the ipsilateral PopLN (Ips PopLN). Figure 10A shows that at 24 h p.i., only the Ips PopLN contained replicating virus, indicating that the Ips PopLN is the initial draining LN. The spleen contained a small amount of replicating virus only at 48 h p.i. and we were unable to detect virus within the Con PopLN at any time points following footpad infection (Figure 10A). To determine if priming the LCMV-specific CD8 T cell response in the draining Ips PopLN resulted in chronic reactivity of this LN, we examined the presence of GC B cells and total cellularity at days 34–40 p.i.. Although the Ips PopLN did not exhibit a heightened/prolonged presence of GC B cells following footpad infection as compared to either the spleen or other LNs examined (Figure 10B), there were significantly (p<0.05) more total cells in the LCMV-immune Ips PopLN as compared to its naïve counterpart (p<0.05; Figure 10C) suggesting that the Ips PopLN is reactive. Next, we wanted to determine if a decreased frequency of CD62L+ LCMV-specific memory CD8 T cells is present in the reactive Ips-PopLN as compared to the non-reactive Con PopLN. Thirty-four to forty days following footpad infection, the Ips PopLN, Con PopLN, MesLN and spleen were harvested and examined for the presence of CD62L+ LCMV-specific memory CD8 T cells. Figure 10D shows a similar low frequency of CD62L+ LCMV-specific memory CD8 T cells in the Ips PopLN, MesLN and spleen. However, there was a significantly (p<0.05) higher frequency of CD62L+ LCMV-specific CD8 T cells in the Con PopLN (Figure 10D). These data suggest that the MedLN does not uniquely attract CD62L− CD8 T cells but rather, this is a property of draining LNs that initiate the virus-specific CD8 T cell response. Furthermore, these data suggest that the prolonged presence of GC B cells is not required for the preferential recruitment of CD62L− memory CD8 T cells.
Guarda et al. previously demonstrated that CD62L− effector and effector memory CD8 T cells accumulate in the reactive LN as compared to the non-reactive LNs following DC immunization in the footpad [13]. To determine if preferential recruitment of CD62L− effector memory CD8 T cells occurs following a systemic viral infection, sorted memory CFSE-labeled CD62L+ and unlabeled CD62L− P14 CD8 T cells were co-transferred into either naïve mice or day 34 LCMV-infected mice. Seventy-two h post-transfer, the reactive MedLNs and non-reactive ILNs and CLNs were harvested and the ratio of CFSE-labeled cells was examined. The CD62L− memory CD8 T cells (CFSE−) exhibited an enhanced capacity to enter the reactive MedLN as compared to the non-reactive LNs in a day 34 LCMV-infected mouse (Figure 11A and B). Importantly, the MedLN of naïve mice displayed a similar ratio of transferred CFSE+∶CFSE− cells as compared to the ILNs and the CLNs (Figure 11). These data indicate that in the setting of a systemic viral infection, the initial draining LN demonstrates an increased capacity to recruit CD62L− effector memory CD8 T cells for an extended time following resolution of the infection.
Much is known about the priming of CD8 T cell responses following localized infection by viral pathogens. In these studies, viral antigen is largely restricted to the infected tissue and some of this antigen is transported to the tissue-draining LN, either by DCs or through the lymph, to prime the virus-specific CD8 T cell response [22], [23]. However, the process by which this occurs following a systemic viral infection where viral antigen is not restricted to a single tissue or draining LN is currently unclear. Our data clearly demonstrates that although all of the LNs examined eventually harbor replicating virus, only the immediate draining MedLN was responsible for priming the majority of the virus-specific CD8 T cell response following an i.p. LCMV infection (Figures 3 and 4). In addition, the draining LN remained chronically reactive (Figure 5) and exhibited a profound impact on the trafficking of memory CD8 T cells, allowing the entry of CD62L− effector memory CD8 T cells (Figure 6). Taken together, these data suggest an intimate link between events that occur early during CD8 T cell priming in the draining LN and how this affects the entry of memory CD8 T cell subsets into the initial draining LN long-term following infection.
Several studies have demonstrated the presence of residual antigen in the priming LN following a localized infection by a number of viral pathogens [17], [24], [25]. For example, Khanna et al [17] showed that there was a high frequency of CD62L− virus-specific CD8 T cells in the MedLN 30 days following an intranasal influenza virus infection [17]. Their studies indicate that the high frequency of CD62L− CD8 T cells in the MedLN is due to long-term depots of influenza virus antigen that constantly stimulate CD8 T cells and cause the downregulation of CD62L. Consistent with this notion, transferred naïve, but not memory, influenza virus-specific CD8 T cells upregulate activation markers such as CD69 and PD-1 [17], [24]. These data indicated that the increased presence of CD62L− memory CD8 T cells in the draining MedLN following LCMV infection may be due to residual antigen within this LN. However, we did not observe any signs of activation (i.e. CFSE-dilution, CD25 upregulation, CD69 upregulation or CD62L downregulation) of naïve P14s upon transfer into day 34 LCMV-infected mice (Figure 7). Furthermore, we co-cultured LCMV-specific memory CD8 T cells with MedLN-derived single-cell suspensions to make antigen available to all cells [17] and we were unable to detect the presence of LCMV-derived antigens in the MedLN (Figure 8). These data suggest that neither the long-term reactivity of the MedLN nor the lack of CD62L expression on CD8 T cells is due to the persistence of viral antigen. However, this assay may not be sensitive enough to detect very low levels (<1 pM) of antigen on a small number of DCs or other antigen presenting cells and does not rule out very low-level persistence of antigen. To utilize a more sensitive assay for LCMV detection [26], [27], we utilized RT-PCR and were unable to detect any residual LCMV GP at day 34 p.i. (Figure 9). Thus taken together, our data indicates that residual LCMV-derived antigen is likely not responsible for either the long-term reactivity of the MedLN or the increased frequency of CD62L− effector memory CD8 T cells in the MedLN. Our results are consistent with a recent study by Takamura et al [28] demonstrating that mice infected i.p. with Sendai virus generate virus-specific memory CD8 T cells that are unable to recognize residual antigen in the draining LN as compared to CD8 T cells generated following an intranasal infection.
Another explanation for the increased frequency of CD62L− effector memory CD8 T cells in the draining LN could be due to the preferential recruitment of these cells. Studies using localized DC immunizations in the footpad demonstrated that transferred CD62L− effector and effector memory CD8 T cells entered the draining LN at an increased propensity as compared to the non-draining LN [13]. In concordance with the above study, we demonstrate that the increased frequency of CD62L− CD8 T cells in the draining LN following a systemic LCMV infection is due to the preferential trafficking of CD62L− effector memory CD8 T cells as compared to the non-draining LNs (Figure 11). These data indicate that although LCMV induces a systemic viral infection in which viral replication occurs in virtually all of the secondary LNs, only the initial draining LN remains reactive and allows preferential access for CD62L− effector memory CD8 T cells. However, the mechanism that accounts for the preferential recruitment of CD62L− effector memory CD8 T cells remains unclear. Guarda et al [13] demonstrated that both CD62L− effector and effector memory CD8 T cells utilize the chemokine receptor CXCR3 to enter reactive LNs [13]. Interestingly, we observed that greater than 90–95% of the LCMV-specific CD8 T cells in the MedLN express CXCR3 (data not shown). However, we do not observe any significant difference in the expression of the CXCR3 chemokine ligands CXCL9 and CXCL10 via either RT-PCR or ELISA (data not shown), suggesting that these cells may utilize other means of entry into the MedLN other than CXCR3. Martin-Fontecha et al demonstrated that CD62L− effector memory CD4 T cells could enter long-term/chronic reactive LNs in a CD62P/PSGL-1 (P-selectin glycoprotein ligand 1)-dependent manner [12]. Recent studies have also shown a role for PSGL-1 in the migration of activated T cells and other leukocytes into LNs [29], [30]. In preliminary experiments, we observed that greater than 90–95% of the LCMV-specific CD8 T cells in the MedLN expressed the CD62P ligand PSGL-1 (data not shown). Other studies report that the activated glycoform of CD43 (CD43glyco) can play a role in leukocyte adhesion to tissue endothelial cells that express E-selectin (CD62E) which in some scenarios can also be expressed on the high endothelial venules of reactive LNs [31]–[34]. We have also observed that LCMV-specific CD8 T cells in the MedLN express higher levels of CD43glyco as compared to LCMV-specific CD8 T cells in the peripheral blood or spleen (data not shown). These data suggest that LCMV-specific CD8 T cells may enter reactive LNs via a PSGL-1-or CD43-dependent manner.
Recent work has suggested that T cells can be “imprinted” upon priming to preferentially traffic to the tissue in which the antigens originated. For instance, CD8 T cells primed in either the gut draining LNs (e.g. MesLN) or the Peyer's Patches are imprinted with the gut homing integrin α4β7 [14], [15] whereas T cells primed in peripheral LNs do not express α4β7, but instead express the α4β1 integrin which plays a role in homing to other inflamed tissues (i.e. the skin and lung) [35]. It is possible that CD8 T cells primed in the MedLN are “imprinted” in this manner to preferentially return to the MedLN. Figure 3 demonstrates that LCMV-specific CD8 T cells in the MedLN 4 days following an i.p. infection with LCMV maintain a CD62L− phenotype. In comparison, LCMV-specific CD8 T cells in other LNs (i.e. the ILN and CLN) largely regain CD62L expression at day 4 post-LCMV infection. These data correlate with what we observed >30 days p.i. where there is a higher frequency of CD62L+ LCMV-specific CD8 T cells in the ILN and CLN as compared to the MedLN. Taken together, these data may provide evidence for “imprinting” of LCMV-specific CD8 T cells primed in the MedLN to return to the MedLN in a CD62L-independent manner following resolution of the infection. However, it is important to note that this imprinting is not due to expression of α4β7 as we observed a similar frequency of β7-expressing memory CD8 T cells within the draining MedLN as we do in the non-draining LNs (Figure S2).
It is unclear if there is an advantage to have CD62L− virus-specific effector memory CD8 T cells in the draining LN long-term after infection. Khanna et al [17] demonstrated that influenza-specific CD8 T cells in the MedLN express activation markers (i.e. CD69 and PD-1) that are often co-expressed with granzyme B after migration into inflamed tissues. Cells with primed effector function may be maintained in the draining LN long-term to provide a first line of defense against pathogens that replicate in secondary lymphoid organs in order to protect against secondary infection via the same route of infection. Figure 2 shows that the MedLN is the first place where replicating virus can be detected following an i.p. infection with LCMV. These data suggest that like other tissues, the draining LN may serve as a reservoir for effector-memory CD8 T cells to serve as local guardians against re-infection.
Overall, our data demonstrates that following a systemic viral infection, the vast majority of virus-specific CD8 T cells are primed within the initial draining LN. Furthermore, our data demonstrates that the long-term trafficking of virus-specific memory CD8 T cells is altered in the draining LN as compared to the non-draining LNs for an extended period of time following resolution of infection, preferentially recruiting CD62L− effector memory CD8 T cells. Our data provides important insight into how vaccines may be manipulated to improve initial CTL responses to particular pathogens. The route of immunization can be controlled to target specific LNs that may be involved in responding to viral infections. For example, either intranasal or i.p. immunization against influenza virus may provide a long-term resident population of effector memory CD8 T cells in the lung draining LNs that are better suited to elicit effector functions after live virus infection [36], [37]. Furthermore, other mucosal immunization routes may be utilized to enhance activated CD8 T cells in the LNs that drain the vaginal tract to protect against either HIV-1 or HSV infection.
All experimental procedures utilizing mice were approved by the University of Iowa Animal Care and Use Committee. The experiments performed in this study were done under strict accordance to the Office of Laboratory Animal Welfare guidelines and the PHS Policy on Humane Care and Use of Laboratory Animals.
The Armstrong strain of LCMV was a gift from Raymond Welsh (University of Massachusetts Medical School, Worcester, MA) and was propagated in baby hamster kidney cells (American Type Culture Collections; ATCC, Manassas, VA). C57BL/6NCr Thy1.2+ mice were obtained from the National Cancer Institute (Frederick, MD). SplnX C57BL/6NCr mice were obtained from the National Cancer Institute and the splenectomy was performed at Charles River Laboratories (Wilmington, MA). SplnX mice were rested for greater than one month following splenectomy prior to LCMV infection. LT-α-KO mice were a gift from Dr. John Harty (University of Iowa, Iowa City, IA). All mice were age-matched and infected i.p. with 5×104 plaque forming units (PFU) of LCMV. T cell receptor transgenic P14 CD8 T cells (specific for the LCMV GP33–41 epitope) were isolated from either the spleen or peripheral blood of Thy1.1+ P14 mice. CFSE labeling of P14 CD8 T cells was performed by incubating 107 splenocytes/ml from P14 mice for 10 minutes at 37°C in the presence of 5 µM CFSE. CFSE-labeled cells were washed twice with RPMI 1640 containing 10% fetal calf serum and twice with sterile PBS. In some experiments, LCMV-infected mice were treated from days +1 to +4 with 50 µg of FTY720 (Cayman Chemical Co., Ann Arbor, MI) in sterile, endotoxin-free H2O. Control mice were administered H2O.
Mice were infected with LCMV and at various times p.i., organs were harvested and placed in sterile, serum-free RPMI 1640. Spleens and LNs were disrupted using a tissue homogenizer (Ultra-Turrax T25, IKA, Wilmington, NC) and tissue homogenates were subsequently centrifuged at 2000 rpm for 10 min. Cell-free supernatants were collected and snap-frozen in liquid nitrogen prior to storage at −80°C. Samples were thawed and virus titers were determined by plaque assay on Vero cells.
Tissues were harvested and mononuclear cells were obtained from the spleen and LNs by pressing the organs between the ends of frosted slides. In some experiments, the spleen and LNs were digested prior to mononuclear cell isolation to ensure maximal liberation of cells. Spleens and LNs were diced and placed in 5 or 1 ml, respectively, of Hank's balanced salt solution supplemented with 125 U/ml of collagenase type II (Invitrogen), 60 U/ml of DNAse I type II (Sigma-Aldrich, St. Louis, MO) and incubated at 37°C for 30 min followed by disruption with frosted glass slides. In experiments where lungs and livers were harvested the mice were first perfused with 20 ml of sterile saline and the tissues were subsequently pressed through a wire mesh screen (Cellector, Bellco Glass, Inc., Vineland, NJ). Blood was collected from isoflourane anethsitized mice by eye bleed into 4% (w/v) sodium citrate and red blood cells were lysed with NH4Cl. Peptides corresponding to the LCMV CD8 T cell epitopes GP33–41, NP396–404 and GP276–286 were purchased from Biosynthesis Inc. (Lewisville, TX). To enumerate the number of LCMV-specific CD8 T cells, mononuclear cells from the spleen, LNs and livers were stimulated in vitro in the presence of 1 µM peptide and 10 µg/ml brefeldin A (Sigma-Aldrich) for 5 h at 37°C. Previous work from our laboratory has shown that mononuclear cells isolated from the lung and peripheral blood require stimulation by exogenous antigen presenting cells coated with peptide to accurately enumerate the number of antigen-specific cells in these locations [38]. Lung and peripheral blood were stimulated with EL4 cells (American Type Culture Collection, Manassas, VA) coated with 1 µM peptide in the presence of brefeldin A for 5 h at 37°C. Cells were subsequently stained for cell surface CD4, CD8, Thy1.2 and intracellular IFN-γ as previously described [39]. In some experiments, cells from LNs and spleens were stained with fluorescein isothiocyanate-conjugated peanut agglutinin (Vector Laboratories, Burlingame, CA), B220 (eBioscience) and CD19 (eBioscience) for the detection of germinal center B cells.
Mice were infected with 5×104 PFU of LCMV i.p. and 34 days p.i., LNs and spleens were harvested and digested in collagenase and DNase as described above. Memory P14 CD8 T cells used as antigen sensors were generated by adoptive transfer of 2×103 naïve Thy1.1+ P14 cells into naïve C57BL/6 Thy1.2+ recipients that were infected with 5×104 PFU of LCMV the following day. Memory P14 CD8 T cells (>60 days p.i.) were positively enriched by staining splenocytes with phycoerythrin-conjugated anti-Thy1.1 (eBioscience) followed by labeling with anti-phycoerythrin-conjugated magnetic beads (Miltenyi Biotec, Auburn, CA) according to the manufacturer's directions followed by separation via AutoMACS (Miltenyi Biotec). MACS-enriched memory P14 CD8 T cells were CFSE labeled with 10 µM CFSE and co-cultured at either a 1∶10 or a 1∶100 ratio with LN or spleen cells from day 34 LCMV-infected mice for 3 days at 37°C and 5% CO2. As a negative control, CFSE-labeled memory P14 CD8 T cells were also co-cultured with either LN or spleen cells from naïve mice. As a positive control for this assay, CFSE-labeled memory P14 CD8 T cells were co-cultured at a 1∶100 ratio with either naïve LN or spleen cells pulsed with 1 µM LCMV GP33–41 peptide.
LNs were harvested from naïve, day 4 and day 34 LCMV-infected mice. LNs were homogenized in 1 ml of TRIzol (Invitrogen Life Technologies) and RNA was collected as previously described [36]. Real-time PCR to detect the GP mRNA of LCMV was performed with TaqMan Universal PCR Master Mix (Applied Biosystems) on an ABI 7300 Real Time PCR System (Applied Biosystems) using universal thermal cycling parameters. Results were analyzed using Sequence Detection System Analysis Software (Applied Biosystems). GP gene primers and probe were previously published [27] and purchased from Integrated DNA Technologies. The probe was synthesized to contain FAM reporter dye and 3′-TAMRA quencher dye. Samples were compared with known standard dilutions of a plasmid containing the GP gene of LCMV [40], a gift from Dr. Juan Carlos de la Torre (Scripps Research Institute, San Diego, CA). The number of GP gene copies per LN was calculated based on the number of copies of the GP gene in the sample and the total RNA isolated from the LNs.
To purify memory CD62L+ and CD62L− P14 CD8 T cells, splenocytes from ≥day 45 LCMV-immune C57BL/6 mice were stained with Thy1.1-PE (Biolegend, San Diego, CA). Cells were subsequently stained with anti-PE magnetic beads (Miltenyi Biotec Inc, Auburn, CA) and positively selected using an AutoMACS (Miltenyi Biotec). AutoMACS-enriched cells were stained for CD8 and CD62L and sorted using a BD FACSAria II (BD Biosciences). Sorted CD62L+ P14 CD8 T cells were labeled with 1 µM CFSE (Molecular Probes, Carlsbad, CA) and mixed with unlabeled CD62L− P14 cells at a 1∶1 ratio and 0.75–1.25×106 CFSE-labeled cells were adoptively transferred i.v. into day 34 LCMV immune C57BL/6 (Thy1.2+) mice.
|
10.1371/journal.pcbi.1006201 | Temporal precision of regulated gene expression | Important cellular processes such as migration, differentiation, and development often rely on precise timing. Yet, the molecular machinery that regulates timing is inherently noisy. How do cells achieve precise timing with noisy components? We investigate this question using a first-passage-time approach, for an event triggered by a molecule that crosses an abundance threshold and that is regulated by either an accumulating activator or a diminishing repressor. We find that either activation or repression outperforms an unregulated strategy. The optimal regulation corresponds to a nonlinear increase in the amount of the target molecule over time, arises from a tradeoff between minimizing the timing noise of the regulator and that of the target molecule itself, and is robust to additional effects such as bursts and cell division. Our results are in quantitative agreement with the nonlinear increase and low noise of mig-1 gene expression in migrating neuroblast cells during Caenorhabditis elegans development. These findings suggest that dynamic regulation may be a simple and powerful strategy for precise cellular timing.
| Cells control important processes with precise timing, even though their underlying molecular machinery is inherently imprecise. In the case of Caenorhabditis elegans development, migrating neuroblast cells produce a molecule until a certain abundance is reached, at which time the cells stop moving. Precise timing of this event is critical to C. elegans development, and here we investigate how it can be achieved. Specifically, we investigate regulation of the molecule production by either an accumulating activator or a diminishing repressor. Our results are consistent with the nonlinear increase and low noise of gene expression observed in the C. elegans cells.
| Proper timing is crucial for biological processes, including cell division [1–3], cell differentiation [4], cell migration [5], viral infection [6], embryonic development [7, 8], and cell death [9]. These processes are governed by molecular events inside cells, i.e., production, degradation, and interaction of molecules. Molecular events are subject to unavoidable fluctuations, because molecule numbers are small and reactions occur at random times [10, 11]. Cells combat these fluctuations using networks of regulatory interactions among molecular species. This raises the fundamental question of whether there exist regulatory strategies that maximize the temporal precision of molecular events and, in turn, cellular behaviors.
A canonical mechanism by which a molecular event triggers a cellular behavior is accumulation to a threshold [3, 4, 12–14]: molecules are steadily produced by the cell, and once the molecule number crosses a particular threshold, the behavior is initiated. The temporal precision of the behavior is therefore bounded by the temporal precision of the threshold crossing. Threshold crossing has been shown to underlie cell cycle progression [3] and sporulation [4], although alternative strategies, such as derivative [9] or integral thresholding [15], oscillation [16], and dynamical transitions in the regulatory network [8], have also been investigated.
Recent work has investigated the impact of auto-regulation (i.e., feedback) on the temporal precision of threshold crossing [12, 13]. Interestingly, it was found that auto-regulation generically decreases the temporal precision of threshold crossing, meaning that the optimal strategy is a linear increase of the molecule number over time with no auto-regulation [12] (although auto-regulation can help if there is a large timescale separation and the threshold itself is also subject to optimization [13]). Indeed, even when the molecule also degrades, the optimal precision is achieved when positive auto-regulation counteracts the effect of degradation, preserving the linear increase over time [12]. However, in many biological processes, such as the temporal control of neuroblast migration in Caenorhabditis elegans [5], the molecular species governing the behavior increases nonlinearly over time. This suggests that other regulatory interactions beyond auto-regulation may play an important role in determining temporal precision. In particular, the impact of activation and repression on temporal precision, where the activator or repressor has its own stochastic dynamics, remains unclear.
Here we investigate the temporal precision of threshold crossing for a molecule that is regulated by either an accumulating activator or a degrading repressor. Using a first-passage-time approach [12, 17–19] and a combination of computational and analytic methods, we find that, unlike in the case of auto-regulation, the presence of either an activator or a repressor increases the temporal precision beyond that of the unregulated case. Furthermore, the optimal regulatory strategy for either an activator or a repressor corresponds to a nonlinear increase in the regulated molecule number over time. We elucidate the mechanism behind these optimal strategies, which stems from a tradeoff between reducing the noise of the regulator and that of the target molecule, and is similar to the fact that a sequence of time-ordered stochastic events becomes more precisely timed with more events. These findings are robust to more complex features of the regulation process, including bursts of molecule production, more complex regulator dynamics, and cell division. Our results are quantitatively consistent with both the temporal precision and nonlinearity of the mig-1 mRNA dynamics of the migrating neuroblast cells in C. elegans larvae [5]. The agreement of our simple model with these data suggests that many molecular timing processes may benefit from the generic regulatory strategies we identify here.
We consider a molecular species X whose production is regulated by a second species, either an activator A or a repressor R (Fig 1A). The regulator undergoes its own dynamics: the activator undergoes pure production at a zeroth-order rate k whereas the repressor undergoes pure degradation at a first-order rate μ, such that in either case the production rate of X increases over time. The activator does not degrade and the repressor is not produced, although we later relax this assumption. For the regulation function we take a Hill function, which is a generic model of cooperative regulation [12, 13, 20],
f + ( a ) = α a H a H + K H ( activator ) , (1)
f - ( r ) = α K H r H + K H ( repressor ) . (2)
Here a and r are the molecule numbers of A and R, respectively, α is the maximal production rate of X, K is the half-maximal regulator number, and H is the cooperativity. First we neglect additional complexities such as bursts of production, more complex regulator dynamics, cell division, auto-regulation, longer regulatory cascades, or transcriptional delay. Later we check the robustness of our results to bursts, more complex regulator dynamics, and cell division, and we speculate upon the effects of auto-regulation, longer regulatory cascades, and delay in the Discussion.
We suppose that a behavior is initiated when the molecule number x crosses a threshold x* (Fig 1B). Because the production of X and the dynamics of the regulator are stochastic, the time at which x first reaches x* is a random variable. We characterize the precision of this event by the mean t ¯ and variance σ t 2 of this first-passage time, which we compute numerically from the master equation corresponding to the reactions in Fig 1A (see Materials and methods). The maximal production rate α is set to ensure that t ¯ is equal to a target time t*, which we assume is set by functional constraints on the initiated behavior. This leaves k, K, and H as free parameters of the regulation (with α a function of these parameters). In principle, these parameters can be optimized to minimize the timing variance σ t 2.
The deterministic dynamics, illustrated in Fig 1C and 1D, neglect fluctuations but give an intuitive picture of the regulation. Whereas the amount of activator increases linearly over time, the amount of repressor decays exponentially from an initial molecule number N:
a ¯ ( t ) = k t , (3)
r ¯ ( t ) = N e - μ t . (4)
In either case, the production rate f± of X increases over time, such that x ¯ increases nonlinearly. N is an additional free parameter in the repressor case.
To investigate the effects of regulation on temporal precision, we consider the timing variance σ t 2 as a function of the parameters k and K, or μ and K. The special case of no regulation corresponds to the limits k → ∞ and K → 0 in the case of activation, or μ → ∞ and K → ∞ in the case of repression. In this case, the production of X occurs at the constant rate α. Reaching the threshold requires x* sequential events, each of which occurs in a time that is exponentially distributed with mean 1/α. The total completion time for such a process is given by a gamma distribution with mean t ¯ = x * / α and variance σ t 2 = x * / α 2 [19]. Ensuring that t ¯ = t * requires α = x*/t*, for which the variance satisfies σ t 2 x * / t * 2 = 1. This expression gives the timing variance for the unregulated process.
In Fig 2 we plot the scaled variance σ t 2 x * / t * 2 as a function of the parameters k and K, or μ and K, for cooperativity H = 3 (color maps). In the case of activation (Fig 2A), the variance decreases with increasing k and K. This means that the temporal precision is highest for an activator that accumulates quickly and requires a high abundance to produce X. In the case of repression (Fig 2B), the variance has a global minimum as a function of μ and K. This means that the temporal precision is highest for a repressor with a particular well-defined degradation rate and abundance threshold. Importantly, we see that for both activation and repression, the scaled variance can be less than one, meaning that regulation allows improvement of the temporal precision beyond that of the unregulated process. We have checked that this result holds for H ≥ 1.
To understand the dependencies in Fig 2, we develop analytic approximations. First, we assume that H → ∞, such that the regulation functions in Eqs 1 and 2 become threshold functions. In this limit, the production rate of X is zero if a < K or r > K, and α otherwise. The deterministic dynamics of X become piecewise-linear,
x ¯ ( t ) = { 0 t < t 0 α ( t - t 0 ) t ≥ t 0 , (5)
where t0 is determined by either a ¯ ( t 0 ) = K or r ¯ ( t 0 ) = K according to Eqs 3 and 4. Then, to set α, we use the condition x ¯ ( t * ) = x *, which results in α = x*/(t* − t0).
Lastly, we approximate the variance in the first-passage time using the variance in the molecule number and the time derivative of the mean dynamics [13]. Specifically, the timing variance of X arises from two sources: (i) uncertainty in the time when the regulator crosses its threshold K, which determines when the production of the target X begins, and (ii) uncertainty in the time when x crosses its threshold x*, given that production begins at a particular time. The first source is regulator noise, and the second source is target noise. We estimate these timing variances from the associated molecule number variances, propagated via the time derivatives,
σt2≈σy2(dy¯dt)−2|t0︸regulator+σx2(dx¯dt)−2|t*︸target, (6)
where y ∈ {a, r} denotes the regulator molecule number.
For the activator, which undergoes a pure production process with rate k, the molecule number obeys a Poisson distribution with mean kt. Therefore, the molecule number variance at time t0 is σ a 2 = k t 0. For the repressor, which undergoes a pure degradation process with rate μ starting from N molecules, the molecule number obeys a binomial distribution with number of trials N and success probability e−μt. Therefore, the molecule number variance at time t0 is σ r 2 = N e - μ t 0 ( 1 - e - μ t 0 ). For the target molecule, which undergoes a pure production process with rate α starting at time t0, the molecule number obeys a Poisson distribution with mean α(t − t0). Therefore, the molecule number variance at time t* is σ x 2 = α ( t * - t 0 ). Inserting these expressions into Eq 6, along with the derivatives calculated from Eqs 3–5 and the appropriate expressions for α and t0, we obtain
σ t 2 x * t * 2 ≈ K x * ( k t * ) 2 + ( 1 - K k t * ) 2 ( activator ) , (7)
σ t 2 x * t * 2 ≈ ( N - K ) x * N K ( μ t * ) 2 + [ 1 - log ( N / K ) μ t * ] 2 ( repressor ) . (8)
As a function of kt* and K, the global minimum of Eq 7 occurs as kt* → ∞ and K → ∞. The path of descent toward this minimum is given by differentiating with respect to K at fixed kt* and setting the result to zero, which yields the curve
K = { 0 k t * < x * 2 k t * - x * 2 k t * ≥ x * 2 , (9)
along which the variance satisfies
σ t 2 x * t * 2 = { 1 k t * < x * 2 x * k t * ( 1 - x * 4 k t * ) k t * ≥ x * 2 , (10)
where the first case comes from the fact that K must be nonnegative. In contrast, the global minimum of Eq 8 occurs at finite μt* and K: differentiating with respect to each and setting the results to zero gives the values
K = e - 2 N , (11a)
μ t * = e 2 x * 2 N + 2 , (11b)
σ t 2 x * t * 2 = x * x * + 4 e - 2 N , (12)
where we have assumed that K/N ≪ 1 (see Materials and methods), which is justified post-hoc by Eq 11a.
These analytic approximations are compared with the numerical results for the activator in Fig 2A (white dashed line, Eq 9) and for the repressor in Fig 2B (white circle, Eq 11). In Fig 2A we see that the global minimum indeed occurs as kt* → ∞ and K → ∞, and the predicted curve agrees well with the observed descent. In Fig 2B we see that the predicted global minimum lies very close to the observed global minimum. We have also checked along specific slices in the (K, kt*) or (K, μt*) plane and found that the analytic approximations generally differ from the numerical results by about 10% or less, despite the fact that the approximations take H → ∞ whereas the numerics in Fig 2 use H = 3.
The success of the approximations means that Eq 6 describes the key mechanism leading to the optimal temporal precision. Eq 6 demonstrates that the optimal regulatory strategy arises from a tradeoff between minimizing regulator and target noise. On the one hand, minimizing only the regulator noise would require that the regulator cross its threshold K with a steep slope d y ¯ / d t and therefore at an early time, meaning that the target molecule would be effectively unregulated and would increase linearly over time. On the other hand, minimizing only the target noise would require that the regulator cross its threshold only shortly before the target time t*, such that the target molecule would cross its threshold x* with a steep slope d x ¯ / d t, leading to a highly nonlinear increase of the target molecule over time. In actuality, the optimal strategy is somewhere in between, with the regulator crossing its threshold at some intermediate time t0, and the target molecule exhibiting moderately nonlinear dynamics as in Fig 1C and 1D.
Eqs 10 and 12 demonstrate that the timing variance is small for large kt*/x* in the case of activation, and small for large N/x* in the case of repression. This makes intuitive sense because each of these quantities scales with the number of regulator molecules: k is the production rate of activator molecules, while N is the initial number of repressor molecules. To make this intuition quantitative, we define a cost as the time-averaged number of regulator molecules,
⟨ a ⟩ = 1 t * ∫ 0 t * d t a ¯ ( t ) = 1 2 k t * , (13)
⟨ r ⟩ = 1 t * ∫ 0 t * d t r ¯ ( t ) = N μ t * ( 1 - e - μ t * ) , (14)
where the second steps follow from Eqs 3 and 4. We see that, indeed, 〈a〉 scales with k, and 〈r〉 scales with N. Thus, Eqs 10 and 12 demonstrate that increased temporal precision comes at a cost, in terms of the number of regulator molecules that must be produced.
We test our model predictions using data from neuroblast cells in C. elegans larvae [5]. During C. elegans development, particular neuroblast cells migrate from the posterior to the anterior of the larva. It has been shown that the migration terminates not at a particular position, but rather after a particular amount of time, and that the termination time is controlled by a temporal increase in the expression of the mig-1 gene [5]. Since mig-1 is known to be subject to regulation [21], we investigate the extent to which the dynamics of mig-1 can be explained by the predictions of our model.
Fig 3A shows the number x of mig-1 mRNA molecules per cell as a function of time t, obtained by single-molecule fluorescent in situ hybridization (from [5]). We analyze these data in the following way (see Materials and methods for details). First, noting that the dynamics are nonlinear, we quantify the linearity using the area under the curve, normalized by that for a perfectly linear trajectory x*t*/2,
ρ = 2 x * t * ∫ 0 t * d t x ( t ) . (15)
By this definition, ρ = 1 for perfectly linear dynamics, and ρ → 0 for maximally nonlinear dynamics (a sharp rise at t*). Then, we estimate x*, t*, and the timing variance σ t 2 from the data. Specifically, migration is known to terminate between particular reference cells in the larva [5], which gives an estimated range for the termination time t*. This range is shown in magenta in Fig 3A and corresponds to a threshold within the approximate range 10 ≤ x* ≤ 25. Therefore, we divide the x axis into bins of size Δx, choose bin midpoints x* within this range, and for each choice compute the mean t* and the variance σ t 2 of the data in that bin. Fig 3B shows the average and standard deviation of results for different values of x* and Δx (blue circle).
The experimental data point in Fig 3B exhibits two clear features: (i) the dynamics are nonlinear (ρ is significantly below 1), and (ii) the timing variance is low (σ t 2 x * / t * 2 is significantly below 1). Neither feature can be explained by a model in which the production of x is unregulated, since that would correspond to a linear increase of molecule number over time (ρ = 1) and a timing variance that satisfies σ t 2 x * / t * 2 = 1 (square in Fig 3B). Furthermore, since auto-regulation has been shown to generically increase timing variance beyond the unregulated case [12], it is unlikely that feature (ii) can be accounted for by a model with auto-regulation alone. Can these data be accounted for by our model with regulation?
To address this question we calculate ρ and σ t 2 x * / t * 2 from our model. For simplicity, we focus on the analytic approximations in Eqs 7 and 8, since they have been validated in Fig 2. In these approximations, since x ¯ ( t ) is piecewise-linear (Eq 5), calculating ρ via Eq 15 is straightforward: ρ = 1 − t0/t*, where t0 is once again determined by either a ¯ ( t 0 ) = K or r ¯ ( t 0 ) = K according to Eqs 3 and 4. For a given ρ and cost 〈a〉/x* or 〈r〉/x*, we calculate the minimum timing variance σ t 2 x * / t * 2. For the activator, we use the expression for ρ along with Eq 13 to write Eq 7 in terms of ρ and 〈a〉/x*,
σ t 2 x * t * 2 = x * 2 ⟨ a ⟩ ( 1 - ρ ) + ρ 2 . (16)
For the repressor, we use the expression for ρ along with Eq 14 to write Eq 8 in terms of ρ and 〈r〉/x*, and then minimize over N (see Materials and methods) to obtain
σ t 2 x * t * 2 = e 3 27 x * ⟨ r ⟩ ( 1 - ρ ) 3 + ρ 2 . (17)
Eqs 16 and 17 are shown in Fig 3B (green solid and red dashed curves, respectively), and we see the same qualitative features for both cases: all curves satisfy σ t 2 x * / t * 2 = 1 at ρ = 1, as expected; and as ρ decreases, each curve exhibits a minimum whose depth and location depend on cost. Specifically, as cost increases (lighter shades of green or red), the variance decreases, as expected. Importantly, we see that at a cost on the order of 〈a〉/x* = 〈r〉/x* ∼ 10, the model becomes consistent with the experimental data: both the low timing variance and the low degree of linearity predicted by either the activator or repressor case agree quantitatively with the experiment. This suggests that either an accumulating activator or a degrading repressor is sufficient to account for the temporal precision observed in mig-1-controlled neuroblast migration.
Our minimal model neglects common features of gene expression such as bursts in molecule production [22] and additional sources of noise. Therefore we test the robustness of our findings to these effects here. First, we test the robustness of the results to the presence of bursts by replacing the Poisson process governing the activator production with a bursty production process. Specifically, we assume that each production event increases the activator molecule count by an integer in [1, ∞) drawn from a geometric distribution with mean b [23, 24]. The limiting case of b = 1 recovers the original Poisson process. The results are shown in Fig 4A for b = 1, 3, and 5 (green solid, cyan dashed, and magenta dashed curves). We see that bursts in the activator increase the timing variance of the target molecule, as expected, but that there remain parameters for which the variance is less than that for the unregulated case, σ t 2 x * / t * 2 = 1 (dashed black line). This result shows that even with bursts, regulation by an accumulating activator is beneficial for timing precision.
We also recognize that whereas the activator can be assumed to start with exactly zero molecules, it is more realistic for the repressor to start with an initial number of molecules that has its own variability. We incorporate this additional variability into the model by performing stochastic simulations [25] of the reactions in Fig 1A and drawing the initial repressor molecule number from a Poisson distribution across simulations. The result is shown by the green dashed curve in Fig 4B. We see that the additional variability gives rise to an increase in the timing variance of the target molecule, as expected (compare with the green solid curve). However, for most of the range of degradation rates, including the optimal degradation rate, the variance remains less than that of the unregulated case, σ t 2 x * / t * 2 = 1 (dashed black line). This result indicates that the benefit of repression is robust to this additional source of noise.
Then, we test the robustness of the results to our assumptions that the activator undergoes pure production and the repressor undergoes pure degradation. Specifically, we introduce a degradation rate μ for the activator, and a production rate k for the repressor, such that either regulator reaches a steady state of k/μ. The blue curves in Fig 4A and 4B show the case where the increasing activator’s steady state k/μ is twice its regulation threshold K, or the decreasing repressor’s steady state k/μ is half its regulation threshold K, respectively. In both cases, we see that the timing variance of the target molecule increases because the regulator slows down on the approach to its regulation threshold. Nonetheless, we see that it is still possible for the variance to be lower than that of the unregulated case. The red curves show the case where the regulator’s steady state is equal to its regulation threshold. Here we are approaching the regime in which threshold crossing is an exponentially rare event. As a result, the variance further increases, to the point where it is above that of the unregulated case for the full range of parameters shown. These results demonstrate that the benefit of regulation is robust to more complex regulator dynamics, but only if the regulator still crosses its regulation threshold at a reasonable mean velocity.
Finally, we test the robustness of the results to a feature exhibited by the experimental mig-1 data: near the end of migration, cell division occurs (Fig 3A, black data). One daughter cell continues migrating (Fig 3A, dark blue data), while the other undergoes programmed cell death [5]. To investigate the effects of cell division, we perform stochastic simulations, and at a given time td we assume that the cell volume V is reduced by a factor of two. For each simulation, td is drawn from a Gaussian distribution with mean t ¯ d and variance σ d 2 determined by the data (Fig 3A, black). At td, we reduce the molecule numbers of both the regulator and the target molecule assuming symmetric partitioning, such that the molecule number after division is drawn from a binomial distribution with total number of trials equal to the molecule number before division and success probability equal to one half. We also reduce the molecule number threshold K by a factor of two because it is proportional to the cell volume via K = KdV, where Kd is the dissociation constant.
Fig 4C shows the dynamics of the mean molecule numbers of the activator (green solid) and its target (blue solid), or the repressor (red dashed) and its target (blue dashed). We see that the activator, repressor, and target drop in molecule number at division but that the abruptness of the drop is smoothed by the variability in the division time. The smoothing is more pronounced in the cases of the repressor and the target because the molecule numbers of these species are smaller at division. Thus, for either the activator or repressor mechanism, we see that the experimentally observed variability in division time is sufficient to smooth out the dynamics of the target molecule number, consistent with the experimental data in Fig 3A.
Additionally, we see in Fig 4D that the timing variance of the target molecule in both the activator and repressor cases is similar to that without division in the region of the experimental division time. This further indicates that either model remains sufficient to account for the low experimental timing variance, even with the experimentally observed cell division. Taken together, the results in Fig 4C and 4D show that the key results of the model are robust to the effects of cell division.
We have demonstrated that regulation by an accumulating activator or a diminishing repressor increases the precision of threshold crossing by a target molecule, beyond the precision achievable with constitutive expression alone. The increase in precision results from a tradeoff between reducing the noise of the regulator and reducing the noise of the target molecule itself. Our minimal model is sufficient to account for both the high degree of nonlinearity and the low degree of noise in the dynamics of mig-1 in C. elegans neuroblasts, providing evidence for the hypothesis that these cells use regulated expression to terminate their migration with increased temporal precision. These results suggest that regulation by a dynamic upstream species is a simple and generic method of increasing the temporal precision of cellular behaviors governed by threshold-crossing events.
Why does regulation increase temporal precision, whereas it has been shown that auto-regulation (feedback) does not [12]? After all, either regulation or positive feedback can produce an acceleration in molecule number over time, leading to a steeper threshold crossing. The reason is likely that positive feedback relies on self-amplification. In addition to amplifying the mean, positive feedback also amplifies the noise. In contrast, regulation by an external species does not involve self-amplification. Therefore, the noise increase is not as strong. The target molecule certainly inherits noise from the regulator (Eq 6), but the increase in noise does not outweigh the benefit of the acceleration, as it does for feedback. Future work could investigate the interplay of regulation and feedback, as well as active degradation of the target molecule, especially as mig-1 is thought to be subject to feedback and degradation in addition to external regulation [5, 21].
Our finding that regulation increases temporal precision is related to the more basic phenomenon that a sequence of ordered events has a lower relative timing error than a single event [19, 26]. Specifically, if a single event occurs in a time that is exponentially distributed with a mean τ, then the relative timing error is σ/τ = 1. For n such events that must occur in sequence, the total completion time follows a gamma distribution with relative timing error σ / τ = 1 / n, which decreases with increasing n. Thus, at a coarse-grained level, the addition of a regulator can be viewed as increasing the length of the sequence of threshold-crossing events from one to two, and thus decreasing the timing error. This perspective suggests that the timing error could be decreased even further via a cascade of regulators.
Although we have demonstrated that our findings are robust to complexities such as bursts and cell division (Fig 4), our model neglects additional features of regulated gene expression such as transcriptional delay. Transcriptional delay has been shown to play an important role in regulation [27, 28] and to have consequences for the mean and variance of threshold-crossing times [29]. If a delay were to arise due to a sequence of stochastic but irreversible steps, then we conjecture that the relative timing error would decrease with the number of these steps, due to the same cascading mechanism mentioned in the previous paragraph. However, it has been shown that if there are reversible steps or cycles within a multistep process, then the first passage time distribution can approach an exponential as the number of steps becomes large [26]. In this case the timing statistics would be captured by our simple model, which assumes single exponentially distributed waiting times. Future work could explore the effects of transcriptional delay in more detail.
Finally, we have shown that the mig-1 data from migrating neuroblasts in C. elegans are quantitatively consistent with either the accumulating activator or diminishing repressor model, but the data do not distinguish between the two models. A direct approach to search for a distinction would be to use genetic knockout techniques to screen directly for regulators of mig-1 and their effects on its abundance. A less direct approach would be to more closely investigate the effects of the cell division that occurs during migration. For example, we assumed in this study that the volume fraction is equal to the average molecule number fraction in the surviving cell after division. However, if they were found to be unequal for either mig-1 or its regulator(s), then the concentrations of these species could undergo an abrupt change after division, which may have opposing consequences for the activator vs. the repressor mechanism. Future studies could use these or related approaches to more concretely identify the role of gene regulation in achieving precise timing during cellular processes.
We compute the first-passage time statistics t ¯ and σ t 2 numerically from the master equation following [12], generalized to a two-species system. Specifically, the probability F(t) that the molecule number crosses the threshold x* at time t is equal to the probability Py, x*−1(t) that there are y regulator molecules (where y ∈ {a, r}) and x* − 1 target molecules, and that a production reaction occurs with rate f±(y) to bring the target molecule number up to x*. Since this event can occur for any regulator molecule number y, we sum over all y,
F ( t ) = ∑ y = 0 Y f ± ( y ) P y , x * - 1 ( t ) , (18)
where Y = {amax, N}. The repressor has a maximum number of molecules N, whereas the activator number is unbounded, and therefore we introduce the numerical cutoff a max = k t * + 10 k t *. The probability Pyx evolves in time according to the master equation corresponding to the reactions in Fig 1A,
P ˙ a x = k P a - 1 , x + f + ( a ) P a , x - 1 - [ k + f + ( a ) ] P a x , (19a)
P ˙ r x = μ ( r + 1 ) P r + 1 , x + f - ( r ) P r , x - 1 - [ μ r + f - ( r ) ] P r x . (19b)
The moments of Eq 18 are
⟨ t m ⟩ = ∑ y = 0 Y f ± ( y ) ∫ 0 ∞ d t t m P y , x * - 1 ( t ) , (20)
where t ¯ = 〈 t 〉 and σ t 2 = 〈 t 2 〉 - 〈 t 〉 2. Therefore computing t ¯ and σ t 2 requires solving for the dynamics of Pyx.
Because Eq 19 is linear in Pyx, it is straightforward to solve by matrix inversion. We rewrite Pyx as a vector by concatenating its columns, P → ⊤ = [ [ P 00 , … , P Y 0 ] , … , [ P 0 , x * - 1 , … , P Y , x * - 1 ] ], such that Eq 19 becomes P → ˙ = M P →, where
M = [ M ( 1 ) M ( 2 ) M ( 1 ) M ( 2 ) M ( 1 ) ⋱ ⋱ M ( 2 ) M ( 1 ) ] . (21)
Here, for i, j ∈ {0, …, Y}, the x* − 1 subdiagonal blocks are the diagonal matrix M i j ( 2 ) = f ± ( i ) δ i j, and the x* diagonal blocks are the subdiagonal matrix M i j ( 1 ) = - [ k ( 1 - δ i a max ) + f + ( i ) ] δ i j + k δ i - 1 , j or the superdiagonal matrix M i j ( 1 ) = - [ μ i + f - ( i ) ] δ i j + μ ( i + 1 ) δ i + 1 , j for the activator or repressor case, respectively. The δ i a max term prevents activator production beyond amax molecules. The final M(1) matrix in Eq 21 contains f± production terms that are not balanced by equal and opposite terms anywhere in their columns. These terms correspond to the transition from x* − 1 to x* target molecules, for which there is no reverse transition. Therefore, the state with x* target molecules (and any number of regulator molecules) is an absorbing state that is outside the state space of P → [12]. Consequently, probability leaks over time, and P → ( t → ∞ ) = ∅ →. Equivalently, the eigenvalues of M are negative.
The solution of the dynamics above Eq 21 is P → ( t ) = exp ( M t ) P → ( 0 ). Therefore, Eq 20 becomes 〈 t m 〉 = V → ⊤ [ ∫ 0 ∞ d t t m exp ( M t ) ] P → ( 0 ), where V → ⊤ is a row vector of length x*(Y + 1) consisting of [f±(0), …, f±(Y)] preceded by zeros. We solve this equation via integration by parts [12], noting that the boundary terms vanish because the eigenvalues of M are negative, to obtain
⟨ t m ⟩ = ( - 1 ) m + 1 m ! V → ⊤ ( M - 1 ) m + 1 P → ( 0 ) . (22)
We see that computing t ¯ = 〈 t 〉 and σ t 2 = 〈 t 2 〉 - 〈 t 〉 2 requires inverting M, which we do numerically in Matlab. We initialize P → as Pax(0) = δa0 δx0 or Prx(0) = δrN δx0 for the activator or repressor case, respectively.
When including cell division, we compute t ¯ and σ t 2 from 50,000 stochastic simulations [25].
The dynamics of the mean regulator and target molecule numbers are obtained by calculating the first moments of Eq 19, ∂ t y ¯ = ∑ y x y P ˙ y x and ∂ t x ¯ = ∑ y x x P ˙ y x, where y ∈ {a, r}. For the regulator we obtain ∂ t a ¯ = k or ∂ t r ¯ = - μ r ¯ in the activator or repressor case, respectively, which are solved by Eqs 3 and 4. For the target molecule we obtain ∂ t x ¯ = 〈 f ± ( y ) 〉, which is not solvable because f± is nonlinear (i.e., the moments do not close). A deterministic analysis conventionally assumes 〈 f ± ( y ) 〉 ≈ f ± ( y ¯ ), for which the equation becomes solvable by separation of variables. For example, in the case of H = 1, using Eqs 1–4, we obtain
x ¯ ( t ) = { α t - ( α K / k ) log [ ( k t + K ) / K ] ( activator ) ( α / μ ) log [ ( N + K e μ t ) / ( N + K ) ] ( repressor ) . (23) Eq 23 is plotted in Fig 1C and 1D.
When including cell division, we compute the mean dynamics from the simulation trajectories (Fig 4C).
To find the global minimum of Eq 8, we differentiate with respect to kt* and K and set the results to zero, giving two equations. kt* can be eliminated, leaving one equation for K,
1 2 log N K = 1 - K N (24)
This equation is transcendental. However, in the limit K ≪ N, we neglect the last term, which gives Eq 11.
To derive Eq 17, we use
ρ = 1 - t 0 t * = 1 - log N / K μ t * (25)
where the second step follows from r ¯ ( t 0 ) = K according to Eq 4; and, from Eq 14,
⟨ r ⟩ = N μ t * ( 1 - e - μ t * ) ≈ N μ t * (26)
where the second step assumes that the repressor is fast-decaying, μt* ≫ 1. We use Eqs 26 and 25 to eliminate μt* and K from Eq 8 in favor of ρ and 〈r〉,
σ t 2 x * t * 2 ≈ x * ( e N ( 1 - ρ ) / ⟨ r ⟩ - 1 ) ⟨ r ⟩ 2 N 3 + ρ 2 . (27)
For nonlinear dynamics (ρ < 1) we may safely neglect the −1 in Eq 27. Then, differentiating Eq 27 with respect to N and setting the result to zero, we obtain N = 3〈r〉/(1 − ρ). Inserting this result into Eq 27 produces Eq 17.
To estimate the time at which migration terminates in Fig 3A, we refer to [5]. There, the position at which neuroblast migration terminates is measured with respect to seam cells V1 to V6 in the larva (see Fig. 4D in [5]). In particular, in wild type larvae, migration terminates between V2 and the midpoint of V2 and V1. This range corresponds to the magenta region in Fig 3A (see Fig. 4B, upper left panel, in [5]). Under the assumptions of constant migration speed and equal distance between seam cells, the horizontal axis in Fig 3A represents time.
To compute ρ for the experimental data in Fig 3A according to Eq 15 we use a trapezoidal sum. For the choices of x* and t* described in the text, this produces the ρ values in Fig 3B.
|
10.1371/journal.pntd.0004449 | Metabolomics-Based Discovery of Small Molecule Biomarkers in Serum Associated with Dengue Virus Infections and Disease Outcomes | Epidemic dengue fever (DF) and dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS) are overwhelming public health capacity for diagnosis and clinical care of dengue patients throughout the tropical and subtropical world. The ability to predict severe dengue disease outcomes (DHF/DSS) using acute phase clinical specimens would be of enormous value to physicians and health care workers for appropriate triaging of patients for clinical management. Advances in the field of metabolomics and analytic software provide new opportunities to identify host small molecule biomarkers (SMBs) in acute phase clinical specimens that differentiate dengue disease outcomes.
Exploratory metabolomic studies were conducted to characterize the serum metabolome of patients who experienced different dengue disease outcomes. Serum samples from dengue patients from Nicaragua and Mexico were retrospectively obtained, and hydrophilic interaction liquid chromatography (HILIC)-mass spectrometry (MS) identified small molecule metabolites that were associated with and statistically differentiated DHF/DSS, DF, and non-dengue (ND) diagnosis groups. In the Nicaraguan samples, 191 metabolites differentiated DF from ND outcomes and 83 differentiated DHF/DSS and DF outcomes. In the Mexican samples, 306 metabolites differentiated DF from ND and 37 differentiated DHF/DSS and DF outcomes. The structural identities of 13 metabolites were confirmed using tandem mass spectrometry (MS/MS). Metabolomic analysis of serum samples from patients diagnosed as DF who progressed to DHF/DSS identified 65 metabolites that predicted dengue disease outcomes. Differential perturbation of the serum metabolome was demonstrated following infection with different DENV serotypes and following primary and secondary DENV infections.
These results provide proof-of-concept that a metabolomics approach can be used to identify metabolites or SMBs in serum specimens that are associated with distinct DENV infections and disease outcomes. The differentiating metabolites also provide insights into metabolic pathways and pathogenic and immunologic mechanisms associated with dengue disease severity.
| Epidemics of dengue fever (DF) and dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS) are overwhelming public health capacity for diagnosis and patient care. Developing a panel of biomarkers in acute-phase serum specimens for prognosis of severe dengue disease would be of enormous value for appropriate triaging of patients for management. Metabolomics offers great potential for identification of small molecule biomarkers (SMBs) for diagnosis and prognosis of dengue virus (DENV) infections. We identified metabolites that were associated with and differentiated DHF/DSS, DF and non-dengue (ND) febrile illness outcomes, primary and secondary virus infections, and infections with different DENV serotypes. These metabolites provide insights into metabolic pathways that play roles in DENV infection, replication, and pathogenesis. Some are associated with lipid metabolism and regulation of inflammatory processes controlled by signaling fatty acids and phospholipids, and others with endothelial cell homeostasis and vascular barrier function. Such metabolites and associated metabolic pathways are potentially biologically relevant in DENV pathogenesis. The diagnostic and prognostic efficacy of differentiating metabolites is currently being investigated. Our goal is to identify the most parsimonious SMB biosignature that, when combined with laboratory diagnostic results, eg., DENV NS1 or RNA detection, will provide the most efficient algorithm for dengue diagnosis and prognosis.
| Epidemic dengue fever (DF) and dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS) have emerged throughout the tropical and subtropical world with devastating consequences and are overwhelming public health capacity for diagnosis and patient care [1, 2]. Upon presentation early after disease onset, it is clinically impossible to differentiate dengue virus (DENV)-infected patients who will have an unremarkable DF disease episode from those who will progress to potentially fatal DHF/DSS [3–7]. Viral biomarkers that correlate with dengue severity include viremia titer and nonstructural protein 1 (NS1) concentration in the blood, secondary DENV infection, and infection with specific virus genotypes [8–11]. Host biomarkers associated with disease severity include multiple immune molecules, biochemical and physiological response indicators, and genetic polymorphisms [3, 4, 12–21]. Algorithms based upon clinical signs and laboratory test results have been proposed to predict dengue severity [22–28]. However, currently there are no standardized biomarkers or algorithms for prognosis of severe disease outcomes.
Current diagnostic tests and approaches are not meeting the challenges posed by dengue [29, 30]. A paradigm shift in diagnosis/prognosis is essential to address the increasing threat of severe dengue disease. Advances in mass spectrometry, metabolite databases, and analytical software provide exciting new opportunities to identify small molecule biomarkers (SMBs) of dengue disease outcome in acute-phase serum specimens. Mass spectrometry-based metabolomics techniques are being applied with increasing frequency for diagnosis, investigation of pathogenic mechanisms, and monitoring the effects of treatments and interventions of infectious diseases [31–36].
Metabolomics is the analysis of low molecular weight biological molecules that result from metabolic processes. Disease states result in changes in metabolism in cells and systems that affect the profile of metabolites [34]. Analysis of metabolite profiles in disease conditions and comparison with the profiles of non-diseased individuals can be used in diagnosis. Metabolites that differentiate DF and DHF/DSS outcomes could potentially be exploited as SMBs for diagnosis of DENV infections and prognosis of disease severity. Liquid chromatography-tandem mass spectrometry (LC-MS/MS) metabolomics approaches have been used to detect and characterize changing metabolite levels in humans and mosquito vectors that are directly attributable to DENV infection and pathogenesis [32, 37]. Primary DENV infection in humans was shown to cause temporally distinct changes in the serum metabolome, particularly in the lipidome, reflecting the pathogenic mechanisms and metabolic pathways perturbed during the time course of DF [32].
Here, hydrophilic interaction liquid chromatography (HILIC)-MS [38, 39] was used to characterize retrospectively the serum metabolome of patients who were diagnosed as DHF/DSS, DF, or non-dengue (ND) febrile disease, as well as for preliminary characterization of the serum metabolome following infection with two different DENV serotypes and after primary or secondary DENV infection. In this exploratory, proof-of-concept study, metabolites that were associated with and differentiated DF and DHF/DSS, that predicted progression to DHF/DSS in serum of DF patients, and that differentiated infecting DENV serotypes and primary and secondary infections were identified. The differentiating metabolites reflect host responses to the pathogen, including tissue damage, inflammation, and other virus-induced pathology and thus provide insights into fundamental metabolic pathways associated with DENV pathogenesis and potentially novel targets for therapeutic intervention [32, 35]. We have identified candidate SMBs to be evaluated in prospective clinical studies for their diagnostic and prognostic efficacy for DENV infections.
Serum samples were obtained from collections of sera from patients who had presented with dengue-like febrile disease in Managua, Nicaragua, and Mérida, México. Nicaraguan serum samples had been collected as part of two ongoing pediatric studies being conducted by the University of California, Berkeley, the Nicaraguan Ministry of Health, and the Sustainable Sciences Institute: the Pediatric Dengue Cohort Study, which is focused upon studying transmission of DENV and identifying immune correlates of protection, and the Hospital-based Dengue Study, which is focused on studying clinical, immunological, and viral risk factors for severe DENV infections. Parents or legal guardians of participants provided written informed consent, participants 6 years of age and older provided assent, and participants in the Hospital-based Dengue Study 12 years of age and older provided written assent. These studies were approved by the UC Berkeley Committee for the Protection of Human Subjects (Protocols # 2010-06-1649 and 2010-09-2245) and the IRB of the Nicaraguan Ministry of Health. Mexican samples had been collected in the Laboratorio de Arbovirología, Centro de Investigaciones Regionales Dr. Hideyo Noguchi or the Unidad Universitaria de Inserción Social (UUIS) San José Tecoh, both from the Universidad Autónoma de Yucatán (UADY), Mérida, Yucatán, México, from patients who were referred by a primary-care physician for diagnostic testing. These samples were procured as part of the normal dengue diagnosis mission of the laboratory and not as part of an experimental protocol. These samples provided an opportunity to determine if metabolites detected in Nicaraguan patients that differentiated dengue disease outcomes could also be detected in dengue patients with different genetic, environmental, and demographic backgrounds. This research was approved by the Bioethics Committee of the Centro de Investigaciones Regionales “Dr. Hideyo Noguchi” (CIR) of the Universidad Autónoma de Yucatán and reviewed by the CSU Institutional Review Board and considered to be an exempt project (Category 4). A portion of serum samples from Nicaraguan and Mexican patients who were diagnosed as DHF/DSS, DF, or ND were de-identified and sent to CSU for metabolomics analysis.
In Nicaragua, 88 serum samples were retrospectively obtained from patients who had been diagnosed as DHF/DSS, DF, or ND (Table 1). The serum sample collection dates for the DF patients ranged from days 1 to 6 of illness and for the DHF/DSS patients from days 3 to 6 of illness. These patients had presented to the study clinic Centro de Salud Sócrates Flores Vivas and met the 1997 WHO case definition for dengue [30] or presented with undifferentiated fever, or to the Hospital Infantil Manuel de Jesús Rivera, the National Pediatric Reference Hospital, with a fever or history of fever <7 days and one or more of the following signs and symptoms: headache, arthralgia, myalgia, retro-orbital pain, positive tourniquet test, petechiae, or signs of bleeding. In Nicaragua, all samples were from pediatric patients <15 years of age, and 50% of the samples were from male and 50% from female patients. Fifty-nine positive samples were included in the analysis, and all were from patients infected with DENV-2. Of the 59 positive samples, 18 (30%) were primary infections and 41 (69%) were secondary infections (Table 1). The majority of both DF and DHF/DSS patients experienced secondary infections.
In México, 101 serum samples were retrospectively obtained from patients who were referred to the UADY clinics and had been diagnosed as DHF/DSS, DF, or ND (Table 2). The sample collection dates for the DF patients ranged from days 2 to 5 of illness, and for the DHF/DSS patients, from days 2 to 6 of illness. Approximately 10% (11) of these samples were from pediatric patients (1–15 years old) and 90% (90) from adult patients (ages 16–71 years); 53% were from female and 47% from male patients. Sixty-eight DENV-positive samples were included in the analysis. In 47 samples (69%), the infecting DENV serotype was determined: 25 patients were infected with DENV-1 (37%), and 22 (32%) were infected with DENV-2. For patients infected with DENV-1, 15 patients were diagnosed as DF and 10 as DHF/DSS. For patients infected with DENV-2, 15 were diagnosed as DF and 7 as DHF/DSS (Table 2). In 21 patients (31%), the infecting virus serotype was not determined (Table 2). For most of the Mexican patients, information on whether the patients experienced a primary or secondary infection was not available, as these clinical specimens were not collected as part of an experimental protocol.
Serum samples from patients in Nicaragua and Mexico were frozen at -80°C until thawed prior to preparation for LC-MS.
In Nicaragua, a case was considered laboratory-confirmed dengue when acute DENV infection was demonstrated by detection of DENV RNA by RT-PCR, isolation of DENV, seroconversion of DENV-specific IgM antibody titers observed by MAC-ELISA in paired acute- and convalescent-phase samples, and/or a ≥4-fold increase in anti-DENV antibody titer measured using inhibition ELISA in paired acute and convalescent samples. In Nicaragua, computerized algorithms based on the 1997 WHO schema were used to classify cases according to disease severity [7, 30, 40]. In Mexico, a case was considered laboratory-confirmed dengue by detection of DENV RNA by RT-PCR, isolation of DENV, or detection of DENV-specific IgM antibodies. Classification of dengue severity was based on the traditional 1997 WHO definitions [7, 30, 41]. The final diagnosis of DHF/DSS, DF, or ND febrile illness based upon clinical and laboratory test results was available for each patient.
In the Nicaraguan hospital study, a medical history was taken upon enrollment, and a complete physical exam was performed. Clinical data were collected every 12 hours for inpatients and every 24 hours for outpatients on Case Report Forms (CRFs) to follow patients’ disease evolution, with vital signs and fluid intake/output recorded more often as appropriate. In the Pediatric Dengue Cohort Study, febrile illnesses that met the WHO criteria for suspected dengue and undifferentiated febrile illnesses were treated as possible dengue cases and followed during the acute phase of illness by study physicians at the clinic. Cases were monitored closely for severe manifestations and were transferred by study personnel to the Infectious Disease Ward of the Hospital Infantil Manuel de Jesús Rivera when they presented with any sign of alarm [42, 43].
For LC-MS analysis, serum samples from Nicaraguan and Mexican patients were thawed on ice, and 25 μl of serum was added to cold LC-MS grade methanol (final concentration 75%) [33]. The extract was dried using a speed vacuum at room temperature, suspended in 25 μl of 100% LC-MS grade acetonitrile (ACN), and incubated at room temperature for 10 minutes (min). Following vortexing for 1 min and centrifuging for 5 min at 4°C at 14,000 rpm, 15 μl of the supernatant was transferred to a glass vial for LC-MS analysis [38, 39]. Biological samples were randomized, and the clinical diagnosis was not considered during sample preparation and data collection.
Abundance measures of MFs produced by LC-MS metabolomics analyses should be considered semi-quantitative in discovery-phase studies [36] and are influenced by instrument and technical variation. To address this, quality control (QC) for SMB measurement followed the recommendations of Dunn et al. [44] for LC-MS analysis of human biofluids. Specifically, human serum (Sigma) was purchased and aliquoted. Each aliquot was processed using the protocol for preparation of human serum samples. After drying, the aliquot was frozen at -80°C until analyzed. For each experimental analysis, the QC sample and the dried experimental samples were reconstituted at the same time. The QC sample was analyzed first by LC-MS and the results compared to QC results obtained in previous analyses. Comparisons included the number of MFs detected, abundance of the MFs, and the baseline of the total ion chromatogram (TIC) among previously analyzed runs. In addition, the QC sample was analyzed after every 15 clinical samples. If differences were detected between QC control results (either previous results or within the analysis), the analysis was stopped, the ionization source and the column (see below) were cleaned, and the mass spectrometer recalibrated. Additional samples were not analyzed until the QC analyses were satisfactory. A reference solution containing ions with m/z (mass-to-charge ratio) values 121.050873 and 922.009798 was infused directly with a capillary pump to ensure mass accuracy; the mass spectrometer continually was normalized to the intensity of these two ions.
To evaluate the reproducibility of the LC-MS analysis, the retention time (RT), and the area under the peaks of ten randomly selected representative metabolites were determined using the Nicaraguan serum specimens (N = 88). All differences in RTs and m/z values were ≤0.25 min and 15 ppm, respectively. All relative standard deviations (RSD) of the peak areas was below 25% (Table 3), confirming acceptable reproducibility of the chromatographic separation and accuracy of the mass measurements.
Analyses of the prepared serum samples were performed using an Agilent 1200 series high performance liquid chromatography (HPLC) system connected to an Agilent 6520 Quadrupole Time-of-Flight (Q-TOF) MS fitted with a dual electrospray ionization (ESI) source (Agilent Technologies, Palo Alto, CA). Metabolites were separated using a Cogent hydrophilic type-C silica diamond-hydride column (particle size 4μm, pore size 100 Å, 2.1 mm x 150 mm) with a Cogent diamond hydride guard column (size 2.0 mm x 20 mm) (Microsolv Technology Corporation, NJ) [45, 46]. A 5-μl aliquot of each processed serum sample was applied to the column that had been equilibrated with 5% solvent A (0.1% formic acid in H2O) and 95% solvent B (0.1% formic acid in ACN). Metabolites were eluted with the following nonlinear gradient formed with solvents A and B at a flow rate of 0.4 ml/min: 0.2 to 30 min, 95–50% B; 30 to 35 min, hold at 50% B; 35 to 40 min 50–20% B; 40 to 45 min 20–95% B.
HPLC column eluent was directly introduced into the Q-TOF instrument for metabolite detection. The MS parameters used were as follows: scan rate: 1.4 spectra/sec; Vcap: 4000V; drying gas (N2): 325°C at 10 l/min; nebulizer pressure: 45 psi fragmentor: 150 V; skimmer: 65V; octopole RF peak: 750 V and 2 GHz extended dynamic range mode; mass range: 100–1700 Da. Reference solution containing ions with m/z (mass-to-charge ratio) values of 121.050873 and 922.009798 was infused directly with a capillary pump to ensure mass accuracy. Mass spectra data were collected in both centroid and profile modes.
HILIC-MS data were analyzed using Agilent’s MassHunter Qualitative Analysis version B.05 software to detect molecular features (MFs) (compounds with defined accurate mass and RT) present in each sample with a minimum abundance of 600 counts, ion species H+, charge state maximum 1, compound ion count threshold 2 or more ions, and all other parameters left at default values.
MFs from the dengue diagnosis groups (DHF/DSS, DF, and ND) were compared using Agilent’s Mass Profiler Professional (MPP), version B.12.01. MFs were aligned with 0.2 min retention time and 15 ppm mass tolerance and filtered based on their presence in at least 50% of samples in at least one diagnosis group. Subsequently, MFs were baselined to the median of all samples and normalized to the 75th percentile shift. The relative abundance of each filtered MF was then compared pairwise between diagnosis groups using ANOVA and Student's t-test. For all comparisons, the false discovery rate was calculated using the Benjamini-Hochberg algorithm, and the fold change (FC) was calculated for metabolites with corrected p-values of <0.05. MFs with a corrected p-value of <0.05 and FC of >2 (positive or negative) were identified in silico when possible by interrogating the neutral mass of each in online databases [36].
The metabolites were putatively identified using Metlin [47], HMDB [48], or the Omics discovery pipeline [49]. Metlin parameters used for identification were neutral charge and mass tolerance of ± 10 ppm. HMDB parameters used were ion mode: positive, adduct type M+H, and molecular weight tolerance ± 0.01 Da. The Omics pipeline parameters used for identification included charge 0 and mass tolerance 0.01 Da. The number of database hits or other possible identities presented in S1 and S2 Tables were obtained using Metlin.
All of the MFs that statistically differentiated the DHF/DSS, DF and ND disease outcomes are listed in S1 and S2 Tables (Nicaraguan and Mexican serum differentiating metabolites, respectively). The chemical formulas were calculated using ChemCalc (Institute of Chemical Sciences and Engineering) [50]. The metabolites are listed by the Metabolomics Standard Initiative (MSI) level of identification [51, 52].
MSI level 1: Identified metabolites (experimental data matched chemical reference standards acquired on the same analytical platform).
MSI level 2: Identified metabolites (without chemical reference standards, based on physicochemical properties and spectrum similarity with public/commercial spectrum libraries).
MSI level 3: Putatively identified metabolites (based on physicochemical characteristics of a chemical class of compounds or by spectrum similarity to known compounds of a chemical class).
MSI level 4: Unidentified metabolite (unidentified or unclassified MFs that still could be differentiated or quantified based on spectrum data). These MFs could not be identified using the databases and Omics discovery pipeline [52].
Based upon their potential biological relevance, selected in silico-identified metabolites were further analyzed by targeted liquid chromatography-tandem mass spectrometry (LC-MS/MS) to corroborate their identities. When available, commercial standards were purchased, and the MS/MS spectra of the standard and the candidate SMB in the native sample were compared. If a commercial standard was not available for the in silico-identified compound, the spectrum obtained from LC-MS/MS analysis of the native sample was compared to spectra available in the NIST commercial library [53]. Many metabolites that differentiated the disease outcomes with strong p-values and FCs remain to be identified at MSI levels 1 and 2 (S2 and S3 Tables). This is due in part to the lack of commercially available standards and to lack of appropriate spectra (e.g., spectra obtained using same platforms and parameters) in the commercially available libraries.
A targeted analysis of a subset of 15 samples was utilized to validate the identification of vitamin D3 isotypes [54, 55]. To each sample, 10 ng of [2H]3-25-hydroxyvitamin D3, 10 ng of [2H]6−1,25-dihydroxyvitamin D3, and 100 ng of [2H]6-vitamin D3 internal standard were added, followed by 1 ml of cold (-20°C) acetone. Each sample was vortexed and centrifuged to precipitate protein. The supernatant was dried using a rotary evaporation device. Just before analysis, each sample was derivatized by adding 50 μl 4-phenyl-1,2,4-triazoline-3,5-dione (PTAD) solution (1 mg/ml of ACN) to each dry sample and reacting for 1 hour at room temperature.
Samples were immediately analyzed by LC-MS/MS (Agilent 6460 QQQ coupled to Rapid Resolution 1200 LC system; Agilent Technologies, Santa Clara, CA). Vitamin D3 concentrations were determined with [2H]6-vitamin D3 internal standard; 25-hydroxyvitamin D3 concentrations were determined with [2H]3-25-hydroxyvitamin D3- internal standard; 1,25-dihydroxyvitamin D3 concentrations were determined with [2H]6−1,25-dihydroxyvitamin D3 internal standard. An Agilent Zorbax C18 2.1x50 mm column was used for analysis. Buffers A and B consisted of 0.1% formic acid + 0.1% methylamine and ACN + 0.1% formic acid + 0.1% methylamine. All data were acquired in MRM mode by monitoring the methylamine adducts [54, 55]. The transitions that were monitored by MRM for the identification of vitamin D3 isotypes and the collision energy used for fragmenting each MF are shown in Table 4. Two transitions (a deuterated and a non-deuterated) were monitored for each compound [55].
Characterization of metabolites in sera by HILIC-MS revealed 15,930 MFs in Nicaraguan specimens and 17,665 MFs in Mexican specimens (Fig 1). These were further analyzed using Mass Profiler Professional (MPP) to select MFs present in at least 50% of samples of at least one diagnosis group. This yielded 744 MFs in Nicaraguan serum specimens and 861 in Mexican samples (Fig 1).
PCA plots demonstrated clustering of specimens by dengue diagnosis group in Nicaraguan samples, notably of the DHF/DSS patients (Fig 2A). In contrast, there was little evidence of clustering by diagnosis group in the Mexican samples (Fig 2B). Potential factors, such as age, gender, or infecting serotype that could contribute to the lack of clustering in the Mexican samples will be addressed below.
Many factors are known to condition dengue disease severity, including primary versus secondary infection and infecting virus serotype and genotype. PCA was used to investigate the role of these potentially confounding factors on the serum metabolome of dengue patients.
Pairwise comparisons of abundances revealed MFs in acute phase serum specimens that statistically (corrected p-value <0.05, FC >2) differentiated the DHF/DSS, DF, and ND diagnosis groups. In Nicaraguan specimens, 83 MFs differentiated DHF/DSS from DF patients, 191 MFs differentiated DHF/DSS from ND patients, and 191 MFs differentiated DF from ND patients (Fig 1). In Mexican serum specimens, 36 MFs differentiated DHF/DSS from DF patients, 313 MFs differentiated DHF/DSS from ND patients, and 309 MFs differentiated DF from ND patients (Fig 1).
MFs that statistically differentiated the dengue diagnosis groups were (when possible) given putative structural identification by interrogation of the Metlin and HMDB databases and the Omics discovery pipeline [47–49, 56, 57]. The metabolites are listed by MSI Level of identification in S1 and S2 Tables. In the Nicaraguan specimens, 13 identified metabolites (MSI Levels 1 and 2), 103 putatively identified metabolites (MSI Level 3), and 101 unidentified metabolites (MSI Level 4) differentiated the dengue diagnosis groups (S1 Table). In the Mexican specimens, 12 identified metabolites (MSI Levels 1 and 2), 120 putatively identified metabolites (MSI Level 3), and 182 unidentified metabolites (MSI Level 4) differentiated the diagnosis groups. Sixty-two of the differentiating metabolites were detected in both Nicaraguan and Mexican serum specimens (S1 and S2 Tables). Thirty-eight of the differentiating metabolites (denoted by **) exhibited a similar FC trend in the two groups; 24 metabolites exhibited an opposite FC trend (denoted by ***) in the two groups.
Thus far, the structural identities of 13 metabolites that statistically differentiate DHF/DSS, DF, and ND disease outcomes in at least one of the pairwise comparisons of the diagnosis groups have been confirmed using MS/MS (Table 4). These metabolites were grouped into biochemical classes including amino acids and lipids such as fatty acids and phospholipids, as well as vitamins.
The identities of six metabolites were confirmed (MSI level 1) by comparing the HILIC-LC-MS/MS spectrum of the candidate metabolite in the native serum with that of a commercial standard (Table 5). The spectra of MSI level 1 compounds that identified proline, α-linolenic acid (ALA), docosahexaenoic acid (DHA), lysophosphatidylcholine (lysoPC) (16:1), lysoPC (18:1) and arachidonic acid (AA), and the International Chemical Identifier (InChl) [58] for each of these metabolites are presented in S1–S6 Figs. The presence of the three vitamin D3 metabolites detected by HILIC-MS was validated (MSI level 1) by comparing the MRM spectrum of the candidate metabolite with that of a deuterated commercial standard using MRM LC-MS/MS [54, 55]. The spectra identifying endogenous vitamin D3, 25-hydroxyvitamin D3, 1,25-dihydroxyvitamin D3, and the InCHl identifiers are shown in S7–S9 Figs. The identities of four additional metabolites, myristoleic acid and three phosphatidylcholines (PCs) (34:1, 34:0, and 36:1) were confirmed (MSI level 2) by spectrum similarity with spectra in the NIST library [59].
The differentiating metabolites that were identified at MSI level 3 in Nicaraguan and Mexican samples are listed in S1 and S2 Tables, respectively. These metabolites (102 in Nicaraguan specimens and 121 in Mexican specimens) were identified in silico by interrogating online databases and libraries [47–49] and were assigned potential identities. These remain to be structurally confirmed. MFs that could not be identified in silico (101 in Nicaraguan specimens and 185 in Mexican specimens) but that were differentiated and quantified based on LC-MS spectrum data (MSI level 4) are also listed in S1 and S2 Tables.
Although the day of defervescence was unavailable for either Nicaraguan or Mexican patients, information regarding progression to DHF/DSS of patients initially diagnosed as DF was available for 31 Nicaraguan patients. These specimens were collected ≤ 4 days post onset of symptoms, presumably before the time of defervescence. Of these, 16 were collected from patients initially diagnosed as DF who later on progressed to DHF/DSS and 15 from patients who did not progress to DHF/DSS. The PCA plot revealed clustering of the patients who experienced unremarkable DF and those who progressed to severe dengue disease (Fig 4).
Statistical analysis of samples from these two patient groups yielded 65 metabolites that differentiated the eventual disease outcomes (S3 Table). Six metabolites were identified at MSI level 1 (Table 6), and all were previously identified (Table 5). The identified metabolites were proline, alpha-linolenic acid, arachidonic acid, docosahexaenoic acid, and two lysoPCs. The metabolites identified at MSI levels 3 and 4, which include 17 MFs at MSI level 3, predominantly lipids, and 44 MFs at MSI level 4, are listed in S3 Table. The potential metabolites (MSI level 3) include phosphatidylcholines, diacylglycerol, phosphatidic acid, phosphatidylserine, triglycerides, and diacylglycerophosphoglycerol.
Relative abundances of the six identified prognostic metabolites are presented in Fig 5. For this analysis, the data were further processed using Agilent Mass Hunter Quantitative Analysis software B.05.0, and the results were imported into PAST (Paleontological Statistics software package version 3.09). The abundances of the respective metabolite in DF and DF patients who later progressed to DHF/DSS patients were statistically compared using two sample t-test for unequal variances. Each of these metabolites was elevated in abundance in the DF patients that progressed to DHF/DSS compared to those who experienced DF disease.
The other metabolites listed in Table 6 and S3 Table are also candidate SMBs for progression to severe dengue disease and will be evaluated for their potential utility in predicting dengue disease outcomes.
Our studies confirm that DENV infection perturbs the human metabolome [32]. Statistical analyses indicated that many metabolites and MFs identified by HILIC-LC-MS had statistically significant differences in abundance in pairwise comparisons of the DHF/DSS, DF, and ND diagnosis groups (Table 5 and S1 and S2 Tables). Cui et al. [32] demonstrated perturbation of many of the same metabolites in DF patients during the time-course of primary DENV infection. Metabolites that were perturbed in DHF/DSS and DF patients in both Nicaraguan and Mexican patients included lysoPCs (14:0, 16:0) and long-chain polyunsaturated fatty acids such as DHA, AA, and ALA. To determine if differentiating metabolites identified by HILIC-LC-MS could be identified using a different LC-MS platform and to more thoroughly explore the metabolome, a subset of serum samples were analyzed in the Purdue Metabolite Profiling Facility (PMPF) using reverse phase (RP)-LC-MS [37]. In confirmation, 54% (117/288) of differentiating metabolites detected by HILIC-LC-MS were also detected using a T3 column (Waters, Milford, MA) in RP-LC-MS. All of the 13 differentiating metabolites whose identities were confirmed by LC-MS/MS (Table 5) were detected by RP-LC-MS and differentiated the dengue diagnosis groups.
These results provide proof of concept that differential perturbation of the serum metabolome is associated with different dengue infections and disease outcomes and that changes in relative concentrations of certain metabolites are associated with dengue diagnosis groups. Unfortunately, in this retrospective proof of concept study, a number of samples were obtained after the presumed time of defervescence and possible progression to severe disease (Tables 1 and 2). Thus, the differentiating metabolites identified in this retrospective study could represent metabolic perturbations reflecting the disease state instead of being predictive of progression to severe dengue disease. To address this issue, we compared the metabolic profiles of a subset of the Nicaraguan DF samples with those of DHF/DSS samples that were initially diagnosed as DF but then progressed to DHF/DSS (Fig 5); all of these samples were collected by day 4 of illness (presumably before the day of defervescence). Despite the small sample size, 65 metabolites differentiated the DF patients from those who progressed to DHF/DSS (S3 Table), including six of the structurally confirmed metabolites reported in Table 5 (proline, ALA, AA, DHA, and lysoPCs (16:0 and 18:1)) (Table 6). These current candidate SMBs are not specific for dengue disease but, when combined with DENV-positive laboratory test results (eg., NS1 antigen or viral RNA detection), may provide diagnosis and prognosis of DENV infection outcomes using acute-phase serum specimens.
Although these proof-of-concept prognostic metabolites are encouraging, they are based upon a small sample size and additional studies with increased numbers of patients will be needed to confirm the results. It must also be noted that these results are restricted to pediatric Nicaraguan patients. It cannot be assumed that the same metabolites will be predictive of progression to DHF/DSS in adult Nicaraguan patients or in patients from other geographic, genetic, and environmental backgrounds. Studies will be necessary to determine if these and/or alternate metabolites are predictive of progression to DHF/DSS in other patient populations.
It was surprising that PCA revealed little clustering of Mexican samples by dengue diagnosis group in contrast to the Nicaraguan samples (Fig 2A and 2B). Many factors have been demonstrated to condition dengue disease severity, including infecting DENV serotype and genotype and primary or secondary infections [24, 60, 61], which could have confounded the analyses. The Mexican patients were infected with either DENV-1 or DENV-2; only DENV-2 was detected in Nicaraguan patients. To explore reasons for the lack of segregation of dengue diagnosis groups in PCA, we analyzed Mexican samples stratified by infecting virus serotype (DENV-1 versus DENV-2). PCA and statistical analyses revealed significant differences in the perturbation of the serum metabolome of Mexican patients attributable to the different serotypes (Fig 3A, S4 Table). In this regard, the different numbers of DENV2 infections in Mexico and Nicaragua (22 and 59, respectively, could have confounded the results (Tables 1 and 2). We also explored the potential role of primary versus secondary infections in perturbation of the serum metabolome of dengue patients. Immune status was only available for the Nicaraguan samples, which were stratified by primary versus secondary infection and analyzed. PCA plots revealed clustering of patients by primary versus secondary infection, and analyses revealed multiple metabolites that differentiated infection by immune status (Fig 3B and S5 Table). This analysis of the Nicaraguan samples clearly demonstrates differential perturbations according to immune status, which are likely occurring in Mexican patients as well. Unfortunately immune status was only known for a few of the Mexican patients. Clearly the differences demonstrated for the Nicaraguan samples could have confounded the analyses of the Mexican samples. In addition the Mexican samples differed from the Nicaraguan samples in age distribution (the effect of age on dengue disease severity and metabolome perturbations is addressed below). All of these and other factors could have contributed to the lack of clustering in the Mexican patients by diagnosis group, and additional metabolomics studies will be necessary to identify the actual mechanisms involved.
Thus, although the available sample sizes were relatively small in this proof-of-concept study, PCA plots revealed clustering of patients by both infecting virus serotypes (Fig 3A) and by primary versus secondary infection, (Fig 3B) and differentiating metabolites were identified for each comparison (S4 and S5 Tables). Interestingly, none of the differentiating metabolites in these two analyses overlapped, suggesting the involvement of different metabolic pathways or mechanisms. It is also interesting that these metabolites differ from those reported in Table 5 that differentiated the dengue disease diagnosis groups (DHF/DSS versus DF, DHF/DSS versus ND and DF versus ND).
Differentiating metabolites identified in this study provide insights into fundamental metabolic mechanisms and pathways that condition DENV infection, replication, and pathogenesis in humans, and several are potentially biologically and physiologically relevant in terms of severe disease outcomes (DHF/DSS). Some are associated with lipid metabolism and regulation of inflammatory processes controlled by signaling fatty acids and phospholipids. Others are associated with immune regulation, endothelial function, and vascular barrier function, which is provocative in the context of the central role of vascular leakage in the pathogenesis of DENV infection and the possible progression to shock in DSS [3, 4, 7, 30].
DENV replication is dependent on host cell lipid biosynthesis and metabolism. Viral replication complexes are enclosed in endoplasmic reticulum-derived double-membrane vesicles that organize and localize the complexes to facilitate the exchange of components with the cytosol for genome replication and virus assembly [37, 62, 63]. Long-chain polyunsaturated fatty acids such as DHA (C22:6) and ALA (C18:3) were increased in abundance in DHF/DSS versus DF and DHF/DSS versus ND groups in Nicaraguan serum samples and in early DF that progressed to DHF/DSS (Tables 5 and 6). Long-chain omega-3 polyunsaturated fatty acids such as DHA and its precursor ALA are potent anti-inflammatory agents and have been previously reported to be elevated during DENV infection [32]. DHA has been shown to decrease the production of inflammatory eicosanoids, cytokines, and reactive oxygen species [64, 65]. This molecule can act both directly by inhibiting AA metabolism and indirectly by altering the expression of inflammatory gene products [66, 67]. DHA also is a precursor of a family of anti-inflammatory mediators called D-series resolvins [66, 67]. The increases in DHA levels we observed in dengue patient serum might represent the host attempt to mitigate immunopathology of dengue disease.
AA and its metabolites have been shown to be elevated in plasma at different stages of infection in dengue patients [32, 68]. This was confirmed in our results; we found AA levels elevated in DHF/DSS patients compared to DF patients in both Mexican and Nicaraguan populations. AA is mobilized from phospholipids in cell membranes and is metabolized by cyclooxygenases and lipoxygenases to pro-inflammatory eicosanoids such as prostaglandins, thromboxanes and leukotrienes [69, 70]. We detected significant changes in abundances of several of these AA metabolites when comparing the dengue disease groups (Tables 5 and 6).
A number of phospholipid metabolites differentiated dengue diagnosis groups in patients from both Nicaragua and Mexico (Table 5 and S1 and S2 Tables). The increases we observed in phospholipid biosynthesis make biological sense given that host cell phospholipid metabolism is known to be influenced by DENV replication in both mosquito and human cells through DENV NS3 protein-mediated redistribution and activation of fatty acid synthase [37, 61]. The prevalent phospholipids found to be increased primarily contain C16 and C18 unsaturated acyl chains. Palmitic acid (C16) and oleic acid (C18) have been found to be increased in DENV-infected mosquito cells and to facilitate production of lysoPCs. Phospholipids are precursors of lipid mediators, such as platelet activating factors (PAFs) and eicosanoids, which are involved in inflammatory responses [70, 71].
LysoPCs (18:1, 16:0) were elevated in acute-phase serum specimens of DHF/DSS patients (Table 5). These single fatty acid chain lipids are involved in alteration of membrane structures and can mediate acute inflammation and regulate pathophysiological events throughout the vasculature and at local tissue sites [72–75]. Interestingly, lysoPCs may alter homeostasis of vascular endothelium, causing endothelial cell instability, barrier dysfunction, and vascular leakage, a major component of the pathophysiology of DSS [18, 76–78]. Previous reports demonstrated perturbation of lysoPC concentrations in DENV-infected human serum [32, 37]. Up-regulation of the phosphatidylcholine biosynthesis pathway in acute DENV infections (days 1–3) was identified as one of the predictors for progression to DHF [19]. Other metabolites from different biochemical classes differentiated dengue disease outcomes. For example, we observed lower levels of 1,25-dihydroxyvitamin D3 (1,25-vitD3) in DHF/DSS versus DF and ND in Nicaraguan patients (Table 5). Reduced levels of 1,25-vitD3, with its roles in immunoregulation and vascular barrier function, could be involved in the immunopathophysiology associated with DHF/DSS [79]. A decrease in serum 1,25-vitD3 levels is associated with increased mortality in sepsis patients [80, 81]. The active form of vitamin D3 (1,25-vitD3) can be synthesized in vascular endothelium following stimulation of vitamin D3 1α-hydroxylase activity by inflammatory cytokines. Interactions of this metabolite with endothelial cells and the reduction of 1,25-vitD3 observed in immune-mediated diseases by others [79, 82] prompts speculation about the potential role of decreased concentrations of this metabolite in patients progressing to DHF/DSS. Interestingly, polymorphisms in the vitamin D receptor gene are linked with severe dengue disease outcomes [14].
Several amino acids or peptides were also found to differentiate disease outcomes. For example, proline, which can act as a modulator of the intracellular redox environment, differed in DHF/DSS and DF patients who were initially diagnosed as DF (Fig 5). Perturbations of proline in endothelial cells could affect endothelium function [83–85].
Clearly, metabolomics provides new opportunities and a powerful approach to investigate potential viral, host, pathogenic, and immunologic determinants of DENV infection and pathogenesis.
Identification of metabolites that differentiate dengue disease outcomes in patients from different geographic areas, environmental conditions, genetic backgrounds, sexes, and ages [34, 35] is an important first step in selecting SMBs for dengue diagnosis and prognosis. We have already identified a large, overlapping set of metabolites that differentiated dengue outcomes in genetically and geographically distinct populations. However, the associations were not always concordant. In some instances, a metabolite differentiated the disease outcomes in one study population but not in the other. For example, lysoPCs (16:0 and 18:1) statistically differentiated DHF/DSS and DF diagnosis groups in Nicaraguan patients but not in Mexican patients (Table 5). In other instances, a candidate SMB was either increased or decreased in abundance in serum from patients from one country and the opposite trend occurred in patients from the other country. Sixty-two of the differentiating metabolites were detected in serum specimens from the Nicaraguan and Mexican patients (see Table 5 and S1 and S2 Tables). Thirty-eight of these differentiating metabolites (denoted by **) had similar FC trends in Nicaraguan and Mexican patients, but 24 (denoted by ***) had opposite FC trends. For example, ALA and DHA exhibited positive fold-changes in differentiating DHF/DSS from DF and ND outcomes in Nicaraguan patient sera but negative trends in Mexican patients.
A number of factors could have contributed to the dissimilar change trends in the two populations. For example in this exploratory metabolomics study, no Mexican patient was officially diagnosed as DSS (although some were hospitalized and diagnosed with DHF), while 15% of the DHF/DSS patients in Nicaragua were classified as DSS. The lack of DSS cases in Mexico is a limitation of our study that may have confounded the identification of SMBs that differentiate DSS from non-DSS disease outcomes. In addition, there were major age differences in the two study populations. The age of the patient can condition dengue disease severity [86, 87]. Severity of symptoms (which can be subjective) that influence clinical diagnosis and disease classification in the two populations could also be strongly influenced by the age of the participants. Ninety percent of the Mexican patients were >15 years of age and may have been less likely to progress to DSS even though they were diagnosed as DHF patients. All of the Nicaraguan patients were children <15 years of age. Human metabolic profiles are age-dependent [88], and DENV pathophysiology and clinical symptomology (e.g., DHF) can differ in different age groups [86, 87, 89] and by sex [60]. DSS is negatively correlated with age [13]. Liver inflammation (an important target organ in DENV infection) is more prevalent in children than in adults [90, 91].
In this regard, we conducted a very preliminary analysis to examine the potential role of age on the DENV-infected serum metabolome. The Mexican patients with DF or DHF/DSS ranged from 1 to 62 years of age. These DENV-infected samples were stratified by age; pediatric patients <15 years of age (N = 11) and adult patients >15 years of age (N = 57), and the serum metabolites were characterized by PCA (S10 Fig). Because of the limited number of pediatric patients, in this preliminary analysis we used a filter of 25% metabolite presence in samples of one of the diagnosis groups instead of our standard filter of 50% metabolite presence. Despite the small number of pediatric patients, clustering of patients by age was evident (S10 Fig). Clearly, age differences could have contributed to the metabolomic differences between the two groups. The effect of age on the serum metabolome during DENV-infections will be a fruitful area of research. Identification of metabolites that differentiate age groups could provide important insights into differences in the pathophysiology of DENV infections in pediatric and adult patients [86, 87].
Other factors could also contribute to the dissimilar change trends in metabolite abundance in the two populations. Dietary differences between Nicaraguan and Mexican patients could confound results with metabolites such as ALA, which is obtained principally from dietary plant sources, and DHA, which is a metabolite of ALA. All of these factors could account for some of the metabolite differences seen in the two study populations. Further studies will be necessary to determine if metabolites such as lysoPCs and DHA are candidate SMBs for progression to DSS in older patients and in patients from other geographic areas.
We are currently conducting a prospective clinical study in Managua, Nicaragua, to determine the diagnostic and prognostic potential of existing and yet to be identified candidate SMBs. MRM analysis [92] will be conducted for accurate quantification of abundance of metabolites in different disease states as part of the evaluation of the potential diagnostic utility of candidate SMBs. Super learner analysis [93] is being used to identify the most parsimonious SMB “biosignature” in acute phase serum specimens that, when combined with other laboratory and clinical information, such as NS1 antigen detection and viral RNA detection by RT-PCR, will provide the most efficient algorithms for dengue diagnosis and prognosis. This would be of enormous value for appropriate patient triaging, management and clinical care. Prospective clinical studies will allow us to increase the number of early acute-phase patients and to identify additional metabolites that are predictive of progression to severe dengue disease. Additional clinical studies potentially will also allow us to increase the number of patients who progress to DSS and to identify metabolites that differentiate DHF and DSS disease outcomes [28]. The studies will also provide insights into metabolic pathways and pathogenic mechanisms that condition DHF and DSS outcomes. Such information could potentially be exploited in the development of new therapeutics for treatment of dengue patients in danger of progressing to DSS [28]. The 3- to 4-day window from dengue disease onset to defervescence provides a unique opportunity for therapeutic intervention [3–5, 7, 30]. Clearly, metabolomics provides new opportunities for diagnosis and prognosis of DENV infections.
|
10.1371/journal.pntd.0007301 | Strategies for tackling Taenia solium taeniosis/cysticercosis: A systematic review and comparison of transmission models, including an assessment of the wider Taeniidae family transmission models | The cestode Taenia solium causes the neglected (zoonotic) tropical disease cysticercosis, a leading cause of preventable epilepsy in endemic low and middle-income countries. Transmission models can inform current scaling-up of control efforts by helping to identify, validate and optimise control and elimination strategies as proposed by the World Health Organization (WHO).
A systematic literature search was conducted using the PRISMA approach to identify and compare existing T. solium transmission models, and related Taeniidae infection transmission models. In total, 28 modelling papers were identified, of which four modelled T. solium exclusively. Different modelling approaches for T. solium included deterministic, Reed-Frost, individual-based, decision-tree, and conceptual frameworks. Simulated interventions across models agreed on the importance of coverage for impactful effectiveness to be achieved.
Other Taeniidae infection transmission models comprised force-of-infection (FoI), population-based (mainly Echinococcus granulosus) and individual-based (mainly E. multilocularis) modelling approaches. Spatial structure has also been incorporated (E. multilocularis and Taenia ovis) in recognition of spatial aggregation of parasite eggs in the environment and movement of wild animal host populations.
Gaps identified from examining the wider Taeniidae family models highlighted the potential role of FoI modelling to inform model parameterisation, as well as the need for spatial modelling and suitable structuring of interventions as key areas for future T. solium model development. We conclude that working with field partners to address data gaps and conducting cross-model validation with baseline and longitudinal data will be critical to building consensus-led and epidemiological setting-appropriate intervention strategies to help fulfil the WHO targets.
| Taenia solium infection in humans (taeniosis and neurocysticercosis) and pigs (cysticercosis) presents a significant global public health and economic challenge. The World Health Organization has called for validated strategies and wider consensus on which strategies are suitable for different epidemiological settings to support successful T. solium control and elimination efforts. Transmission models can be used to inform these strategies. Therefore, a modelling review was undertaken to assess the current state and gaps relating to T. solium epidemiological modelling. The literature surrounding models for other Taeniidae family infections was also considered, identifying approaches to aid further development of existing T. solium models. A variety of different modelling approaches have been used for T. solium including differences in structural and parametric assumptions associated with T. solium transmission biology. Despite these differences, all models agreed on the importance of coverage on intervention effectiveness. Other Taeniidae family models highlighted the need for incorporating spatial structure when necessary to capture aggregation of transmission stages in the environment and movement of animal hosts.
| Infection by the cestode Taenia solium contributes to a significant and underreported public health and economic burden in low and middle-income countries [1, 2]. A transmission cycle including humans and pigs is facilitated by the free-roaming behaviour of pigs in subsistence and minimal biosecurity farming environments [3, 4]. Humans become definitive hosts when consumption of raw or undercooked cyst-infected pork leads to the tapeworm infection taeniasis (henceforth referred as taeniosis as per Kassai et al. [5]). Humans can also act as accidental intermediate hosts when T. solium eggs are ingested. In this instance, migration of the larval stage of T. solium to the central nervous system can result in neurocysticercosis (NCC) [6]. Human cysticercosis especially occurs in high-risk settings where poor hygiene and sanitation standards prevail [7, 8].
NCC is associated with epilepsy and a recent review found that 31.5% of epilepsy cases could be due to NCC in endemic settings [9]. The Foodborne Disease Burden Epidemiology Reference Group (FERG) under the World Health Organization (WHO) estimated that NCC-associated epilepsy accounted for approximately 2.8 million disability-adjusted life years (DALYs) globally in 2010, concluding that NCC contributed the largest number of DALYs in a list of priority foodborne parasites [10]. In addition to its impact on public health, T. solium infection in pigs is associated with a substantial economic burden due to the decreased market value of infected pigs [11, 12] and market distortion resulting from farmers adopting informal avenues for selling infected meat and animals [13, 14].
Combatting the burden associated with T. solium infection was initially recognised in the WHO “Global Plan to Combat Neglected Tropical Diseases (2008–2015)” [15] and by WHO Member States at the World Health Assembly [16]. More specifically, the 2012 WHO roadmap on neglected tropical diseases (NTDs) [17] set out the goal of scaling up interventions for T. solium in selected countries by 2020. This target was predicated on having achieved, by 2015, the establishment of a validated strategy to meet such a goal. Despite the declaration by the WHO of being ‘tool ready’ for pig-, human-, and environment-orientated interventions [18], the effective implementation of such intervention tools in endemic settings will present considerable challenges. It is likely that interventions will need to be tailored to local epidemiological circumstances, local pig husbandry practices and socio-cultural behaviours [19]. Even with epidemiological setting-appropriate strategies identified, a framework for supporting and implementing needs to be present within a control strategy. Braae et al. have proposed such a framework towards the control and elimination of T. solium [20].
Infectious disease modelling can support T. solium control and elimination strategies by improving understanding of the key transmission dynamics processes that shape epidemiological patterns and by comparing, optimising, and estimating the cost-effectiveness of tailored strategies applicable for control in local settings [21,22]. Following the 2012 London Declaration on NTDs [23], an international collaboration of infectious disease modellers emerged under the umbrella of the NTD Modelling Consortium (https://www.ntdmodelling.org/) to provide modelling and quantitative support to address questions surrounding the feasibility of achieving the WHO 2020 call targets with current or alternative/complementary strategies. For example, outputs using multi-model comparisons and field data have improved knowledge of epidemiological processes, such as examining the feasibility of Onchocerca volvulus elimination in Western Africa [24], or cross-validation with epidemiological data has enabled consensus-based evidence to emerge, as seen with the development of alternative mass drug administration guidelines to target lymphatic filariasis elimination [25].
In order to develop a comprehensive research agenda towards formulation of cost-effective strategies for the control and elimination of T. solium taeniosis/cysticercosis in the context of the WHO NTD 2015/2020 call for T. solium, this article seeks to compare and identify gaps in existing T. solium transmission dynamics models. We follow the approach of Nouvellet et al. [26] and Pinsent et al. [27] who have synthesized and compared a wide range of models for Chagas disease and trachoma, respectively. By assessing the current state of the field, we highlight differences in structure of published models, sources of uncertainty and the data used to motivate, inform and parameterise such models. We compare the main conclusions drawn from each model and uncover knowledge gaps related to model complexities and data needs. In addition to a comparison of T. solium transmission models, we review models representing the other members of the Taeniidae family to consider where future development of existing T. solium models may be focussed. We hope this work will therefore form the basis for improved dialogue between field epidemiologists, programme managers, and modellers.
We conducted a systematic review of modelling studies to understand population dynamics or effects of interventions caused by members of the Cestoda: Taeniidae family (i.e. Echinococcus, Taenia). A systematic review, conducted by Atkinson et al. [28], had focussed on assessing Echinococcus models only and has been consulted to corroborate our findings in this review. We performed a search for eligible studies in PubMed, without date or language restrictions, in January 2018, using the search terms: (Taeni* OR Echino* OR Cesto* OR cysticerc* OR hydatid*) AND (model OR models OR modelling OR modeling OR simulat*) AND (dynamics OR transmission OR control).
The PubMed search output was reviewed by the following method: 1) title and abstracts were reviewed and articles were excluded if they were related to parasites or diseases different from those relating to the Taeniidae family; 2) all full texts were retrieved from those abstracts that met the inclusion criteria; 3) each article was reviewed for descriptions of mechanistic transmission models with specifications that addressed parasite prevalence, incidence, or intensity. Models addressing only spatial distribution or parasite abundance within a single host (i.e. not considering transmission between host species), or risk assessment models that did not consider explicitly transmission processes, were excluded. Literature found through the systematic search was supplemented by specific searches of references and papers known to the authors or cited in the papers obtained (Supplementary S1 Flow Diagram). Papers based on re-application or minimal modifications to the original models were excluded. Identified models were divided into groups based on model type and characteristics (Table 1). Geographic distribution of locations where models have been developed and applied are presented in Fig 1. This review is compliant with the PRISMA checklist for systematic reviews [29] and available in Supplementary S1 Table.
Identified studies were then analysed and data extracted based on the following headings: Reference, Year, Title, Journal, Parasite genus, Parasite species, Motivation, Type of model (including further specificities of model type), Nature of model (including whether the model represents or not the totality of the transmission cycle, i.e. Full transmission vs. partial model), Role of stochasticity, Representation of population dynamics, Explicit representation of spatial transmission, Spatial design, Parameterisation/calibration for specific setting(s), Hosts (states) represented, Explicit representation of Environment, Source of parameters, Major assumptions and model simplification(s), Assessment of parametric uncertainty, Interventions modelled, Model validation (informal/formal), and Main findings. The full data extraction tool is available in Supplementary S1 File.
A systematic search of the literature yielded 23 papers plus two papers known to authors, and three identified through additional searches, for inclusion in the analysis. Of these, four studies modelled T. solium exclusively; 20 modelled infection by Echinococcus spp., one focussed solely on Taenia ovis, one on T. ovis and Taenia hydatigena, and the remaining two addressed Echinococcus spp. and Taenia spp. (other than T. solium) infections (T. ovis and T. hydatigena).
Results are first presented with an in-depth analysis of T. solium dynamic transmission models, followed by an assessment of the other Taeniidae family transmission models to identify possible modelling gaps and areas for future development of T. solium dynamic transmission models.
Analysis of the T. solium papers revealed four models that could be classified as dynamic transmission models (Table 2). Different modelling approaches are used to simulate T. solium transmission, including a decision tree/stochastic simulation approach in Gonzalez et al. [35]; deterministic and stochastic versions of a Reed-Frost model in Kyvsgaard et al. [42]–a chain binomial model whereby chains of infection are generated by the assumption that infection spreads between individuals in discrete units of time under the binomial probability distribution [58]; an individual-based, stochastic model, cystiSim [53]; and a population-based, deterministic model, EPICYST [56]. Representation of the T. solium life cycle is captured with varying degrees of complexity within each model. Similarities and differences between the four dynamic transmission models are subsequently compared based on T. solium life-cycle and transmission features.
Kyvsgaard et al. [42] incorporates compartments for human taeniosis and porcine cysticercosis but does not consider heterogeneity in host infection states, such as age dependency or infection burden. Equally Kyvsgaard et al. [42] assume a human to pig transmission probability of 0.01, without providing evidence to support this. While Gonzalez et al. [35] also omits any infection heterogeneity, more complexity is introduced as human states include those infected with maturating stages of the adult tapeworm, and infection and antibody presence in pig compartments. In addition, a pig-population stochastic sub-model is implemented to simulate population dynamics in the absence of infection [35]. The cystiSim model [53] features heterogeneity in both exposure and infection by modelling high (resulting from direct coprophagia) and low (resulting from indirect environmental exposure) burden infections in pigs, along with age-dependent human infection. The EPICYST model [56] assigns a proportion of the infected pig population into high or low burden states and incorporates different transmission mechanisms in the life cycle; a density-dependent process for pig and human exposure to eggs and a frequency-dependent process [59] for human exposure to cysts in pork.
Both the Kyvsgaard et al. [42] and Gonzalez et al. [35] models do not explicitly model infection in the environment, although an ‘infection potential’, analogous to environmental contamination, is generated in Gonzalez et al. [35]. This is based on the number of adult tapeworms and humans in the ‘post-infection contamination’ stage, with the latter produced by a fixed-delay in transmission reduction, which can be varied depending on different climatic and hygienic conditions as specified by parameter inputs. The number of eggs in the environment is explicitly modelled in EPICYST [56], while cystiSim [53] defines environmental contamination as an attribute of previous tapeworm carriers with removal of eggs implemented using an exponential decay function based on environmental studies of Taenia saginata egg survival on pastures and expert opinion [60]. Considerable uncertainty surrounds the rate at which T. solium eggs decay in the environment, reflected in the use of egg survival studies from other Taeniidae species to inform parameterisation (Table 3). The sensitivity analysis conducted in EPICYST [56] of the model output (cumulative number of human cysticercosis cases) to model parameters, indicated egg death rate as a highly influential and uncertain parameter, highlighting the need for more research into T. solium egg environmental viability and whether heterogeneity exists between settings.
Several consistencies and differences emerge across the T. solium models in relation to assumptions on host immunity (Table 3). For example, there is no inclusion of human immunity for taeniosis across the models, although Kyvsgaard [42] indicate that “spontaneous elimination of the parasite” occurs without providing details. Natural recovery (in the absence of interventions) from porcine cysticercosis is only modelled in Kyvsgaard et al. [42], with pigs transferred to a recovered compartment given a certain probability, and subsequently develop assumed “life-long” immunity given short life-expectancies. A breeding sow would likely live much longer and outlive this period, although determining the contribution of these animals to transmission is unclear given this sub-population does not generally represent slaughtering stock. Equally, the presence of natural recovery from porcine cysticercosis is unclear, given that average pig life expectancy is low in many settings and needs further clarification from field data. Protective immunity is only included following treatment and recovery of infected pigs [61,62] in both cystiSim [53] and EPICYST [56] for a period of 3 months.
The pig host immune response is more directly modelled by Gonzalez et al. [35]. Firstly, if born to a serologically positive sow, pigs produce antibodies (modelled as being enzyme-linked immunoelectrotransfer blot [EITB] positive); it is assumed that these antibodies persist for a period of 8 months, although the pig may still acquire infection during this time (therefore being already EITB positive when infected). Secondly, pigs born to serologically negative sows become EITB positive with a delay of 15 days following larval infection with immature cysts. These infected pigs progress to infection with mature cysts after a delay of 75 days and remain EITB positive. Only (simulated) treatment clears infection as pigs move to and remain in the treated state, indicating that they are resistant to re-infection following treatment [63]. The modelling approach taken by Gonzalez et al. [35] also calls into question the need to include a “diagnostic layer” in the other T. solium transmission models to represent outcomes from serological data (both antibody and antigen in human and pig hosts) which may not directly equate to underlying true infection status in the hosts as performed in onchocerciasis [24] and Chagas disease modelling [64].
Gonzalez [35] and cystiSim [53] model the maturation of the adult tapeworm, from infected humans (with taeniosis) harbouring immature and mature adult tapeworms, while EPICYST [56] considers only mature tapeworms (for the human taeniosis infected compartment) and ommiting the pre-patent period as this is assumed to be 5–10 weeks compared to a significantly longer human life expectancy duration. Kyvsgaard et al. [42] uses the prepatent period to set the time-step for the chains of infection. Across the Gonzalez [35], Kyvsgaard et al. [42] and cystiSim [53] models, the pre-patent period is defined as 3 months although this is based on data from other Taeniidae species including T. saginata and Echinococcus multilocularis (Table 3).
Further parameters related to the adult tapeworm life history also vary between the models including the egg production rate, which is identified as an influential and uncertain parameter in the EPICYST sensitivity analysis [56], and the assumed average life span of the adult tapeworm reflecting the limited data associated with the adult tapeworm dynamics (Table 3). This has a direct bearing on the estimated basic reproduction number (R0) of T. solium and accounts for some of the variability in the estimates of R0 between EPICYST [56] (R0 of 1.4, 95% credible Interval: 0.5–3.6) and the Kyvsgaard et al. [42] (R0 of 1.75) models. The R0 estimated in Kyvsgaard et al. [42] does not consider pig infections, with the calculation based on the summation of new infected humans over time, although there is no distinction between new human cases and those continually re-infected, and this definition erroneously produces units of time for R0. By contrast, R0 calculated from EPICYST [56] reflects the whole system of transmission among pigs, humans and the environment. Further noting that T. solium is not dioecious but it is hermaphrodite species, the classical R0 for helminths is strictly only compatible with an intensity-based modelling framework. The R0 however as estimated in EPICYST [56] still provides a useful and valid threshold quantity for comparison, given that R0 for T. solium has been estimated exclusively to date using so-called microparasitic prevalence modelling frameworks.
Human-directed interventions are simulated in all four dynamic transmission models (Table 4). Mass drug treatment irrespective of infection status are simulated in Gonzalez et al. [34], Kyvsgaard et al. [42] and cystiSim [53], while EPICYST [56] currently models a hypothetical test-and-treat (T&T) intervention based on the possible future availability of a specific and sensitive point-of-care test for taeniosis, although current diagnostics lack either or both sensitivity and specificity [66, 67], or in the case of a highly specific coproantigen test [68], are not commercially available yet. For example, the human-directed intervention modelled in EPICYST [56] is a hypothetical approach based on the rES33 EITB for antibody detection [69,70], which has substantially lower specificity than that currently modelled and most intervention studies measuring human taeniosis use the coproantigen ELISA test [71,72]. Another potential limitation across models is that human treatment efficacy may be lower in field settings compared to currently assumed estimates. However, efficacy can be adjusted by the user in cystiSim [53] and EPICYST [56] with the available code, allowing adaptation of the model to a given treatment efficacy.
In the pig host, mass treatment (using oxfendazole) and/or vaccination (e.g., the TSOL18 vaccine [73]) are simulated in all models except Gonzalez et al. [35], where only pig mass treatment is simulated. Pig-directed interventions achieve high efficacy from field studies [74,75] and this is reflected in the models. For example, the treatment efficacy of oxfendazole is assumed to range from 90% in cystiSim [53] to 99% in EPICYST [56] to 100% in both Gonzalez et al. [35] and Kyvsgaard et al. [42]. Pig vaccination efficacy has also been assumed to be high, having been set to 100% in Kyvsgaard et al. [42], 99% with an adjustment to account for the fact that some piglets may become infectious before a full course of vaccine can be administered in EPICYST [56], and 90% in cystiSim [53], where vaccination was combined with treatment in all modelled scenarios. While cystiSim [53] and EPICYST [56] permit user-specified efficacy changes, cystiSim [53] has the added benefit of allowing for age-targeted interventions in both pigs and humans.
Coverage of human- and pig-targetted interventions is included as a parameter in all models, with coverage levels fixed at 90% Kyvsgaard et al. [42], but varied in the EPICYST [56] sensitivity analysis and across intervention scenarios for cystiSim [53] and Gonzalez et al. [35]. Behavioural and environmental focussed interventions have also been simulated using EPICYST [56] and Kyvsgaard et al. [42], including improved sanitation, husbandry, meat inspection and cooking practices, by modifying nominal values of certain transmission parameters. A key finding across the models is that human- and pig- targeted interventions are generally sensitive to coverage levels (Table 4), although these interventions are more robust to changes in coverage compared to behavioural and environmentally focussed interventions in the EPICYST sensitivity analysis [56]. One important quantity that could therefore be estimated is the minimum fraction of pigs to be vaccinated to achieve transmission interruption and infection elimination. Limitations with current modelling of interventions, especially on the effectiveness of human-directed intervention approaches and on the realistic, achievable coverage levels, emphasise the need to design intervention simulations in conjunction with research groups involved field intervention trials. Equally, simulations need to be compared with data collected during interventions implemented in the field. For example, it is planned that cystiSim [53] predictions will be compared with data collected in Zambia as part of the CYSTISTOP programme and used to update model inputs from longitudinal infection data and to inform parameters of interest including pig population turnover and actual coverage [76].
The existing models need to be tested to determine their ability to accurately model field-specific targeted interventions. For example, cystiSim was used to test targeted anthelmintic treatment in school-age children given the age-structure of the host human, replicating the approach taken in Braae et al [77]. Requirements to model other targeted interventions, such as the inclusion of spatially explicit structure to capture ring screening/treatment strategies, as applied in northern Peru [71], will need further consideration.
The FoI models of Roberts et al. [31,32] were also used to estimate the R0 of E. granulosus, T. ovis, and T. hydatigena and to determine the equilibrium steady-state. Similar FoI models could be fitted to T. solium age-prevalence data and, if available, age-abundance data from pigs and humans, as already performed in Ecuador and Zambia [78,79], but applied to a wide variety of epidemiological settings to support setting-specific model parameterisation. The egg to human and egg to pig transmission coefficients were identified in the EPICYST sensitivity analysis [56] and could therefore be informed through FoI estimation. This approach could also be used to investigate different assumptions on age-exposure patterns, for example by implementing age-dependent, age-truncated or dynamic FoI modifications to the FoI models [43, 52] and acquisition of immunity. Age-dependent infection is incorporated into human dynamics in cystiSim [53], however FoI modelling could help to inform further age-dependent infection processes in pig and human populations in cystiSim [53] and EPICYST [56]. For example, there is some evidence for specific age trends in taeniosis infection, with the highest prevalence’s found in younger age groups as identified in the Democratic Republic of Congo [80], Peru [81], and Guatemala [82], which could be a result of protective immunity in older individuals or age-specific meat consumption trends. Age-stratified taeniosis prevalence data could support FoI modelling to better understand the rate of recovery from taeniosis, identified as an influential and uncertain parameter in the EPICYST sensitivity analysis [56]. The rate of human pork meal procurement was also considered a significant parameter in the EPICYST sensitivity analysis [56], so risk-factor analyses, such as those conducted in Western Kenya [83], could refine the uncertainty around this nominal parameter value in different settings. Fitting appropriate distributions such as the negative binomial distribution to T. solium cyst abundance data from pigs, could help to better determine the degree of infection aggregation, possibly indicative of heterogeneous exposure and support modelling overdispersion explicitly as performed for E. multilocularis worm burden in foxes [45].
Transmission dynamics models of Echinococcus spp. reveal an interesting split between modelling approaches. Deterministic, population-based transmission models have been used primarily for E. granulosus incorporating dogs as the definitive host, sheep or other livestock as the intermediate host and humans acting as accidental intermediate hosts [49, 50, 54]. The exception is Huang et al. [48] where an individual-based model of E. granulosus was developed to study dynamics in a small community, an approach also applicable to the simulation of T. solium in small communities. Wang et al. [54] extends the models previously developed [49,50] by devising an approach to tackle parameter estimation issues concerning egg dynamics in the environment. The model incorporates infection delays as distributed time delays of infection between hosts, with different distribution functions chosen to reflect differences in the range of host movement (e.g., livestock, humans), and may provide an alternative approach for modelling transition between infection stages in the T. solium transmission models.
E. multilocularis transmission models were initially structured within deterministic, population-based frameworks [34, 38], also extending these approaches to consider optimal control through an economic lens [47]. Takumi & Van der Giessen, 2005 [40] also present an E. multilocularis deterministic model which tracks the mean number of transmission stages in hosts rather than measuring prevalence to replicate more accurately the rebound to pre-control of adult worm prevalence seen after cessation of a deworming campaign, even when substantial reductions are initially achieved. The impact on the rate at which average worm burdens return to pre-control levels, following cessation of community chemotherapy interventions has been further demonstrated for other helminths [84]. Modelling of E. multilocularis transmission dynamics diverges significantly from E. granulosus modelling through the development of individual-based stochastic dynamics [37, 44, 57] in the definitive host (e.g., foxes) to capture stochasticity in the demographic and infection processes in the wild animal populations that drive transmission. Heterogeneities in local pig populations and differential pig foraging behaviours [85] may be better captured by similar individual-based techniques, although these behaviours may be difficult to parameterise reliably and will likely be seasonally- and husbandry/management system-specific.
Recognising that environmental contamination is spatially aggregated [40], E. multilocularis transmission models [37, 38, 45, 57] and a specific T. ovis deterministic transmission model [51] have introduced spatial dynamics by a variety of approaches (Table 5). Spatial heterogeneity in T. solium transmission is undoubtable and has been identified in a number of settings, with the detection of local clusters of pig cysticercosis prevalence and incidence [86,87], and clustering of pig cysticercosis infection (or seropositivity) near to human taeniosis carriers [88–90]. This may indicate the presence of spatially-aggregated environmental contamination of T. solium eggs and suggests spatially heterogeneous transmission. There is some evidence to suggest that other mechanisms are involved in the spatial distribution of T. solium eggs in the environment, such as the possible role of dung beetles acting as mechanical vectors for egg dispersal [91] and could be involved in a complex interplay with pig behaviour and seasonal factors [4]. Movement of individuals (humans and/or pigs) between communities may also play an important role in T. solium transmission and will influence the likelihood of sustaining elimination or experiencing resurgence [35]. Inclusion of spatial dynamics, however, should not detract from resolving the structural and parametric uncertainties that affect the current non-spatial models.
Another feature explored in the E. multilocularis modelling papers is the impact of seasonal variation, by seasonal forcing of transmission models to account for differences in egg viability and movement of wild animal populations between seasons, for example by describing egg decay as a function of temperature [57]. The Gonzalez et al. [35] T. solium transmission model begins to consider the possibility that T. solium dynamics may be influenced by climate; however, there is little information available to estimate the effect of temperature (and other environmental variables) on T. solium egg viability in natural conditions. These factors might also affect transmission differently depending on the endemicity level, e.g. the proportions of infections in pigs resulting from indirect transmission. The role of peak pork consumption periods [92] could provide a more realistic way of implementing T. solium seasonal dynamics and would be interesting to explore with relevant longitudinal data. Advanced statistical modelling approaches have also been adopted in the wider Echinococcus modelling literature to improve predictive ability where periodicity in human echinococcosis prevalence data is observed [93]; however, fairly detailed time-series data are required for model fitting. Further seasonal heterogeneities may exist including seasonal slaughter patterns in areas where more pigs are slaughtered due to specific holidays [92], to obtain capital ahead of planting season, or the free capital for school fees. Likewise, seasonal variation in local crop production systems have a potential impact on transmission dynamics [3]. Less predictable events such as funerals can additionally lead to increased slaughter activity and movement of pigs.
A number of data gaps are evident to inform modelling efforts and develop a comprehensive research agenda for T. solium control and elimination efforts, with Fig 2 summarising data needs described across this paper. It is clear that one of the limitations of existing T. solium transmission models is uncertainty surrounding biological parameter estimates, for example, for those associated with egg dynamics and the adult tapeworm lifespan, identified as influential parameters (egg production rate/ death rates) in the EPICYST sensitivity analysis [56]. Direct measurement is often difficult through experimental design, for example for egg production rates; therefore, it could be possible to use the existing or improved T. solium models to infer these values from observable data, such as fitting to baseline prevalence data. Transmission rate (FoI) parameterisation with FoI model fitting for different settings as applied for E. granulosus [36, 48] and Trypanosoma cruzi [94] could be facilitated with collection of detailed age-stratified prevalence and incidence data, using diagnostics with field-validated sensitivity and specificity estimates to perform suitable adjustments. Necropsy of pigs, which is the assumed gold standard diagnostic methodology, would provide the most robust and reliable data for model fitting; however, issues associated with cost and feasibility of obtaining reasonable sample sizes, longitudinal measurements, and utility in the control phase of a programme with low infection prevalence levels pose barriers to the use of these data. Determining serological diagnostic markers which represent true infection status will be important, as performed for validation of B158/B60 Ag-ELISA with necropsied animals in Zambia [95], to establish effectiveness of interventions where necropsy is unavailable.
Development of spatial transmission models, when the current uncertainty is addressed in existing models, will require spatially-resolved infection datasets, including variables on pig movement between communities and/or households, household georeferenced data, and data on human movement, as demonstrated for developing a spatially-explicit network model of endemic schistosomiasis in Senegal using mobile phone data [96].
Although not necessary for accurate transmission modelling, dynamic modelling of neurocysticercosis (NCC) to understand how interventions influence longer term burden of disease estimates would be useful for economic assessments. The main challenges associated with NCC modelling include simulating the proportions of individuals with cysticercosis that have neurocysticercosis, and the proportion subsequently developing morbidity and when this occurs (rather than those that are asymptomatic or presenting with mild symptoms), which would require temporal data [97]. Burden of disease modelling would also require data to capture the variation of infection-related morbidity. Clinical neurocysticercosis, for example, is highly pleomorphic, with a range of factors influencing clinical outcomes including the location of lesions within the central nervous system (e.g. extra- compared to intra- parenchymal), the cyst stage and the intensity of the immune response to cysts [98]. Bhattarai et al. [99] have included the DALYs for NCC associated headache in their burden of disease estimation, but more generally modelling efforts have focussed on morbidity associated with epilepsy and seizures. Relevant to transmission, the EPICYST model [56] also contains a compartment for humans infected with both cysticercosis and taeniosis, for which there are very limited data.
Finally, it is clear that simulated interventions need improved parameterisation in terms of efficacy and coverage and require longitudinal intervention datasets for validation. Reliable intervention modelling will require data on age-structured interventions, especially for pig-directed strategies such as vaccination and oxfendazole treatment (to model that animals close to slaughtering age should not be treated), but also for human-directed strategies such as school-based treatment programmes [77]. This type of intervention modelling is already implementable in cystiSim [53] and there are plans to integrate these interventions using an age-structured version of EPICYST [56]. A ‘logical model’ of pig cysticercosis infection risk in different age cohorts by Lightowlers & Donadeu [92] clearly outlines some of the considerations for an age-structured model. For example, the authors suggest restricting oxfendazole use in animals approaching the average age of slaughter, as oxfendazole treatment mandates a 21-day withholding period before human consumption. Equally, testing how the average age at which pigs are slaughtered impacts onward transmission risk and, therefore, intervention efficacy would be important to consider.
Working closely with field partners, stakeholders and strengthening collaboration between T. solium modelling groups will facilitate opportunities to harmonise models and compare projections through cross-validation based on longitudinal field data from intervention trials [100]. This approach will improve confidence in the predictive abilities and utility of T. solium transmission models for evaluating whether the WHO NTD roadmap targets, especially relating to the development of a validated strategy for control and elimination, will be achievable in the near future.
|
10.1371/journal.pcbi.1004651 | Operational Principles for the Dynamics of the In Vitro ParA-ParB System | In many bacteria the ParA-ParB protein system is responsible for actively segregating DNA during replication. ParB proteins move by interacting with DNA bound ParA-ATP, stimulating their unbinding by catalyzing hydrolysis, that leads to rectified motion due to the creation of a wake of depleted ParA. Recent in vitro experiments have shown that a ParB covered magnetic bead can move with constant speed over a DNA covered substrate that is bound by ParA. It has been suggested that the formation of a gradient in ParA leads to diffusion-ratchet like motion of the ParB bead but how it forms and generates a force is still a matter of exploration. Here we develop a deterministic model for the in vitro ParA-ParB system and show that a ParA gradient can spontaneously form due to any amount of initial spatial noise in bound ParA. The speed of the bead is independent of this noise but depends on the ratio of the range of ParA-ParB force on the bead to that of removal of surface bound ParA by ParB. We find that at a particular ratio the speed attains a maximal value. We also consider ParA rebinding (including cooperativity) and ParA surface diffusion independently as mechanisms for ParA recovery on the surface. Depending on whether the DNA covered surface is undersaturated or saturated with ParA, we find that the bead can accelerate persistently or potentially stall. Our model highlights key requirements of the ParA-ParB driving force that are necessary for directed motion in the in vitro system that may provide insight into the in vivo dynamics of the ParA-ParB system.
| Segregating genetic material is essential for cell survival over multiple generations. The process underlying the required spatio-temporal organization of DNA is mediated by the ParA-ParB-parS system. Recently, experiments have shown that directed motion can be reconstituted in vitro. In these experiments, a magnetic bead was covered with the protein ParB and was able to move ballistically over a surface of DNA that was bound by the protein ParA. How does this active transport spontaneously emerge? In this paper we present a deterministic model for the dynamics of ParA-ParB proteins. We show how spatial noise in surface bound ParA is sufficient for the creation of a gradient in ParA that can drive motion of ParB in vitro. The model explains certain key aspects of the in vitro ParA-ParB system and leads to testable predictions.
| A variety of mechanisms exist within bacteria to spatially localize proteins within the small confines of a bacterial cell, from reaction diffusion processes that set up waves [1] to spatial occlusion due to the highly crowded environment [2–5]. One such protein system that has been observed to display highly dynamic spatial localization within bacteria is the ParA-ParB system. These two proteins are responsible for actively transporting DNA, whether it be the replicating chromosome [6, 7] or much smaller plasmids [8], within the cell. With respect to the chromosome, it has been observed that ParB bound to sites near the origin of replication processively moves from one pole to the other via a gradient in ParA that is bound on the nucleoid surface [9]. When the system is used to segregate plasmids, ParB bound plasmids are seen to move along the ParA bound nucleoid, eventually settling into positions that are equally spaced along the cell length [10].
The exact mechanism by which the ParA-ParB system generates directed transport is not entirely resolved. Biochemical experiments have shown that ParA dimerizes in the presence of ATP and is able to to bind to DNA as ParA-ATP [11]. ParB is able to bind a specific DNA sequence known as parS and aids the autohydrolysis of ParA-ATP, causing it to unbind from the DNA [12, 13]. Recent studies have also shown that the structure of the bacterial nucleoid plays a role as well. Specifically, it has been shown that conformational changes in the nucleoid can disrupt plasmid positioning [14] and a model suggests that chromosomal elasticity could provide the required translocation force that transports partition complexes [15].
Based on the experimental evidence, several models have emerged to explain the operation of the ParA-ParB-parS system and rule out mechanisms that generate predictions which are inconsistent with observations. One model assumes that ParA-ATP binds into longitudinal filaments over the nucleoid, and that ParB initiates depolymerisation and is then carried along by a depolymerisation force [16, 17]. The filamentous structure provides directionality to the segregation of DNA. Another model considers that ParA dimers bind uniformly over the nucleoid surface and that ParB linked DNA moves via a diffusion-ratchet mechanism as it creates a wake of ParA [18]. A chemotactic force is hypothesized to exist between the two proteins that biases the otherwise free diffusion of ParB bound DNA [19]. Recent work argues that the precise nature of how ParA binds to the nucleoid surface is inconsequential for the equi-positioning of plasmids in-vivo [14]. It showed that the plasmid’s ability to move along a gradient in ParA concentration was sufficient to explain their resulting positioning. Additionally a diffusion/immobilization mechanism where freely diffusing ParB complexes are immobilized through interactions with ParA was also ruled out. Another model has ruled out freely diffusing ParB complex biased by ParA concentrations by computing that such a mechanism was insufficient to provide directionality to the motion of the partition complex [15].
Recent experimental work has managed to amazingly reconstitute the ParA-ParB system in vitro which provides a new context in which to explore the mechanism of how this two protein system produces active transport. Specifically, it was found that a ParB coated magnetic bead can undergo directed motion when confined to move along the plane (without rolling) of a DNA coated flow cell that was bound with ParA-ATP protein [20, 21]. Multiple beads which were identically prepared were observed to show either directed motion with differing speeds or underwent free diffusion. The beads with directed motion had diffusion constants that were around 3 times lower than that of the freely diffusing beads and had paths with persistence lengths which were many times (∼20) the radius of the micron sized bead. ParA in the vicinity of the bead was hydrolyzed and released from the surface as the bead moved and was later recovered when the bead had moved away. After an initial lag time, the bead would begin to move across the surface, creating a wake in ParA. Thus from the experimental observations, persistent motion seems to result from the creation of a ParA gradient that can provide a chemotactic force which can drive the ParB covered bead [19]. However, how does this gradient form? And how does the motion of the bead depend on the various system parameters?
Here we develop a completely deterministic model for the operation of the in vitro ParA-ParB system that complements prior modelling work on the in vivo system. We consider the ParB decorated bead to be an over-damped particle under the influence of attractive forces from ParA proteins on the surface (see Fig 1). As shown experimentally, beads experiencing directed motion had little diffusion [21], and so we consider a bead’s motion to be completely deterministic and proportional to this chemical force. ParA kinetics on the surface is also completely deterministic and is driven by the presence of the bead which removes ParA within its vicinity. The only noise we consider is in the initial spatial distribution of ParA and we find that this is sufficient to generate a spontaneous ParA gradient which can drive the motion of the bead. Under most conditions, we find that the bead moves with constant speed, and that this depends on the ratio of the range of the force to that of ParA removal by ParB. Interestingly, we find that the bead can attain a maximum speed which depends on this ratio.
We also consider ParA rebinding to the surface since in the in vitro experiment, depleted ParA regions would recover once the bead had moved away [21]. ParA surface diffusion may also serve as a possible mechanism for recovery of ParA. There are two regimes for ParA recovery on the surface: undersaturated, where there are excess binding sites on the DNA for ParA or saturated, where there is more free ParA than there are binding sites on the surface. We find that in the undersaturated regime persistent acceleration of the bead is possible. In the saturated regime, depending on the rate of rebinding and the degree of ParA cooperativity, the bead can be made to stall. From our modeling we find that distinguishing cooperative binding or ParA surface diffusion from that of non-cooperative rebinding using the current in vitro assay would be challenging as their effects on bead motion are all qualitatively similar. Nevertheless, the model does make predictions that could be readily testable using the in vitro system and suggests ways to tune the operation of the ParA-ParB system.
In Fig 1 we show a schematic of our minimal model for the in vitro ParA-ParB system. In the experiment by Vecchiarelli et al. [21], a ParB decorated bead was put into contact with a surface that was covered with strands of DNA on which ParA-ATP was bound. The observed motion of the bead was predominantly unidirectional, and so we begin by considering a 1D model for the system. We represent the surface bound ParA-ATP with a concentration, a(x, τ). It is initialized with a mean concentration that fluctuates uniformly from position to position with a magnitude δa. The bead of radius R is located at a position xp, and we assume there is a central force that acts on it due to the interactions between ParA-ATP and ParB. ParB on the bead also stimulates the removal of ParA-ATP in the vicinity of the bead. We assume that bead motion is in the overdamped regime so that the drag force balances the net force due to ParA-ParB interaction. Both the chemotactic force and the rate of removal decay with distance from the center of the bead. In the absence of surface diffusion or ParA rebinding the system dynamics for our minimal model are given by the following deterministic equations(for further details, see Methods):
∂ a ( x , τ ) ∂ τ = - e - ( x - x p ) 2 / 2 c 2 a ( x , τ ) , (1) v = d x p d τ = A 0 ∫ d x e - ( x - x p ) 2 / 2 x - x p 1 + ( x - x p ) 2 a ( x , τ ) . (2)
Here the parameter A0 combines several system parameters: the initial mean ParA concentration, the amount of ParB on a bead, the ParA-ParB interaction force, and the drag on the bead (see Methods). The parameter c is the ratio of the lengthscale over which the bead removes nearby ParA-ATP to the lengthscale over which it experiences a force due to the surface bound ParA. We assume that both the rate of removal and force decay as Gaussian functions and c is the ratio of their standard deviations σr and σf respectively. Physically, c should be less than 1, since ParA can not be removed over distances greater than that from which it exerts an attractive force on the bead.
An estimate for these two lengthscales can be inferred from several experimental observations. First, it was found from in vivo measurements that active motion of plasmids require a ratio of ParB to ParA of around 5 to 1 (there are about 580 ParB bound proteins to 120 ParA in a cell [15]). We assume that the in vitro system requires a similar ratio to generate motion of the ParB decorated beads. The experiment conducted by Vecchiarelli et. al [21] estimates the number of ParB molecules on a bead that can interact with ParA on the surface to be 4800. They also found the number of ParA molecules present per square micron on the surface near the bead to be 400. Given the required ratio of ParA to ParB molecules, a bead with 4800 ParB molecules would need to interact with ∼ 1000 ParA molecules to generate motion. Given the ParA molecule surface density, 1000 ParA molecules would cover 1000/400 ∼ 2.5 μm2 of the surface leading to an effective force range, σf ∼ 0.9 μm. From the experiment, it was also observed that the radius over which the bead removed ParA from the surface (σr in our model) was on average 225 nm. Thus given these experimental observations, we estimate that c = σr/σf ∼ 0.2−0.4 given the lower and upper bounds on the measured values. As we will show below, this value for c is sufficient for generating directed motion of a bead.
The only source of noise in our system is in the initial conditions that describe the ParA concentration at every point. Starting with the bead at rest, integrating the above deterministic equations show that after a short time lag the bead begins to move (Fig 2A). This lag period has also been observed in the in vitro experiments [21]. The cause of this movement is the non-zero net force that builds along a particular direction due to the noise in the initial bound ParA distribution which breaks the symmetry around the bead. As the bead travels it removes ParA-ATP leaving a wake behind itself and creates a wavefront of bound ParA in front (Fig 2B). Since the bead is over-damped it has no inertia from its previous step and movement along the chosen direction is sustained because the backward pulling force due to the ParA behind the bead is less than that due to the ParA in front.
In two dimensions, a similar lag period is observed during which the bead reduces the ParA concentration (Fig 3B). After symmetry breaking the bead maintains motion along a particular direction as the attractive forces along the sides of the bead roughly cancel out whereas the forward motion is sustained due to the depleted concentration behind the moving bead.
For the model given by Eqs 1 and 2, following the symmetry breaking the bead attains a uniform speed as can be seen by the linear displacement of the bead with time (Fig 2A). The average speed is independent of the magnitude of the noise in the initial ParA distribution (Fig 2C), though we find that the standard deviation of the speed does increase with the noise level. We also find by increasing A0 (for example by increasing the initial mean ParA concentration), the bead’s speed increased, keeping c fixed. The constant of proportionality between speed and A0 is found to be dependent on c.
Given that the bead attains a uniform speed we analytically solved Eqs 1 and 2 in this limit (see S1 Text). The steady state ParA distribution for a bead moving with constant speed was found that was then used to solve for the steady state net force acting on the bead (see S1 Fig). This force balances the drag force on the bead moving at constant speed, v, that leads to a non-linear equation that can be solved for the speed in terms of the two free parameters, A0 and c.
In Fig 3A, we show that the analytical results for the predicted speed, v, as a function of c, match well with the simulated results found from integrating Eqs 1 and 2. We observe that the speed of the bead is maximized at a particular value of c for a fixed value of A0 and this value of c at which the speed is maximized changes with changing A0. The presence of a maximum speed was not unexpected, since in the limit c → 0, no ParA is removed and hence there is no motion and when c → 1, too much ParA is removed and the gradient is weakened, lessening the speed.
The two dimensional simulation shows a typical trajectory on a 2d substrate that has a noisy initial distribution of ParA (see Fig 2B). Now the bead can spontaneously move in any direction and moves roughly in a directed fashion over the surface. Similar dependences on A0 and c were also found. The analytical solution was extended to 2d by considering the entire speed of the bead to be along an axis (see S1 Text for details). It predicted similar features of maximum speed at a particular value of c, which was confirmed by simulation. Interestingly, for values of c less than ∼ 0.3 the bead is faster when placed in a 1d system. But for values of c greater than ∼ 0.3 the bead would attain higher speed on a 2d surface (Fig 3C). This can be intuitively understood as follows: when c < 0.3 on a 2d surface the wake of reduced ParA behind the bead is thinner than the bead’s effective radius that is attracted by the remaining ParA. This implies that there is a backward pulling force component due to unremoved ParA behind the bead. This force component decreases as c > 0.3 and the wake width nears the effective bead radius. This leads to lower speeds for c < 0.3 and higher speeds for c > 0.3 in 2d.
Next we investigated the inclusion of ParA rebinding to a one-dimensional substrate. We assumed that at each position x along the surface there was a certain concentration of binding sites for ParA on the DNA substrate, d(x), whose average is D0, and fluctuates an amount δd (see Fig 4). We now also consider that there is ParA in the buffer, given by the quantity ab(τ). Diffusion is quick in the buffer and so ParA in the buffer does not depend on position. The amount of ParA in the system is limited by the initial amount in the buffer which is set to ab(0) = As. For non-cooperative binding of ParA to the substrate, rebinding depends on the amount of unbound sites available at a given location, namely (d(x) − a(x, τ)) and the rebinding rate, kr. We included cooperative rebinding with a term that depends on the amount of ParA on the surface, and is governed by a cooperative rate, kc. Including this favors rebinding to sites that possess larger amounts of bound ParA. Including rebinding changes Eq 1 to the following equation,
∂ a ( x , τ ) ∂ τ = - e - ( x - x p ) 2 / 2 c 2 a ( x , τ ) + a b ( τ ) [ d ( x ) - a ( x , τ ) ] [ k r + k c a ( x , τ ) ] . (3)
While the equation for the ParA per binding site available in the buffer, ab(τ) is given by:
d a b ( τ ) d τ = ∫ d x e - ( x - x p ) 2 / 2 c 2 a ( x , τ ) - a b ( τ ) [ d ( x ) - a ( x , τ ) ] [ k r + k c a ( x , τ ) ] / L (4)
Since the ParA released due to the bead diffuses rapidly in the buffer it adds to the free concentration ab uniformly. As the ParA released by the bead is distributed over the entire system, the system size (L) and rate of rebinding (kr) are coupled in our model. Increasing the length of the system would effectively decrease the rate of rebinding as the ParA released would be diluted over a larger area. The finite size effects of our model also become evident when the bead nears a boundary of the surface. In the 1d model, on reaching an edge the bead stopped there until the ParA concentration was recovered enough in its wake and then it would begin to move back to the other end. In 2d, our model found that a bead would change directions when it encountered a boundary but would never come to a halt, unlike in 1D (see S2 Fig).
Depending on the initial amount of ParA in the buffer, ab(τ = 0) = As, there are two possible regimes defined by the quantity ϕ = As/D0: ϕ > 1 is the saturated regime where there is an excess of ParA and ϕ < 1 in which the system is undersaturated, and there is always an excess of binding sites for ParA.
When the system was initialized with limited ParA such that ϕ < 1, there was still the possibility of further rebinding at every point. On introducing the bead into the system, a spontaneous gradient forms and the bead starts traveling in a particular direction as in the previous sections. As the sites in front of the bead have the capability to bind more ParA, the wavefront attracting the bead increases in magnitude and the bead gains further speed (Fig 4B). Thus in the undersaturated regime, we predict that it may be possible to observe persistent acceleration of the bead (Fig 5A).
In the regime ϕ > 1, there is excess ParA in the buffer and during the equilibration phase (before the ParB decorated bead is introduced), the bound ParA is nearly equal to the saturating limit d(x) while free ParA in the buffer still remains. Bead motion in this regime is similar to that described in the first section. A wavefront forms due to hydrolyzation and as all sites ahead of the bead are saturated the leading edge can not grow (Fig 4C). Indeed, we find that the bead attains a uniform speed, experiencing only an initial short burst of acceleration.
Within these two regimes a variety of bead behaviours can be observed on varying ϕ. In the undersaturated regime, when ϕ ≪ 1 the bead accelerates persistently, never reaching a saturated speed within the surface length L. We carried out a simple analytical calculation (see S3 Text) in this limit to determine the dependence of the bead’s speed on time which matched our simulation results (see S3 Fig). For values of ϕ ≲ 1 it is possible for the bead’s speed to increase and saturate to a constant value. This happens when the bound ParA wavefront in front of the bead rises to saturate all the binding sites. In the oversaturated regime, when ϕ ⩾ 1 there is a possibility of the buffer ParA filling the wake region completely. If this occurs rapidly enough, we expect that there should be potential to stall the bead. For a fixed value of kr, we observed that increasing the ϕ led to reduced bead speeds until a value ϕstop was reached, at which the bead did not commence motion (Fig 5C).
Besides depending on ϕ, the dynamics of the bead also depend on the rate of rebinding, kr. We find that in the undersaturated regime when ϕ ≲ 1 the final value to which the bead speed tends, decreases with increasing kr to a saturated value (Fig 5B inset). Hence, it is not possible to stall the bead no matter how high the rate of the rebinding is. In the saturated regime however the bead can be made to stop by increasing ϕ and this ϕstop depends on kr. Some analytics that show this dependence are given in the S3 Text section of this paper and agree well with our simulated results (S3 Fig).
We then considered the case of cooperative rebinding, where we assumed that rebinding depended not only on the amount of free binding sites (d(x) − a(x, τ)) but also on the amount of ParA bound at a given location. This cooperative rebinding was governed by a rate constant, kc while a non-zero kr gave the rate of non-cooperative rebinding from the buffer (Eq 3).
In the regime where ϕ < 1 the bead behaviour was similar for simulations with and without any cooperativity in rebinding. As the amount of ParA in the buffer was scarce, it immediately redistributed to all unsaturated sites. In the ϕ < 1 regime, introducing cooperative rebinding increases the acceleration of the bead as lesser ParA rebinds to the wake and more ParA rebinds to the wavefront ahead. In the ϕ > 1 regime the role of the rates of rebinding become more prominent as there is ParA available for rebinding, but only in the regions from where the ParA is hydrolyzed. Increasing the cooperative rebinding rate led to the depletion zone filling in faster, making it possible for the wake to recover and stall the bead at lower values of ϕ (Fig 5D). Hence, introducing cooperativity in ParA rebinding led to the bead stopping at lower values of ϕ than when there was no cooperative binding.
Lastly, we considered the effect of surface diffusion of ParA in the absence of rebinding to determine if there were any significant differences in bead behaviour from that of rebinding from a well mixed buffer alone. We again included a saturating limit for the amount of ParA concentration that could exist at every point x. This coupled the amount of available ParA in the system to the binding site distribution, d(x) as ParA could diffuse to a point only if it had the capacity to bind more ParA. We assumed that ParA surface diffusion was governed with a diffusion coefficient, κ. The equation describing the bead dynamics remains the same while the equation describing a(x, τ) in the presence of surface diffusion and no rebinding is given by (see S2 Text for details):
∂ a ( x , τ ) ∂ τ = - e - ( x - x p ) 2 / 2 c 2 a ( x , τ ) + κ [ d ( x ) ∂ 2 a ( x , τ ) ∂ x 2 - a ( x , τ ) d 2 d ( x ) d x 2 ] . (5)
As the concentration of binding sites is spatially noisy this coupling maintains the spatial noise necessary for the formation of the spontaneous gradient in ParA that initiates bead motion.
The system was initiated in a state such that all sites on the surface have bound ParA equal to the binding distribution. When ParA protein’s surface diffusion is low, the wake can not fill in rapidly enough to stall the bead and so it attains a steady state speed. By increasing the diffusion constant, just as was the case for rebinding rates, the bead could be made to stall (Fig 6). Thus qualitatively, the behaviour is nearly identical to the situation where only rebinding occurred from the buffer.
In this paper we have presented a minimal model for the operation of the in vitro ParA-ParB system. The model involved a tug-of-war between attractive forces exerted on a ParB decorated bead by surface bound ParA in front of and behind the bead. ParB on the bead would remove ParA from the surface, tilting the balance in favor of one side, leading to directed motion. For a range of parameter values, we found that spatial noise in the initial ParA distribution was sufficient to break spatial symmetry, causing the spontaneous formation of a ParA gradient between the front and back of the bead, leading to motion.
It was found experimentally that identical beads could display different speeds when placed on the same DNA substrate. Our model provides some insight into parameters that influence a bead’s speed. The first such parameter is A0 which depends on both the initial ParA concentration as well as the amount of ParB on the bead. A simple explanation for the observed differences in speed, would be that the amount of ParB may not be the same on each bead, and hence, even though the surface bound ParA concentration that each bead sees is the same, the effective A0 would be different. Another contributing factor that could influence the amount of ParB on a bead that can interact with the surface, is bead bound ParA. Experimentally it was found that the ParA content on the beads undergoing directed motion was 25 ± 5% less than on beads which diffused freely [21]. This suggests that increased ParA on the bead lowers its ability to interact with surface bound ParA and transforms the bead into a freely diffusing particle. In our model, ParA bound to beads would affect A0 through the effective change in the amount of ParB that can interact with the surface.
The speed of a bead also depends on the parameter c which is the ratio of the lengthscale of the ParA removal kinetics to the lengthscale of the ParA-ParB attraction force. We speculate that a change in this ratio could be experimentally achieved by changing the size of the DNA linker that binds ParB to the micron-sized bead. The changes in bead speed due to different c could be characterized by looking at the shape and size of their wakes.
The model displayed a rich variety of behaviour when rebinding of ParA to the surface was considered. We showed that when the surface was saturated and there is always free ParA in the buffer that it should be possible to stall the bead. For the situation where the surface is unsaturated and free ParA can always find free sites to bind, we predict that persistent acceleration of the bead results, with the counter intuitive result that lesser total amount of ParA can actually lead to higher speeds. In order to potentially see acceleration, one would likely have to study beads on much narrower tracks so that the released ParA could have an appreciable effect on the wavefront when it rebinds. These predictions that only depend on the amount of ParA in relation to binding sites should be readily testable experimentally.
A topic of some debate about the operation of the ParA-ParB system is the role of cooperative binding for ParA and whether the formation of filaments or ParA clusters is essential. We included cooperative rebinding of ParA in our model and found that it was qualitatively indistinguishable from non-cooperative rebinding in regards to the bead dynamics. More complex dynamics, that include having multiple beads (the in vitro version of multiple plasmids in a cell) may be able to disentangle whether cooperative rebinding has any detectable effect. This will be a topic for further exploration. We also allowed for ParA surface diffusion and found that it too led to dynamical behaviours that would be hard to distinguish it from non-cooperative rebinding.
Although our model was developed to capture essential features of the in vitro ParA-ParB system, we feel that it may serve as a useful coarse grained model for studying the in vivo system. Future work towards this end would allow for ParB to diffuse and include discretizing the system to consider stochastic kinetic effects. Neither of these is currently in the continuum model presented here, but likely play a role in vivo. In summary, the model presented here makes several non-trivial predictions that should further aid the dissection of the operational mechanisms of this active transport system.
To derive the dimensionless Eqs 1 and 2 we start with a 1d version of our model in real space-time coordinates X and t. The concentration of ParA at every point on the surface is given by Am(X, t) which can only be removed by the ParB decorated bead. The rate of removal depends on the bead’s position, Xp and decays with distance from the bead. We assume that the rate has a Gaussian form, centered on the bead with a characteristic range of removal given by the parameter, σr. The bead has attractive forces acting on it due to its interactions with the ParA on the surface. The force between the bead and ParA on the surface decays with distance from the bead, and is directed along the surface in proportion to the X component of the vector connecting the bead to the surface location. Similar to the rate of removal, we consider the magnitude of the force to have a Gaussian form with a characteristic range given by σf. The total force is found by integrating over the entire surface. We consider that the dynamics of the bead is in the over damped regime and so the net force, F is proportional to the bead’s speed, dXp/dt. Putting all these assumptions together we start with the following dimensionful equations for the ParA-ParB system:
∂ A m ( X , t ) ∂ t = - γ 0 e - ( X - X p ) 2 / 2 σ r 2 A m ( X , t ) (6) β d X p d t = F = ∫ d X F 0 e - ( X - X p ) 2 / 2 σ f 2 X - X p R 2 + ( X - X p ) 2 A m ( X , t ) . (7)
Here γ0 gives the rate of ParA removal, F0 is a multiplicative constant that scales the force per unit ParA and ParB concentration exerted on the bead and β is the drag coefficient of the bead given by β = 6πηR. The above equations are reducible to a dimensionless version by making the following transformations:
X → R x , X p → R x p , t → τ / γ 0 (8) A m ( X , t ) → a 0 a ( x , τ ) (9)
Where x, xp and τ are dimensionless variables and a0 is a multiplicative constant to scale the ParA in the system. Under these transformations and introducing the ratio c = σr/σf the dimensionless equations governing the dynamics become:
∂ a ( x , τ ) ∂ τ = - e - ( x - x p ) 2 / 2 c 2 a ( x , τ ) (10) v = d x p d τ = A 0 ∫ d x e - ( x - x p ) 2 / 2 x - x p 1 + ( x - x p ) 2 a ( x , τ ) (11)
Here A0 = F0 a0/βR and c are the primary parameters of the system on which v depends. It should be noted that varying the constant A0 might imply varying the magnitude of initial ParA concentration a0 or magnitude of the force of attraction exerted by the ParA per unit ParB present on the bead, F0 or inversely varying the radius of the bead R. All these dependencies have been suitably combined into the single dimensionless parameter A0, which when varied reflects changes in concentration of initial ParA, since both the strength of the attractive force and radius of the bead are assumed fixed and not readily changeable. Apart from A0 the only other parameter affecting bead speed is the ratio, c.
We numerically integrate the above equations for a 1d system with surface length L = 70 and spacing dx = 0.02 using the Euler step method in steps of dτ = 0.01. For each simulation we fixed an average initial ParA concentration and obtained the ParA concentration for every site by adding uniform noise with a magnitude δa. The resultant profile was a spatially noisy distribution about the mean ParA concentration. The bead was placed in the center of this surface to replicate the in vitro experimental process such that the bead is surrounded by ParA in all directions. The forces attracting the bead from the ParA are integrated using Simpson’s rule to obtain the total vector force on the bead and the change in its position can be calculated using Eq 2. Bead speeds were obtained by doing a linear fit to the position vs. time graphs, ignoring the initial lag period.
Similar transformations can be extended to obtain the following set of equations for a 2d system:
∂ a ( x , y , τ ) ∂ τ = - e - ( r - r p ) 2 / 2 c 2 a ( x , y , τ ) (12) ( r 2 = x 2 + y 2 , r p 2 = x p 2 + y p 2 ) d x p d τ = A 0 ∫ ∫ d x d y e - ( r - r p ) 2 / 2 x - x p 1 + ( r - r p ) 2 a ( x , y , τ ) (13) d y p d τ = A 0 ∫ ∫ d x d y e - ( r - r p ) 2 / 2 y - y p 1 + ( r - r p ) 2 a ( x , y , τ ) (14)
|
10.1371/journal.pcbi.1000192 | Nonidentifiability of the Source of Intrinsic Noise in Gene
Expression from Single-Burst Data | Over the last few years, experimental data on the fluctuations in gene activity
between individual cells and within the same cell over time have confirmed that
gene expression is a “noisy” process. This variation is in
part due to the small number of molecules taking part in some of the key
reactions that are involved in gene expression. One of the consequences of this
is that protein production often occurs in bursts, each due to a single promoter
or transcription factor binding event. Recently, the distribution of the number
of proteins produced in such bursts has been experimentally measured, offering a
unique opportunity to study the relative importance of different sources of
noise in gene expression. Here, we provide a derivation of the theoretical
probability distribution of these bursts for a wide variety of different models
of gene expression. We show that there is a good fit between our theoretical
distribution and that obtained from two different published experimental
datasets. We then prove that, irrespective of the details of the model, the
burst size distribution is always geometric and hence determined by a single
parameter. Many different combinations of the biochemical rates for the
constituent reactions of both transcription and translation will therefore lead
to the same experimentally observed burst size distribution. It is thus
impossible to identify different sources of fluctuations purely from protein
burst size data or to use such data to estimate all of the model parameters. We
explore methods of inferring these values when additional types of experimental
data are available.
| Recent experimental data showing fluctuations in gene activity between individual
cells and within the same cell over time confirm that gene expression is a
“noisy” process. This variation is partly due to the small
number of molecules involved in gene expression. One consequence is that protein
production often occurs in bursts, each due to the binding of a single
transcription factor. Recently, the distribution of the number of proteins
produced in such bursts has been experimentally measured, offering a unique
opportunity to study the relative importance of different sources of noise in
gene expression. We derive the theoretical probability distribution of these
bursts for a wide variety of gene expression models. We show a good fit between
our theoretical distribution and experimental data and prove that, irrespective
of the model details, the burst size distribution always has the same shape,
determined by a single parameter. As different combinations of the reaction
rates lead to the same observed distribution, it is impossible to estimate all
kinetic parameters from protein burst size data. When additional data, such as
protein equilibrium distributions, are available, these can be used to infer
additional parameters. We present one approach to this, demonstrating its
application to published data.
| The regulation of gene activity is essential for the proper functioning of cells,
which employ a variety of molecular mechanisms to control gene expression. Despite
this, there is considerable variation in the precise number and timing of protein
molecules that are produced for a given gene under any particular set of
circumstances. This is because gene expression is fundamentally a
“noisy” process, subject to a number of sources of randomness.
Some of these are intrinsic to the biochemical reactions that
comprise the transcription and translation of a particular gene [1],[2]. Several
of the reactions involve very small numbers of molecules. There are only one or two
copies of the DNA for the gene, and in its vicinity inside the cell there are likely
to be only a few copies of the relevant transcription factors and of RNA polymerase.
Similarly, for each mRNA molecule, the processes of ribosome binding and of mRNA
degradation are typically highly stochastic.
Recent advances in experimental technology have shown that such single molecule
effects can lead to protein production occurring in bursts of varying size, each due
to a single transcription factor binding event [3],[4]. Other sources of
variability are extrinsic to the specific reactions, and include
fluctuations in relevant metabolites, polymerases, ribosomes, etc. [1],[2]. These
will not be considered further here.
It is of considerable interest to determine the various contributions of such
different sources of variability. Within the last few years, experimental techniques
for addressing this question have increasingly become available. Elowitz et al.
[1]
observed fluctuations in the expression level of genes tagged both with cyan and
yellow fluorescent proteins in monoclonal Escherichia coli cells
under identical environmental conditions. Similar work was carried out by Raser and
O'Shea [5] in the eukaryote Saccharomyces
cerevisiae. Such dual-reporter experiments are able to distinguish between
intrinsic and extrinsic sources of stochasticity. More recently, single molecule
data has become available [6],[7], which monitors the expression of a gene a single
protein at a time and provides the distribution of the sizes of bursts. It had been
hoped that data of this kind would answer many of the remaining questions about the
origin of noise in gene expression and in particular distinguish between the
different contributions of transcription and translation to intrinsic noise.
Intuitively, one might expect that randomness due to transcription would play the
more significant role than translation, since typically there will be more than one
mRNA molecule, and the fluctuations due to translation from each of these might to
some extent average out. To test this hypothesis and to put it on a quantitative
basis, it is necessary to employ mathematical models of gene expression. These also
provide a valuable tool for the analysis of experimental data, and in particular of
the burst size distributions reported in the literature, e.g., [6],[7].
A great deal of work has gone into modelling gene expression in both prokaryotic and
eukaryotic systems, with some of the earliest papers predicting fluctuations in mRNA
and protein levels published 30 years ago [8],[9]. McAdams and Arkin [3]
provided the first model of bursting at the translation level. They showed that the
number of protein molecules produced by a single mRNA transcript is described well
by a model which considers whether the next event is the production of a further
protein, or the degradation of the mRNA molecule. Such competitive binding between
ribosomes and RNase results in a geometric distribution for the protein number. Such
an analysis can also be applied to transcription following the binding of a
transcription factor to a gene and also results in a geometric distribution. The
joint analysis of these two stochastic processes forms the basis of the present
paper.
The integration of simple stochastic (Markov) models of transcription factor, RNA
polymerase, ribosome and RNase binding leads to what is now widely regarded as the
standard model of gene expression for prokaryotes [4]. The analysis of this
model using a master equation allows the determination of the moments of the
distribution of the number of protein molecules when the system is in steady state.
Further analysis of this equilibrium distribution was carried out by Paulsson [10]–[12] who used the master
equation and the fluctuation–dissipation theorem to obtain predictions
about the mean and variances of molecule numbers and lifetimes and the contribution
made by transcriptional and translational bursting. Other studies have been carried
out by Höfer [13] who used a rapid-equilibrium approximation to
compare mRNA levels for genes with one and two active alleles, and by Friedman et
al. [14]. The drawback of these approaches is that the master
equation that describes the temporal evolution of the probability distribution of
protein (and mRNA) numbers is too complex to be solved analytically. Furthermore,
the burst size distribution necessary for comparison with recent experimental data
[6],[7] cannot be
obtained directly from the master equation. Such difficulties with master equation
based approaches are exacerbated in the case of more complex models of gene
expression such as multi-step models that include intermediate stages such as the
formation of DNA–RNA polymerase complexes, phosphorylation events, and
mRNA–ribosome binding. Both deterministic and stochastic simulation
studies of these models have been performed, e.g., [15] and [16], but none of these
approaches have been useful for the analysis of burst size data.
In the present work we avoid the problems associated with the master equation
approach, which are at least in part due to the explicit incorporation of time
evolution. Instead, we ignore time and directly derive an expression for the burst
size distribution by extending the analysis of [3]. In many ways this
approach is similar to that used for the analysis of multi-stage queues [17]. The
distribution of the number of mRNA molecules produced in a single burst is geometric
and the distribution of the number of protein molecules produced by a single mRNA is
also geometric [3]. The overall burst size distribution is therefore
given by the compound distribution of two geometric distributions [17]. This
can be readily computed using generating functions [17] and is itself
not geometric. However, experimentally it is not possible to
detect bursts that produce no protein molecules at all, and therefore the published
data [6],[7] are in fact
the relevant conditional distributions, assuming at least one protein molecule is
produced in a burst. Surprisingly, it turns out that when we condition the compound
distribution in this way, we again obtain a geometric distribution. This is
determined by a single parameter, which we can derive in terms of physically
meaningful constants such as binding and unbinding rates. This shows that different
combinations of noise levels in the translation and transcription parts of the
process can give the same overall burst size distribution. Mathematically, this
means that the standard model of gene expression (described in detail below) is
nonidentifiable
[18],[19] from
burst size data alone. This in turn implies that it is not possible to identify the
relative contributions of translation and transcription to the burst size
distribution of protein numbers only using this data.
We also show that our approach is applicable to a variety of more detailed models
that incorporate additional steps to provide more realistic descriptions of
expression [16]. These still yield a single parameter geometric
conditional distribution. This shows that within the context of a very large class
of models, experimental burst size data on its own cannot identify the relative
contributions of different reactions to the overall noise level. However, by
simulating the equilibrium distribution of protein numbers for different parameter
combinations giving the same burst size distribution we demonstrate that a
combination of burst size distribution and equilibrium distribution can discern
different sources of noise. The difficulty with such an approach is that the
determination of the equilibrium distribution requires the knowledge of two
additional kinetic parameters: the transcription factor binding rate and the protein
degradation rate. Estimates of these are not easy to obtain independently, so that
we now have to estimate six unknown parameters from the combined burst size and
equilibrium distribution data. Initial simulations (not shown here) suggested that
it is difficult to do this reliably.
It is possible however, by using independent estimates of one of the parameters to
reduce the parameter space from six to five dimensions. Using the relationship
between the remaining parameters determined from the burst size distribution allows
the elimination of a further parameter, leaving four kinetic parameters to be
estimated from the equilibrium distribution. We show below that by using the
Nelder-Mead algorithm to maximize the empirical likelihood, useful estimates of the
four remaining parameters can be obtained. We carry out this process twice, first
using independent measurements of the mRNA degradation rate and then of the protein
half-life. In the first case we obtain unrealistically short estimates of the
protein half-life, and in the second a considerably faster mRNA degradation. This
suggests that when in the repressed state, mRNA may be degraded at a faster rate
than when the gene is active.
In principle, this method can be applied to any gene where burst size and equilibrium
distributions are available, providing a new approach to the estimation of
parameters estimates for the ever more sophisticated models increasingly being used
in computational biology.
In the so called “standard model” of gene expression, Figure 1, an inactive gene can
be activated by a promoter or transcription factor. This allows molecules of RNA
polymerase to bind and produce mRNA. This in turn can bind to ribosomes leading
to the production of protein molecules. Eventually the transcription factor
unbinds, terminating the production of mRNA, and each mRNA molecule is degraded,
which stops protein production.
Each of these processes is modelled as a transition in a continuous time Markov
chain with a particular rate. Such a rate is interpreted as the probability of
an event occurring in a unit time interval. Thus, if we denote the rate of
transcription factor binding by α0 then the
probability of this occurring in an interval of length
δt, assuming that the transcription factor is not bound
at the start of the interval, is
α0δt.
Integrating over time, this means that the probability of the event having
happened by time t, is , whilst the average time for the event to happen is
1/α0. The same holds for the other
transitions in the model, with the rate of transcription factor unbinding
denoted by β1. Whilst the transcription
factor is bound, RNA polymerase binds at a rate
α1, and each such binding event is assumed
to produce one molecule of mRNA. More detailed models that allow the polymerase
to unbind before it has produced mRNA are considered later and will have no
effect on our overall conclusions.
Each mRNA molecule binds to a ribosome at rate
α2 and is degraded at rate
β2. When the last mRNA has decayed no
more protein will be produced. We define the number of proteins produced between
the transcription factor binding and the last mRNA decaying as a
“burst”. Note that since a burst begins once the
transcription factor has bound, we expect the distribution of burst sizes to be
independent of the transcription factor binding rate
α0. This is confirmed by the rigorous
derivation below.
Mathematically, the Standard Model of Gene Expression is a continuous time Markov
chain model. Each particular combination of number of mRNA molecules, number of
protein molecules and state of binding of the transcription factor constitutes a
single state of the model. It is possible to derive an (infinite) set of coupled
ordinary differential equations (called the Kolmogorov forward equations or
master equation) that govern the probability at any given time of the system
being in any given state. However, the analysis of a such a complex set of
equations is difficult. On the other hand, using the same approach as for
multi-stage queues, it is relatively easy to derive the distribution of protein
burst sizes.
We begin with the analysis of McAdams and Arkin [3] for the
distribution of the number of proteins produced by a single mRNA molecule. If a
certain number (possibly 0) of protein molecules has been produced, the
probability that the next event in which the mRNA molecule participates is the
production of another protein molecule is
p = α2/α2+β2)
(see Text
S1 for derivation). Conversely, the probability that the next event is
the degradation of the mRNA molecule is
1−p = β2/(α2+β2).
In order to produce precisely n molecules of protein, we need
n events of the first type to occur, followed by a final
degradation event. The probability of this happening is
pn(1−p), giving the
distribution Q(n) of the number of protein
molecules produced by a single mRNA molecule(1)
Here
A2 = α2/β2
is the expectation of Q. Contrasting this with [3],
the parameter A2 defining the distribution is now
expressed in terms of physically measurable rate constants. Exactly the same
argument applies to the distribution of the number of RNA molecules produced
between the successive binding and unbinding of the transcription factor. In
particular, the probability of producing one more mRNA molecule before the
transcription factor unbinds is
α1/(α1+β1)
and the probability of the transcription factor unbinding is
β1/(α1+β1).
In order to produce precisely m mRNA molecules before the
transcription factor unbinds we need m independent production
events with probability
α1/(α1+β1),
followed by the unbinding event with probability
β1/(α1+β1).
Thus the probability distribution, R(m), of the
number of mRNA molecules produced in one burst is(2)where
A1 = α1/β1
is the expectation of R(m). In order to derive
the overall protein burst size distribution for the Standard Model in Figure 1 we need the
probability generating functions [17] of the
distributions Q(n) and
R(m) which we denote as
Q*(z) and
R*(z), respectively. These are
simply obtained by summing the relevant geometric seriesand
The distribution P(n) of the total number of
proteins produced in a single burst is simply the compound distribution of
R and Q
[17].
This is easily computed using probability generating functions (see below), and
is not a geometric distribution. However, it is of relatively little interest
since it includes the possibility that the transcription factor unbinds before
any proteins have been produced (either because no mRNA is produced, or because
this mRNA is degraded before binding to a ribosome). Such events cannot be
observed in the experimental protocol used in [6],[7], and hence
P(n) cannot be directly compared to the
data in these papers. However, we can re-scale
P(n) to give the probability distribution
Pˆ(n) = P(n)/(1−P(0))
of protein numbers conditional on at least one protein being produced. An
approximate calculation of this distribution was given in the supplementary
material of [7]. This replaced the discrete geometric distribution
Q(n) by a continuous exponential
distribution of the same mean and then used the Laplace transform to obtain the
(continuous approximation to the) compound distribution. Here we present an
exact derivation for the discrete distribution using generating functions (which
are closely related to the Laplace transform). Furthermore we relate the
parameter of the final burst size distribution to the original kinetic
parameters α1,
α2,
β1, and
β2.
Thus, let X(i) be the random
variable, with distribution Q(n), giving the
number of proteins produced by the ith mRNA transcript and let
Y be a random variable, with distribution
R(n) giving the number of mRNA molecules
produced. Then the random variablegives the total number of proteins in a burst. Denote the
distribution of X by P(n),
with generating function P*(z). Then a
standard result on generating functions of compound distributions [17] gives(3)To obtain the distribution conditional on at least one protein
molecule being produced, we subtract P*(0) and
normalise (divide) by 1−P*(0) to giveThis is the generating function of a conditional geometric
distribution with (dimensionless) parameter
Â2 = A2(1+A1),
so that Pˆ(n) has the distribution(4)where the parameter Â2 can
be expressed in terms of the mean number A1 of mRNA
molecules produced and the mean number A2 of protein
molecules produced from a single mRNA molecule as(5)(6)We thus see that the burst size distribution is determined by a
single parameter, and that many different combinations of the parameters
α1,
α2, β1,
and β2 will lead to the same burst size
distribution. In mathematical language this says that the Standard Model with
parameters α1,
α2,
β1, and
β2 is nonidentifiable from
burst size data. In fact we can only estimate a single parameter (or a single
linear combination) and the three remaining parameters can be arbitrarily
chosen.
It might be hoped that such nonidentifiability is a particular pathology of the
Standard Model. We thus next consider a number of generalisations of this model,
which provide a more detailed description of the process of gene expression. We
find that for a wide range of generalisations we can still derive the burst size
distribution in a similar manner the above. It turns out to be geometric in each
case and hence all such models are also nonidentifiable.
One common extension is to include an additional step in the model of the
transcription process [13], as shown in Figure 2. This accounts for the fact that
after the transcription factor has bound, one still requires the RNA polymerase
to bind to the transcription initiation complex, and this may not always happen
successfully. A similar modification could be made to the translation loop to
describe the binding of the mRNA transcript to the ribosome in more detail. Both
of these additions can be considered individually, or in combination.
Doing this results in distributions R and Q
which are still geometric, but with the parameters
A1 and A2 given by more
complex combinations of the individual rates. We illustrate this for the
transcription loop, where we find that in order to produce exactly
m mRNA molecules, the system can pass through state
G* any number
i≥m times. On
i−m of these occasions the
polymerase unbinds before an mRNA molecule is produced, returning to
G with rate δ1, and on the
remaining m occasions an mRNA molecule is produced, with rate
α1. The m productive
steps can be interspersed in any order amongst the i visits,
giving possible choices. The probability of producing
m mRNA molecules is thuswith A1 now given by
A1 = α1γ1/β1(α1+δ1).
A similar derivation holds for the translation loop. We see that carrying out
either or both of these modifications still results in a geometric distribution
in the form of Equation 4 for Pˆ(n),
with
Â2 = A2(1+A1),
but A1 and A2 now given
by
A1 = α1γ1/β1(α1+δ1)
and
A2 = α2γ2/β2(α2+δ2).
As a consequence the overall conditional protein size distribution,
Pˆ(n), will still be given by
Equation 4, with the parameter
Â2 = A2A1+A2
as before.
An alternative generalisation is to add additional loops with the same structure
as the current transcription and translation loops. We prove in the Supporting
Information (Text S1) that if we have k−1 such loops,
the final conditional protein size distribution Pˆ
k
(n) will still be geometric.
We thus conclude that all of these models yield the same geometric protein burst
size conditional distribution, determined by a single parameter. In particular,
models which include additional steps to account for DNA–RNAP complex
formation and mRNA-ribosome complex formation give distributions that are
mathematically indistinguishable from those from the Standard Model. It is thus
impossible to differentiate between these models using experimentally observed
burst size distributions. Similarly we cannot use such data to differentiate
between the contributions to noisy gene expression from transcriptional versus
translational bursting.
We can compare the probability distribution derived above directly with
experimental data. We consider recently published data of burst sizes for two
fluorescently tagged proteins in the bacterium Escherichia coli
[6],[7]. In
[6], a
novel fluorescent imaging technique is used to determine the distribution of
protein molecules per transcription factor binding event in live E.
coli cells. The specific protein studied was a fusion of a yellow
fluorescent protein variant (Venus) with the membrane protein Tsr. The
tsr-venus gene is incorporated into the E.
coli chromosome, replacing the lacZ gene. This
modified gene is then under the control of the lac promoter. In
a second publication [7], the same group used a different imaging
technique to determine the distribution of protein molecules per transcription
factor binding event of β-gal in live E.
coli cells.
Such experimental data can be compared to the predicted distribution
Pˆ(n) in two ways. One
possibility is to use maximum likelihood estimation to find the value of
Â2 for which
Pˆ(n) best fits the data. This
is illustrated in Figure 3,
which shows that it is possible to obtain excellent agreement between the
theoretical and experimental distributions. The estimated value of
Â2 for Tsr-Venus is
Â2 = 3.57,
whilst for β-gal,
Â2 = 20.96.
The difference in magnitude between these two estimates may be partially due to
the fact that β-gal is only active as a tetramer. Thus,
each burst of activation measured experimentally (and thus available for
fitting) corresponds to the production of 4 monomers. The disadvantage of
fitting the model in this way is it can only provide an estimate of the single
parameter Â2, but not of the underlying
kinetic parameters α1,
α2,
β1, and
β2.
An alternative approach to verifying the model would be to obtain independent
estimates of the model parameters from which we can calculate
Â2 using Equation 6. The resulting
geometric distribution can then be compared to the observed burst size data.
Unfortunately, as is common for most models in cell and molecular biology,
direct experimental measurements of many of these rates are not available. For
the β-gal data, β2
can be obtained from the reported mRNA half life [7],[20], but the other
three parameters corresponding to the off-rate of the transcription factor and
to the binding rates of RNA polymerase to DNA and of mRNA to ribosome
respectively are not available.
We have shown that it is possible to use results from queuing theory to derive the
burst size distribution of protein molecules produced by a single transcription
factor binding event in terms of physically measurable kinetic rate constants for
both the simplest model of gene expression, the so-called Standard Model, and for a
number of natural extensions.
Furthermore, we have shown that the mathematical form of these models is
nonidentifiable, and all such burst size distributions are actually determined by a
single parameter. This implies that it is impossible to use burst size data alone to
determine the relative contributions of transcription and translation to the
variability in gene expression.
One possible way of overcoming this limitation is to use a combination of burst size
data and steady-state data. However, this requires estimates of a further two
parameters (which are not needed when using burst-size data alone). We were unable
to estimate all six parameters directly from the combined data. However, using
independent estimates of either the mRNA lifetime or the protein lifetime reduces
the number of parameters by one, and enables successfully estimation of the
remaining five parameters by maximizing an empirical likelihood using the
Nelder-Mead simplex algorithm. Although this suffers from the common problem of
occasional convergence to a local maximum, by using computing repeated estimates it
was possible to identify and exclude such cases and hence obtain good estimates of
the desired five kinetic parameters under the different constraints.
|
10.1371/journal.ppat.0030004 | In Vitro Derived Dendritic Cells trans-Infect CD4 T Cells Primarily with Surface-Bound HIV-1 Virions | In the prevailing model of HIV-1 trans-infection, dendritic cells (DCs) capture and internalize intact virions and transfer these virions to interacting T cells at the virological synapse. Here, we show that HIV-1 virions transmitted in trans from in vitro derived DCs to T cells principally originate from the surface of DCs. Selective neutralization of surface-bound virions abrogated trans-infection by monocyte-derived DCs and CD34-derived Langerhans cells. Under conditions mimicking antigen recognition by the interacting T cells, most transferred virions still derived from the cell surface, although a few were transferred from an internal compartment. Our findings suggest that attachment inhibitors could neutralize trans-infection of T cells by DCs in vivo.
| Dendritic cells (DCs) patrol peripheral mucosal sites, capturing and processing potential pathogens into antigenic peptides for presentation to T cells of lymphoid organs, and thereby initiating an immune response. HIV-1 had been proposed to use DCs as “Trojan horses,” hiding inside the DCs and surviving the degradation pathway to gain access to the lymph nodes and spread to the T cells. Our study challenges this “Trojan horse” model by showing that only HIV-1 virions bound to the surface of DCs, and not internalized virions, are transmitted to T cells. Even when T cells specifically recognized the antigen presented by DCs, the infection of T cells was principally mediated by virions remaining at the surface of the DCs. Interestingly, in this context of antigen-specific recognition, which increases the trafficking toward the immunological synapse of DC internal vesicles, where HIV-1 virions seem to hide, a few internal virions could infect T cells. Our findings suggest that in vivo transmission to T cells of HIV-1 virions captured by DCs should be more sensitive to neutralization than previously expected.
| To ensure their survival, microbial pathogens have evolved strategies to subvert the action of cellular components of the host immune system, including dendritic cells (DCs). DCs patrol peripheral mucosal sites, capturing and processing potential pathogens into antigenic peptides for presentation by major histocompatibility complex (MHC) class II to CD4 T cells in lymphoid organs, initiating an immune response (for a review, see [1]). HIV-1 has been proposed to usurp this natural function of DCs to spread efficiently. HIV-1 entering the body via the mucosa and other peripheral sites may be transported by DCs to CD4 T cells deeper in the mucosa or in lymphoid organs [2–4]. HIV-1 that reaches lymphoid organs can also take advantage of the formation of DC–T-cell conjugates to promote its replication and spread [5–7].
DCs can transmit HIV-1 to T cells via two pathways. In the de novo pathway, DCs are actively infected with HIV, leading to the budding and spread of new virions to neighboring CD4 T cells. In the prevailing model of the trans pathway, intact HIV-1 virions are captured by alternative HIV-1 receptors, which bind virions without triggering fusion, and internalized into clustered compartments resembling late endosome/multivesicular bodies (MVBs) [8,9]. After interacting with a CD4 T cell, HIV-1–loaded DCs redistribute the virion-containing vesicles to the virological synapse [8,10,11]; CD4, CXCR4, and CCR5 receptors on T cells are recruited to this region, facilitating trans-infection [10].
How HIV-1 virions survive the uptake pathway designed to capture and cleave pathogens into peptides for antigen presentation remains unknown. HIV-1 could divert the intracellular trafficking of immunological synapse components to avoid degradation and thus survive until later transmitted to T cells. Alternatively, external virions could be transmitted to CD4 T cells since some HIV-1 virions remain deeply tangled in membrane protrusions and microvilli of the plasma membrane [8]. Using functional assays that detect virion fusion and productive infection of CD4 T cells, we investigated whether trans-infection is mediated through internalized or external HIV-1 virions in monocyte-derived DCs (MDDCs) and CD34-derived Langerhans cells (LCs).
The potential effects of the state of DC maturation and coreceptor utilization by HIV virions in the trans and the de novo pathways in HIV-1 transmission from DCs to T cells were evaluated. These studies were performed with immature MDDCs or MDDCs matured with tumor necrosis factor α and poly(I:C) [12] and with two laboratory-adapted viral strains, CXCR4-tropic NL4–3 and CCR5-tropic 81A (Figure 1). MDDCs were incubated with virions at 4 °C to promote viral binding and were then either added to autologous activated T cells immediately or incubated for 1 to 5 d at 37 °C before mixing the autologous T cells. The MDDCs were then incubated with T cells for 24 h to allow virion transfer to T cells. After an additional 2 d of incubation in the presence of azidothymidine (AZT), productive infection of T cells was measured by immunostaining with anti-p24Gag (Figure 1A). Transmission of 81A (R5-tropic) virions from immature MDDCs to T cells was biphasic, as reported [11]. The early phase (0 to 1 d) involved the trans pathway; the later phase (1 to 5 d) involved the de novo pathway and was sensitive to the HIV protease inhibitor amprenavir (not shown). During the first day, 25% of R5-tropic 81A virions were transmitted by the trans pathway; 75% were transmitted by the de novo pathway over the ensuing 4 d (Figure 1B). In mature MDDCs, however, approximately 93% of virions were transmitted by the trans pathway during the first day. X4-tropic NL4–3 virions were transmitted by both immature and mature MDDCs principally by the trans pathway. Similar results were obtained when MDDCs were analyzed from nine different normal donors (Figure 1C).
In vivo, DCs may not immediately interact with T cells after virion capture. Accordingly, we investigated how a delay in T-cell contact might affect transmission through the trans pathway with a virion-based HIV-1 fusion assay [13,14]. MDDCs loaded with HIV-1 virions containing β-lactamase-Vpr (BlaM-Vpr) were incubated with autologous T cells, and fusion to CD4 T cells was monitored by the changes in fluorescence of CCF2, a BlaM substrate loaded into the cells (Figure 1D). When NL4–3 virions were presented immediately after binding to mature MDDCs, up to 24% of CD4 T cells displayed BlaM activity, indicating virion fusion. Transmission was less efficient when virions were presented by immature MDDCs. Fewer 81A than NL4–3 virions were transmitted, likely because there were fewer CCR5- than CXCR4-expressing cells in resting peripheral blood lymphocytes (PBLs). When virions were presented by MDDCs after incubation at 37 °C for up to 120 min, transmission efficiency decreased sharply (Figure 1E) in both immature and mature MDDCs. This rapid decrease was not due to a relative lack of sensitivity of the fusion assay in the context of trans-infection. As in our previous studies of T-cell infection with free virions, the fusion assay proved to be both sensitive and quantitative over a broad range of viral inputs in these DC–T-cell mixing experiments [13] (Figure S1). Thus, HIV-1 trans-infection from MDDCs to autologous CD4 T cells is efficient only for a limited time after virion capture.
Our results show that immature DCs preferentially transmit R5-tropic HIV-1 by the de novo pathway, as described [11,15–17], and X4-tropic HIV-1 by the trans pathway. Mature DCs transmit both R5- and X4-tropic virions mainly by the trans pathway. Since immature DCs are present at mucosal sites of viral entry, while mature DCs reside in lymph nodes and in the gut-associated lymphoid tissue (GALT), HIV-1 may exploit different transmission strategies at different anatomic sites in vivo. In healthy mucosa, immature DCs are likely to transmit R5-tropic HIV-1 principally via the de novo pathway, especially since the efficiency of the trans pathway declines rapidly (Figure 1E and [11,16]); the trans pathway might contribute to the local spread of virus from mucosal immature DCs to macrophages and CD4 T cells. However, in inflamed mucosal epithelium, which contains a greater proportion of mature DCs, HIV-1 transmission might preferentially involve the trans pathway, as in human cervical explants [4]. In lymph nodes and GALT, the proximity of mature DCs to T cells would further favor the trans pathway. Since DC–T-cell conjugates are major sites of HIV-1 production [5–7], trans-infection could be critical in the intense viral replication that characterizes the acute and chronic phases of untreated HIV-1 infection.
To identify the cellular compartment from which HIV-1 is transmitted in trans, we selectively neutralized surface-bound HIV-1 virions with truncated recombinant soluble CD4 (sCD4; AIDS Reagent Program [18]), which binds the HIV-1 envelope gp120 protein and prevents engagement of CD4 on T cells. Cells were treated at 4 °C to protect internalized virions from sCD4 exposure. NL4–3 virions containing BlaM-Vpr were bound to MDDCs and allowed to internalize, and cell-surface virions were neutralized with sCD4. In the absence of an internalization step, surface-bound virions on immature and mature MDDCs effectively fused to CD3+CD4+ cells (Figure 2A; bars 1 and 3); this fusion was effectively blocked by sCD4 (bars 2 and 4). However, when HIV-1 virions were internalized at 37 °C for 30 min before treatment (bars 5 to 8), sCD4 still inhibited virion fusion (bars 6 and 8). To confirm that sCD4 neutralized surface-bound virions without impairing subsequent virion transfer, two sets of HIV-1 virions were successively bound to immature MDDCs, but only the first was neutralized with sCD4 (Figure 2B). Similar amounts of HIV-1 fused to T cells regardless of the presence of previously neutralized cell-surface virions on the DCs (bars 3 and 5). Thus, sCD4 does not interfere with virological synapse formation or the transfer of virions bound after sCD4 treatment.
To further confirm the absence of virion transfer from internal cellular compartments, we performed sequential loading of HIV-1 expressing green fluorescent protein (GFP) (GFP-HIV) and then HIV-1 expressing cyan fluorescent protein (CFP) (CFP-HIV) [19] (Figure 2C). GFP-HIV was bound to MDDCs and incubated at 37 °C to allow virion internalization. Some residual GFP-HIV virions remained at the surface. Next, CFP-HIV was bound but kept at 4 °C to prevent virion internalization. MDDCs were then incubated with autologous T cells for 48 h. sCD4 treatment after binding and internalization of GFP-HIV but before CFP-HIV binding fully blocked transmission of GFP-HIV (Figure 2C, middle panel), indicating that residual surface-bound GFP-HIV was the source of virus for transmission to T cells (left panel). Under these conditions, CFP-HIV was still transferred to T cells, confirming that sCD4 treatment did not affect subsequent transfer of virions and was not inherently harmful. In the absence of treatment, many double-positive GFP+CFP+ T cells were observed, indicating that more than one virion can be transmitted to each T cell (left panel).
Thus, the neutralization studies showed that only surface-bound, but not internalized, HIV virions mediated trans-infection from MDDCs to T cells. sCD4 likely neutralizes HIV-1 virions by competing for viral binding to the CD4 receptor on T cells, as intact HIV-1 virions, including gp120, remained associated with the MDDCs after treatment (unpublished data). Although sCD4 has been reported to allow HIV-1 fusion to cells that do not express CD4 [20], sCD4 did not induce fusion to CD4– cells in our experiments, as demonstrated by the absence of BlaM transfer to CD8 T cells or B cells (not shown).
Next, we stripped surface HIV virions from MDDCs by proteolytic digestion (Figure 2D). GFP-HIV was again allowed to bind and internalize. CFP-HIV was restricted to the surface of MDDCs and served as a control for neutralization by the proteolytic enzymes. MDDCs loaded with GFP-HIV and CFP-HIV were then treated with trypsin as described [10,21] or pronase and incubated for 2 d with T cells. Trypsin treatment did not effectively remove externally bound CFP-HIV virions (unpublished data). However, in the presence of increasing amounts of pronase, fewer surface-bound CFP-HIV virions were transmitted to T cells, indicating increasingly effective removal the surface-bound HIV-1 by this protease cocktail. However, pronase had the same effect on the transfer of GFP-HIV, whether it had been internalized or not. Thus, successfully transmitted GFP-HIV virions appear to originate from the cell surface rather than from an internal, pronase-resistant compartment in MDDCs.
We considered the possibility that the low levels of transmission in the presence of sCD4 could correspond to transmission events from internalized HIV-1, masked under our experimental conditions. However, changes in viral input, internalization time, and viral strains failed to reveal significant transfer from internal compartments (Figure S2). We also studied a second type of DC, LCs derived from CD34+ cord blood cells (MatTek; http://www.mattek.com). As observed in MDDCs, the transfer of HIV virions from LCs to allogenic T cells again was mediated by virions bound at the surface of the LCs (Figure S3). Since MDDCs and LCs derived from CD34 progenitors are excellent surrogates of in vivo DCs, we conclude that most virions transmitted in trans in vivo likely originate from the cell surface.
These results in MDDCs and LCs sharply contrast with the report that formed the basis for the prevailing model of trans-infection [21]. In that report, surface-bound virions were neutralized by proteolytic digestion with trypsin. Although in our hands trypsin did not digest surface-bound virions as potently as pronase, the discordance of results likely lies in the use of different reporter systems. Kwon et al. [21] used a luciferase reporter that does not allow the distinction between infection of T cells or MDDCs. Since the immature MDDCs used in that study are highly susceptible to fusion with the R5-tropic BaL envelopes [12] and efficiently replicate CCR5-tropic HIV-1 [15,22], the luciferase activity may have derived from infected immature MDDCs, not T cells, in the coculture. Our flow-based assays, which permit a clear distinction between infection of T cells and MDDCs in coculture, reveal that HIV-1 virions transmitted in trans are sensitive to pronase treatment.
Our results also differ from three later studies, further supporting the notion that HIV infection of T cells by DCs involves the transfer of internalized virions from DCs to interacting CD4 T cells [9,10,15]. McDonald et al. [10] showed the recruitment of “trypsin-resistant” HIV-1 virions to the immunological synapse; however, functional assays were not performed to confirm that the interacting CD4 T cells were actually infected by the recruited virions. We suspect that, while virions may be transported to the synapse, these virions are not successfully transmitted. Wiley et al. [9] showed the release of infectious virions from HIV-1–loaded MDDCs, even after surface-bound virions were removed with trypsin. However, the efficacy of the trypsin treatment was only controlled in the experiments measuring the release of p24Gag in the supernatant, not in studies measuring the infectivity of these virions. Finally, Ganesh et al. [15] observed the transfer of some virions from MDDCs to T cells in the presence of neutralizing antibodies, a surprising result in light of our findings with sCD4. In their study, the efficacy of the antibody neutralization was measured with free virions but not with MDDCs bearing only surface-bound HIV-1 virions. The amount of antibody required to neutralize free virions might be lower than the amount needed to neutralize surface-bound virions on MDDCs, a possibility that would explain our divergent results. Of note, our findings are supported by a recent study showing that, in Raji cells expressing DC-SIGN (DC-specific ICAM-3 grabbing nonintegrin), surface-bound rather than internalized virions are transmitted in trans to 293T cells expressing CD4 and CCR5 [17].
Several reports have suggested that captured HIV-1 virions are stored in MVBs, raising the possibility that HIV-1 mediates trans-infection of T cells by highjacking a pathway involved in the trafficking of internal vesicles to the immunological synapse. In DCs, the transport of MHC class II from the MVB to the immunological synapse requires a T-cell–mediated signal [23]. Only T cells of the appropriate antigen specificity trigger this transport. Since antigen recognition could mobilize the release of HIV-1 virions from the MVB, we investigated trans-infection in the context of stimulation with a superantigen, staphylococcal enterotoxin B (SEB) (Figure 3). SEB activates T cells by crosslinking the variable region of T-cell receptor β-chain and the MHC class II molecule expressed on the DC surface [24].
NL4–3 virions containing BlaM-Vpr were bound at 4 °C to SEB-pulsed MDDCs. Viral transfer to autologous purified CD4 T cells was measured after allowing virion internalization, or not, by MDDCs. Again, sCD4 completely blocked HIV-1 transmission (Figure 3A). However, since the fusion assay does not require T-cell activation to generate a positive signal, trans-infection could be detected in the absence of engagement of the T-cell receptor and MHC class II. To further ensure that transfer was analyzed only when the MDDCs and T cells were effectively engaged, we measured productive infection of resting T cells by immunostaining for p24Gag (Figure 3B). Only T cells stimulated by SEB-loaded MDDCs are rendered permissive by releasing a postentry restriction block created by APOBEC3G (apolipoprotein B mRNA-editing enzyme, catalytic polypeptide-like 3G) [25]. Again, the vast majority of transmission events were neutralized by sCD4. However, under these experimental conditions, a few virions were transmitted from an sCD4-resistant compartment, as evidenced by the slight increase in transfer when SEB-loaded MDDCs were allowed to internalize HIV-1 virions at 37 °C. These transmission events were not due to new virion production by MDDCs since similar results were observed when AZT was added after 24 h of coincubation of MDDCs and T cells. When SEB-pulsed MDDCs were allowed to internalize HIV-1 virions for a longer period, HIV-1 virion transfer slightly increased peaking at 1 h of internalization (Figure 3C). Subsequently, the efficiency of transfer from internal compartments decreased, likely as a consequence of degradation or inactivation of the internalized virions.
In conclusion, HIV-1 virions transmitted in trans from DCs to T cells principally originate from the surface of DCs, except during antigen recognition, when a few internalized virions may also be transmitted to the antigen-specific T cells. Whether these rare events contribute to the preferential infection and elimination of HIV-specific T cells in vivo [26] is not known. Nevertheless, even within this context, the vast majority of transmitted virions are derived from the surface of DCs. Our results do not challenge the prevailing view that HIV-1 virions are internalized in the DCs. Indeed, we detected large amounts of internalized HIV-1 virions by microscopy. However, unless HIV-1–loaded DCs encounter T cells of the appropriate specificity, virion internalization appears to be a dead end for HIV-1 trans-infection. Since the C-type lectin receptors involved in trans-infection are localized in lipid rafts [27], surface-bound HIV-1 likely exploits the clustering of lipid rafts at the immunological synapse to enhance trans-infection of CD4 T cells. Because trans-infection principally involves surface-bound virions, our findings suggest that attachment inhibitors could be used to limit trans-infection of T cells by DCs, in vivo.
To study HIV-1 transmission from DCs to T cells, MDDCs (2 × 106) were incubated with 81A or NL4–3 virions (50 μg of p24Gag /ml) for 1 h at 4 °C, washed four times in cold PBS, incubated for 0 to 5 d at 37 °C, diluted 1:10, and added to autologous phytohemagglutinin-activated PBLs (2 × 106). Cocultures were maintained for 3 d in RPMI with 10% FBS, 20 IU/ml IL-2, 25 ng/ml IL-4, 50 ng/ml GM-CSF, and penicillin and streptomycin (100 μg/ml each); at 24 h, AZT (10 μM) was added to prevent further infection. Infected T cells were identified by intracellular immunostaining for p24Gag combined with antibodies against CD3, CD4, and CD1a. Productively infected T cells represent the percentage of p24Gag+CD4– cells in the CD3+CD1a– population and correspond to infected CD4 T cells that had effectively downregulated CD4 receptors due to expression of select viral gene products, including nef, vpu, and env [28]. In some experiments, an HIV protease inhibitor, amprenavir (40 nM) (Division of Acquired Immunodeficiency Syndrome, National Institute of Allergy and Infectious Diseases; http://www.niaid.nih.gov), was added during the binding step and maintained for the rest of the experiment.
MDDCs derived from CD14+ monocyte were induced to mature with poly(I:C) and tumor necrosis factor α [12]. The 81A or NL4–3 virions containing BlaM-Vpr (500 ng of p24Gag) [12–14] were incubated with MDDCs (2 × 106) or with CD34-derived LCs for 1 h at 4 °C, washed four times in cold PBS, and incubated at 37 °C for the indicated time to allow virion internalization or kept at 4 °C. Aliquots (2 × 105 cells) were added to autologous resting PBLs (2 × 106), and incubated at 37 °C for 1 h. HIV-1 fusion to CD4+CD3+ cells was measured using the virion-based fusion assay combined with immunostaining with CD1a-APC, CD4-PE Cy7, and CD3-APC Cy7 antibodies [13,14]. Cells were analyzed by flow cytometry (BD LSRII; Becton Dickinson, http://www.bd.com) and analyzed with FlowJo software (Treestar Software, http://www.flowjo.com).
GFP-HIV virions were bound to MDDCs for 1 h at 4 °C and allowed to internalize at 37 °C for 30 min; CFP-HIV virions were only bound to MDDCs. As indicated, surface virions were neutralized with sCD4 before or after the binding of CFP-HIV or by pronase after the binding of GFP-HIV and CFP-HIV. MDDCs were then incubated with autologous T cells for 48 h. Cells were immunostained with CD1a-APC, CD4-PE Cy7, and CD3-APC Cy7 antibodies and analyzed by flow cytometry.
To neutralize surface-bound virions, MDDCs or CD34-derived LCs loaded with HIV-1 virions were incubated for 90 min at 4 °C with 20 μg/ml sCD4 in RPMI and 10% FBS and extensively washed with PBS before MDDCs or CD34-derived LCs were added to T cells. To neutralize virions with pronase, the HIV-1–loaded MDDCs were incubated for 30 min at 4 °C with 50 to 400 μg/ml pronase (Roche, http://www.roche.com).
MDDCs (2 × 106) were pulsed with SEB (0.5 μg/ml) at 37 °C for 1 h. NL4–3 virions containing BlaM-Vpr (500 ng of p24Gag) were allowed to bind at 4 °C to the MDDCs; as indicated, cells were incubated for 30 min to 4 h at 37 °C to allow internalization. Surface-bound virions were then neutralized or not with sCD4. The HIV-loaded MDDCs were added to autologous purified resting CD4 T cells, and trans-infection was measured with the fusion assay at 2 h or by measuring productive infection after 3 d of coculture. Productively infected T cells were identified by intracellular immunostaining for p24Gag combined with antibodies against CD3, CD4, and CD1a. We then measured the percentage of p24Gag+CD4– cells in the CD3+CD1a– population. |
10.1371/journal.pntd.0005113 | Isothermal Diagnostic Assays for Monitoring Single Nucleotide Polymorphisms in Necator americanus Associated with Benzimidazole Drug Resistance | Soil-transmitted helminths (STHs) are the most prevalent intestinal helminths of humans, and a major cause of morbidity in tropical and subtropical countries. The benzimidazole (BZ) drugs albendazole (ABZ) and mebendazole (MBZ) are used for treatment of human STH infections and this use is increasing dramatically with massive drug donations. Frequent and prolonged use of these drugs could lead to the emergence of anthelmintic resistance as has occurred in nematodes of livestock. Previous molecular assays for putative resistance mutations have been based mainly on PCR amplification and sequencing. However, these techniques are complicated and time consuming and not suitable for resource-constrained situations. A simple, rapid and sensitive genotyping method is required to monitor for possible developing resistance to BZ drugs.
To address this problem, single nucleotide polymorphism (SNP) detection assays were developed based on the Smart amplification method (SmartAmp2) to target codons 167, 198, and 200 in the β-tubulin isotype 1 gene for the hookworm Necator americanus.
Diagnostic assays were developed and applied to analyze hookworm samples by both SmartAmp2 and conventional sequencing methods and the results showed high concordance. Additionally, fecal samples spiked with N. americanus larvae were assessed and the results showed that the Aac polymerase used has high tolerance to inhibitors in fecal samples.
The N. americanus SmartAmp2 SNP detection assay is a new genotyping tool that is rapid, sensitive, highly specific and efficient with the potential to be used as a field tool for monitoring SNPs associated with BZ resistance. However, further validation on large numbers of field samples is required.
| Hookworms are amongst the major STHs and the second most prevalent intestinal helminth of humans. Large-scale treatment with the benzimidazoles (BZs) albendazole or mebendazole is the major control strategy against STHs in mass drug administration (MDA) programs. Prolonged and repeated treatment with the same anthelmintics has led to the emergence of widespread BZ resistance in veterinary parasites which is caused by a single nucleotide polymorphism at codon 200, 167 or 198 in the β-tubulin gene. There is a considerable concern that prolonged use of the same anthelmintics with suboptimal efficacy against hookworms, may select for resistant parasites and favour the development of resistance. We developed a novel genotyping assay to screen for β-tubulin polymorphisms in N. americanus, using the SmartAmp2 method. SmartAmp2 is a unique genotyping technology that detects a mutation under isothermal conditions with high specificity and sensitivity. The N. americanus SNP detection assay is rapid, sensitive and highly specific and has the potential to be used in the field for the detection of SNPs associated with BZ resistance.
| Intestinal helminths cause a major burden on human health in developing countries, infecting more than 2 billion people worldwide [1]. Hookworms are one of the major STHs and the second most prevalent intestinal helminth of humans [2], infecting an estimated 438.9 million people in resource-constrained countries in the tropics and subtropics [3], and causing 22.1 million disability-adjusted life years [4]. Maternal hookworm anemia can complicate pregnancy, placing both mothers and children at higher risk of mortality. Infected children are compromised with anemia, stunted growth, and physical and intellectual growth deficits [5, 6]. Hookworm infections are mainly caused by N. americanus and Ancylostoma duodenale, with N. americanus being the most prevalent species, representing ~85% of all hookworm infections, and are associated with more morbidity worldwide than any other STH [7].
Current efforts to control morbidity caused by hookworms rely heavily on large-scale administration of ABZ or MBZ in MDA programs [8] which have been greatly expanded in recent years by massive donations of these anthelmintics. However, a single dose of either drug shows suboptimal efficacy against hookworms [9–13]. Treatment with these drugs is the major hookworm control strategy recommended by the World Health Organization (WHO) [14, 15] as there is no vaccine available. A major concern is that MDA over prolonged periods using the same anthelmintics would exert selection pressures on hookworm parasite populations and favour the development of resistance [8, 16]. In veterinary nematodes, BZ resistance has been developed and is caused by a single nucleotide polymorphism (SNP) in the β-tubulin isotype 1 gene that substitutes tyrosine (Tyr) for phenylalanine (Phe) at codon 167 or 200 (TTC>TAC) or alanine (Ala) for glutamate (Glu) at codon 198 (GAG>GCG) [17–21].
Benzimidazole resistance-associated mutations have been found in many veterinary nematodes particularly in nematodes in the same phylogenetic clade as hookworms [12, 22]. Such SNPs have already been observed in N. americanus and Trichuris trichiura [23, 24]. Furthermore, the presence of resistance-associated SNPs at codon 200 and 198 increased with treatment and were significantly higher in individuals who showed a poor response to ABZ (low efficacy) than in individuals who responded well to ABZ (good efficacy) in T. trichiura [24]. With the possibility that BZ susceptibility could be decreased by repeated rounds of MDA, it is important to monitor the level of resistance-associated SNPs in STHs before resistance becomes clinically manifested. The significant success and expansion of MDA programs for the control of human STHs, including hookworms, increases the urgency to monitor for drug resistance [25]. In human medicine, the egg reduction rate (ERR) is the gold standard for measuring drug efficacy and detection of resistance [26]. Relying on this efficacy measure alone is likely to be insensitive for detecting the early stages of the development of resistance and may only detect resistance when resistance allele frequencies are at high levels and treatment failures have already occurred [27, 28]. Furthermore, the ERR could be inappropriately interpreted as evidence of resistance, if standards similar to those used in the guidelines for anthelminthic of the World Association for the Advancement of Veterinary Parasitology were applied [10].
Molecular-based diagnostic tools are accurate and reliable [29, 30]. Diagnostic sequencing techniques are reliable and sensitive but are lengthy, complex, and too expensive for large-scale screening programs. With the rapid development of mutation-detection methods, several PCR based techniques have been developed to identify anthelmintic resistance, such as allele-specific PCR [18, 31, 32], RFLP-PCR [33, 34], real-time PCR [2, 35–37] and pyrosequencing [24, 30]. These methods are accurate and highly sensitive; however, they are unsuitable for large-scale screening of BZ resistance due to their complexity and the need for expensive equipment. Therefore, developing a rapid, simple, and accurate molecular method for monitoring for BZ resistance, without the need for expensive equipment, is highly desirable.
Here we report the development of a novel genotyping assay to monitor for the presence or absence of β-tubulin polymorphisms in N. americanus, using the SmartAmp2 method (Smart Amplification Process). SmartAmp2 is a unique DNA amplification method for rapid detection of target DNA or genetic polymorphisms under isothermal conditions, in a single step which eliminates the need for PCR amplification or a thermocycler [38]. SmartAmp2 is similar to LAMP, but LAMP technology uses symmetrical primer design. In SmartAmp2, however, the asymmetrical primer design is a key feature responsible for the suppression of the mismatch amplification, which minimizes the free-primer hybridization, other priming events and alternative mis-amplification pathways [39].
This method can detect a SNP with high specificity and sensitivity within 30 min [38]. The SmartAmp2 method uses Aac DNA polymerase which has strand displacement activity, combined with asymmetric primer design and Thermaus aquaticus MutS (Taq MutS) enzyme, which give the assay high specificity [38, 39]. Taq MutS is a mismatch binding protein [40], which recognizes a mismatched pair between the target DNA and the discrimination primer. This protein binds to mismatched nucleotides and blocks the dissociation of mismatched DNA by the Aac polymerase, which inhibits further amplification from non-target DNA [39].
The aim of this study was to develop molecular genotyping assays for the detection of the three β-tubulin polymorphisms associated with BZ resistance in other nematodes and validate their specificity and reliability in human hookworm samples and in fecal samples.
Hookworm eggs and larvae were collected during a study conducted in Sri Lanka on the comparative efficacy of different MBZ polymorphs for the treatment of hookworm infections and molecular markers of drug resistance in hookworms. Ethical clearance for the study was granted by the Ethics Review Committee of the Faculty of Medicine, University of Kelaniya (P39/04/2010).
Ethical approval (study 2535) was also obtained by Dr. Patrick Lammie, CDC, Atlanta, GA, and included the collection and examination of fecal samples from Haiti for helminth eggs, and DNA analysis of helminth eggs. Oral informed consent was obtained from all human adult participants and from parents or legal guardians of minors, as described previously [24, 30].
N. americanus adult worms, larvae and eggs were available in our lab [22, 28]. Additional larvae (L3) were cultured using the Harada-Mori technique from fecal samples collected in the field in Sri Lanka. All egg and larval samples were preserved in 70% ethanol after collection. Eggs and larvae were isolated under a dissecting microscope using a 10 μl pipette. Genomic DNA was extracted from larvae and eggs as described [41]. Lysis buffer was prepared as follow (KCl [50 mM], Tris [10 mM] with pH 8.3, MgCl2 [2.5 mM], 0.45% Nonidet P-40, 0.45% Tween 20 and 0.01% gelatine). Ten μl proteinase-K [10 μg / ml] (Invitrogen, Life Technologies; Burlington, ON, CA) and β-mercaptoethanol (Sigma-Aldrich, ON, CA) were added to 1 ml of this buffer just before use. Twenty-five μl of lysis buffer mix was added to previously isolated eggs and larvae and then tubes were incubated at 60°C for 2 h. Genomic DNA was extracted from adult N. americanus using QIAamp DNA mini kit (Qiagen, Hilden, Germany) according to the manufacture’s protocol.
To assist with SmartAmp2 development, control plasmids were engineered by site-directed mutagenesis as previously described [23]. Extracted genomic DNA from individual adult worms was used to amplify a fragment of the N. americanus β-tubulin isotype 1 gene including the codon positions 167 (exon 4), 198 and 200 (exon 5). Specific forward primer 5’-AAGAAGCTGAAGGATGTGACTG-3’ and specific reverse primer 5’-GAAGCGA AGACAGGTAGTAACAC-3’ (Invitrogen), were designed in the exonic regions of N. americanus genomic DNA sequence (GenBank accession no. EF392851). The PCR master mix contained 2 μl 10×PCR buffer, 1 μl (50 mM) MgSO4, 1 μl dNTP [10mM], 1 μl of each forward and reverse primer [10 μM], 1 U Platinum Taq DNA polymerase High Fidelity (Invitrogen), 2 μl genomic DNA and distilled H2O to reach a final volume of 20 μl. Negative controls (no template) were also included for quality control. The PCR reaction conditions were 94°C for 3 min, followed by 35 cycles at 94°C for 45 s, 57°C for 45s and 68°C for 1 min and a final extension at 68°C for 10 min. The resulting PCR fragments were Sanger sequenced to confirm the presence of sensitive alleles at codon positions 167, 198, and 200. Plasmids carrying mutations at position 167, 198, or 200 (MT) were engineered by site-directed mutagenesis. Primers for MT plasmids (outer primers and inner primers carrying the mutant alleles) are shown in Table 1. Amplified WT or MT fragments were cloned into TOPO-TA-Cloning vector (Invitrogen). Plasmid DNAs were extracted and purified using QIAprep miniprep plasmid kit (Qiagen) and subsequently sequenced by Sanger sequencing at the McGill University/Genome Quebec Innovation Centre, Montreal, Quebec. The purity and quantity of DNA in clones was measured using a Nano Drop Spectrophotometer, ND-1000 (Implen, Munich, Germany). Diluted WT and MT plasmids were used for assay optimization and development, and to determine the detection limit of each assay.
Primer sets were designed to amplify and detect putative β-tubulin mutations in the hookworm N. americanus. The online software version 1.1 (SMAPDNA), was made available by KK. DNAFORM, Japan (http://www.dnaform.jp/en/), and used initially to design the primers. Several primer sets were suggested by the software and further refinement in primer design was made by trials and evaluation tests and the best candidate primer set was selected for each assay. A primer set consists of five specific primers, the folding primer (FP), turn-back primer (TP), boost primer (BP), and two outer primers (OP1 and OP2), designed to recognize six different sequences on the target sequence. A set of primers has two discrimination primers, specific for either the WT or MT allele, differing only in one nucleotide at the 3’-end or the 5’-end. In this work, BP primer was selected to be the discrimination primer. The location and the sequences of primers for each SNP target are illustrated (Fig 1).
The SmartAmp2 assay was optimized using different concentrations of primers, MgSO4, and betaine and carried out in 25 μl reactions containing (2–3 μM) TP/FP, (1–1.5 μM) BP, (0.25–0.4 μM) OP1/OP2 (Invitrogen) 1.4 mM dNTPs (Invitrogen), (0.8–1.4 M) betaine (Sigma- Aldrich), 1x isothermal buffer (20 mM Tris-HCl (pH 8.6), 10 mM KCl, 10 mM (NH4)2SO4, (4–8 mM) MgSO4, 0.1% Tween 20, 1/100,000 dilution SYBR Green I (Invitrogen), 1 μg Taq MutS (Nippon Gene, Toyama,Japan/Wako Chemicals, Richmond, VA, USA) and 12 U Aac DNA polymerase (KK. DNAFORM, Yokohama, Japan). Control plasmids corresponding to WT or MT alleles were used to develop each assay and to evaluate the accuracy of genotyping between different primer sets. One microliter of WT or MT plasmids (~10 ng) was heated at 95°C for 3 min before being added to the assay. Reactions were incubated at 60°C for 60 min. The Rotor-Gene Q system (Qiagen) was used to maintain isothermal conditions and to monitor the change in fluorescence intensity of the intercalating SYBR Green I during the reaction. Assays were evaluated in terms of amplification (full match) and non-amplification (mismatch) within 60 min.
Further optimization was performed to estimate the sensitivity and reproducibility of the assays in individual samples and pools. Assays were tested on individual eggs/larva and pools (10–20 eggs/larva per pool). After DNA extraction from eggs and larvae using lysis buffer and proteinase K, 3 μl of this crude lysate was added to each reaction and then tubes were incubated at 60°C for 90 min. For evaluation of the sensitivity and the specificity of the assay to detect MT alleles in a background of WT DNA, MT plasmid DNA (~5 ng) was mixed with WT plasmid (~5 ng) in serial dilutions of 1:1, 1:9, 1:99 and 0: 100 and assays were carried out using the MT detection primer sets. These experiments were repeated twice and each DNA sample was analysed in duplicate.
Validation on field samples was performed using 110 individual samples obtained from Sri-Lanka. Pools of 20 larvae previously collected under microscopy for each individual sample were analyzed. Larval samples were digested using 25 μl of previous lysis buffer mix. From this crude lysate, 3 μl were added to each reaction after a DNA heating step at 95°C for 3 min. Assays were carried out in 25 μl reactions as previously explained. Positive (adult worm genomic DNA) and negative controls were always included as a reference in each experiment. Tubes were incubated in a real-time PCR at 60°C for 90 min.
PCR amplification of β-tubulin gene around codon positions 167, 198 and 200 were performed. Primers were designed as follow. For codon 167, forward and reverse primers were respectively 5’-AAGAAGCTGAAGGATGTGACTG-3’ and 5’-GGGTGGTTCCAGGCT GATGC-3’. For codons 198/200 (exon 5), forward and reverse primers were respectively, 5’-GGTTTCCGACACTGTGGTTG-3’ and 5’-GAAGCGAAGACAGGTAGTAACAC-3’. The PCR master mix contained 5 μl 10×PCR buffer, 1.25 μl (50 mM) MgSO4, 1 μl dNTP [10mM], 1 μl of each forward and reverse primer [10μM] (Invitrogen), 1 U Platinum Taq DNA polymerase High Fidelity, 3 μl of each DNA sample and distilled H2O up to 50 μl. The PCR reaction conditions were 94°C for 4 min, followed by 35 cycles at 94°C for 45 s, 57°C for 45 s and 68°C for 1 min and a final extension at 68°C for 5 min using a thermocycler (Biometra, Göttingen, Germany). Amplicons were identified on 2% agarose gel and visualized under UV (Bio-Rad Molecular Imager Gel Doc XR System). PCR products were sent to McGill University/Genome Quebec Innovation Centre, Montreal, Quebec for conventional Sanger sequencing to confirm and validate the genotyping results previously obtained with SmartAmp2 assay. Electropherograms were analyzed with Sequencher software (version 4.10.1) to identify the genotypes.
To validate the specificity of the assay on DNA extracted from fecal samples, genomic DNA was extracted from parasite-free fecal samples spiked with N. americanus larvae (~1000), or was extracted from the same fecal samples (with no larvae added) and used as a negative fecal control. As a positive control, DNA was extracted from purified L3 larvae (~1000). A protocol of QIAamp fast DNA stool mini kit (Qiagen) was used with some modifications. The fecal suspension (~ 1g) was kept on at 30°C to evaporate the ethanol. Then the InhibitEX buffer (QIAamp kit) was added together with approximately 150 mg of 0.5 mm glass beads (Sigma-Aldrich) and the sample vortexed on a Vortex-Genie 2 at 3000 rpm for 3 cycles of 5 min each, to disrupt the cell wall of the helminth samples. The fecal suspensions were then heat-shocked at 95°C for 5 min and placed in liquid nitrogen for 2 min, for 5 cycles. Then the proteinase K and AL buffer (QIAamp kit) were added and heated at 60°C for 1 h. The suspension was centrifuged at 14,000 rpm for 1 min and the procedures outlined in the QIAamp kit were then followed. Three μl of the spin column eluate (100 μl) was taken for each SmartAmp assay. The reaction mix for the SmartAmp assay was prepared and carried out as previously mentioned. Negative controls were used in all experiments.
Various sets of candidate primers were designed to genotype the β-tubulin gene at the three SNP positions. Screening of these primer combinations and assay conditions yielded an ideal primer set for each assay, which completed the amplification within 20–30 min from the target DNA sequence in a plasmid. Primer sets were chosen based on first, the speed and yield of the amplification, and second, the primer efficiency to discriminate between the full match and mismatch amplification. In the initial optimization of the assay and without the inclusion of Taq MutS, a 15 min delay for the mismatch amplification was achieved. With the inclusion of Taq MutS, a complete suppression of the mismatch amplification was observed up to 60 min (S1 Fig). All primer sets that displayed late full match amplification or a short delay between the full-match and mismatch amplification were omitted. The location and sequences of primers for each SNP target are shown (Fig 1). The SNP 167T/A (Phy167Tyr) occurs in exon 4 of the β-tubulin gene. A set of primers was designated TP, FP, BP, and OP2. The 5’-end of the BP discriminates the polymorphism 167T or 167A. In this assay, one of the outer primers was omitted to avoid the design of primers in the intron region. As the assay is highly specific, any polymorphism in the intron regions could affect the reproducibility of the assay.
Both 198A/C SNP (Glu198Ala) and 200A/T SNP (Phy200Tyr) reside on exon 5 of the β-tubulin gene. One set of primers was designed (TP, FP, OP1 and OP2) for both assays but a specific BP (discrimination primer) was designed in which the 3’-end of 198BP or 200BP discriminates the polymorphism at 198A/C or 200A/T, respectively. For each assay, BP primers were designed to be specific either for a MT variant or a WT allele. This unique primer design for SNP 198 and 200 makes reaction setup simple and easy to perform within a short time for a large number of samples.
Before testing the assay on hookworm samples, constructed WT and MT plasmids were used as DNA templates for assay optimization and development. Sequencing the MT plasmids revealed that the desired mutations at codons 167T/A, 198A/C and 200T/A of the β-tubulin gene were generated. The optimal amplification results were obtained when the reaction mixture contained 2 μM each TP/FP, 1 μM BP and 0.25 μM OP1/OP2 with 0.8 M betaine and 8 mM MgSO4. Primer sets with a specific BP for detecting the 167T/A, 198A/C, or 200T/A point mutations rapidly amplified the MT plasmid within 20–30 min, whereas the same primer sets failed to amplify the WT plasmid within 60 min. The WT primer sets amplified the WT plasmid but not the MT plasmid. Each assay was run in duplicate and all negative control reactions included in the experiments showed no amplification. These results confirmed that the SmartAmp2 assays were optimized as they accurately discriminated the full match amplification from the mismatch with complete suppression of the mismatch amplification. As an example, Fig 2 shows 2 different assays using a WT primer set (2A) and a MT primer set (2B) for the 200T/A β-tubulin SNP.
Further optimizations on single eggs/larva and pools of eggs or larvae were performed and full-match amplification using the WT primer sets was achieved with complete suppression of the mismatch amplification when the MT primer sets were used. Our assays amplified and genotyped DNA from single eggs and larva with high sensitivity and specificity within 40–50 min (Fig 3B). The WT primer set also amplified DNA from pools of eggs/larvae within 30–40 min with complete suppression of the amplification when the MT primer set was used (Fig 3B). The SmartAmp2 genotyping results and the Sanger sequence were always consistent.
To determine the specificity of each assay for detection of the MT alleles in a mix with WT DNA, serially diluted plasmids representing MT alleles at codon 167, 198 or 200, and wild-type plasmids, were used. The results of SmartAmp2 assays, using full-match MT primer sets for each SNP target, are illustrated (Fig 4). The MT alleles were detected even when present at only 1% of the WT DNA approximately at 50 min for the SNP 198 and 200 assays and at 60 min for SNP 167 assay. Up to 90 min, no background amplification was observed from the WT DNA or the negative controls. These results show that the assay is highly sensitive as it allowed detection as low as 1% of the MT alleles in a mix of WT DNA and the ability to genotype single and pooled hookworm eggs and larvae was also demonstrated. High sensitivity and specificity of the assay are particularly important in screening for mutations in individual samples with low levels of STH infections.
To validate the accuracy and specificity of the SmartAmp2 assays to handle field samples, we analyzed 110 individual field samples and a pool of 20 larvae per sample was examined. We detected the presence or absence of WT and MT alleles in all the samples within 30–40 min after incubation at 60°C. No background amplification (mismatch) was observed within 90 min. No amplification was observed from the negative controls., Three assays were performed for each sample to screen the three codon positions 167, 198 and 200. For codon 167 and 200, the WT primer set amplified the DNA target within 30–40 min. No amplification was observed with the MT primer set. None of the larval samples revealed significant levels of polymorphisms either at position 167 or 200. However, a polymorphism was identified at codon position 198A/C in some samples. The MT primer set for codon 198 allowed the amplification of the DNA target within 30–40 min (full match amplification).
The SNP198 detection primers recognized the SNP 198A/C of the β-tubulin gene to discriminate homozygous 198A/A (WT), mixed 198A/C (WT/MT), and homozygous 198C/C (MT) in genomic DNA samples. From the 110 samples examined by SmartAmp2 for the 198A/C assay, 90 samples were homozygous WT, 12 were mixed, and 8 were homozygous MT. Selected results for each of the three different genotypes using the WT and MT primer sets are illustrated (Fig 5A). The accuracy of our SmartAmp2 results was tested by amplifying the SNP target by PCR. The resultant amplicon was Sanger sequenced and showed concordance with the SmartAmp2 results (Fig 5B).
Positive (spiked with L3) and negative fecal samples were assessed in triplicate by SmartAmp2 assay. Full-match amplification was obtained only from positive fecal samples using the WT primer set within 40–45 min. High amplification efficiency was achieved when samples were diluted 1:4 in distilled H2O. The negative fecal samples and the negative control remained at baseline for at least 90 min. DNA extracted from purified L3 (positive control) produced a slightly faster amplification signal than genomic DNA from spiked fecal samples (Fig 6).
Diagnostics are vital to achieve successful elimination of parasitic infections and to aid against emerging pathogen resistance to the limited number of anti-parasitic drugs. The low sensitivity of current field diagnostic tools could miss the early stages of resistance. The lack of rapid, simple and reliable diagnostic tools for intestinal nematodes prevents accurate estimation of the distribution of BZ-resistant populations, the determination of at-risk populations and the burden of disease [42]. In this study, we developed a new SNP genotyping assay based on the SmartAmp2 method for monitoring β-tubulin polymorphisms. Primer sets were selected and optimized specifically to target F200Y, F167Y (TTC/TAC) and E198A (GAG/GCG) SNPs in the hookworm N. americanus. SNP-detection primer sets were able to efficiently and rapidly discriminate MT and WT genotypes using plasmids as DNA templates. Assays showed high reproducibility and sensitivity for detecting genomic DNA from pools and single egg/larval DNA in a single amplification and detection step. Compared with PCR-based methods, in order to genotype single egg/larval DNA, a nested PCR using the same forward and reverse primers is required, followed by gel electrophoresis and pyrosequencing [24]; a multistep technique that is time consuming and increases the risk of contamination as a result of manipulating PCR products. Additionally, the SmartAmp2 assays showed high specificity and allowed the detection of as little as 1% of the MT alleles in a mix with WT plasmid.
Assays detected the E198A (GAG/GCG) SNP in the N. americanus larvae. The MT-detection primer set detected and identified the MT alleles in pools of larval samples from Sri Lanka. To our knowledge, this is the first time that 198SNP has been detected in N. americanus. Amplification was evident within 30–40 min using as few as 20 larvae per pool. In addition, genotyping from single larva and eggs was achieved within 40–50 min.
Fecal samples spiked with N. americanus larvae were processed in the SmartAmp2 assay and the results showed high tolerance of the Aac polymerase to fecal inhibitors. No significant difference in the amplification efficiency between spiked fecal extracts and purified larval DNA was observed and any slight difference could be explained by the presence of remaining fecal inhibitors in the fecal samples. SmartAmp2 assays for genotyping hookworm samples are highly sensitive and specific using the Aac polymerase which is tolerant to inhibitors in stool samples; however, the assay sensitivity in stool samples could be compromised by the capacity of the extraction method to obtain good quality and quantity of purified DNA, the number of eggs in stool samples particularly in individuals with low level of infection, the amount of stool used for DNA extraction and an uneven distribution of eggs in the stool. To improve the performance of SmartAmp2 assay in fecal samples, sample collection and processing should be simple and ensure high DNA yields. Preliminary enrichment of eggs through sugar flotation followed by repeated cycles of freeze/boiling to crack the egg shell and liberate DNA [43] may significantly improve DNA recovery, assay sensitivity and reproducibility. Commercial DNA extraction methods are reliable and efficient for removing PCR inhibitors present in fecal samples; however, these methods use a small amount of feces and this could affect the sensitivity of the detection assay.
SmartAmp2 assay is relatively inexpensive; the main costs are for the Taq MutS and Aac polymerase. SmartAmp2 Primers used in this study were regular primers not HPLC primers. Additionally, reducing the reaction mixture to 10 μl and using in-house prepared buffers and reagents also reduces cost. Our data were generated on a Real-Time PCR system to follow the formation of double stranded DNA in real time, using SYBR green. However, end point detection system for monitoring fluorescence that would allow high throughput analysis of samples in a 96-well microplate format could be employed. Other approaches for visualizing the formation of DNA residues could be applied using fluorescence dyes that allow colorimetric inspection of the results.
In a SmartAmp2 assay, the presence of both the WT and MT alleles can be detected in a sample, as amplification would occur with both WT and MT primer sets. The relevance would be that if the MT is detected, as well as the WT, there would be the potential for some parasites being resistant and that with further anthelmintic selection the frequency of the MT allele might increase, increasing the risk of phenotypic resistance. The absence of definitive evidence of anthelmintic resistance in human parasites does not mean that resistance alleles are not present, at least at low frequencies. Such resistance alleles could increase with prolonged and repeated selection pressure. The lack of detection of phenotypic resistance may, in part, be due to the lack of a reliable and sensitive method to monitor for resistance alleles before and after BZ treatment in control programs [44], a low frequency of resistance alleles, and the probability that BZ resistance is recessive, as it is in veterinary parasites [45].
The present study provides evidence that the SmartAmp2 method targeting β-tubulin polymorphisms in N. americanus allowed direct detection of SNPs of a target DNA sequence in fecal samples. Additionally, these results indicate that our SNP genotyping assays are rapid, simple, very sensitive and highly specific which provide a unique tool for investigating the possibility of developing BZ resistance in the hookworm N. americanus. The development of sensitive and practical methods for early detection of resistance using molecular diagnostic tools that could be adapted to the field is urgently needed in order to sustain the benefits of helminth control programs.
|
10.1371/journal.ppat.1006172 | Substrate-analogous inhibitors exert antimalarial action by targeting the Plasmodium lactate transporter PfFNT at nanomolar scale | Resistance against all available antimalarial drugs calls for novel compounds that hit unexploited targets in the parasite. Here, we show that the recently discovered Plasmodium falciparum lactate/proton symporter, PfFNT, is a valid druggable target, and describe a new class of fluoroalkyl vinylogous acids that potently block PfFNT and kill cultured parasites. The original compound, MMV007839, is derived from the malaria box collection of potent antimalarials with unknown targets and contains a unique internal prodrug principle that reversibly switches between a lipophilic transport form and a polar, substrate-analogous active form. Resistance selection of cultured P. falciparum parasites with sub-lethal concentrations of MMV007839 produced a single nucleotide exchange in the PfFNT gene; this, and functional characterization of the resulting PfFNT G107S validated PfFNT as a novel antimalarial target. From quantitative structure function relations we established the compound binding mode and the pharmacophore. The pharmacophore largely circumvents the resistance mutation and provides the basis for a medicinal chemistry program that targets lactate and proton transport as a new mode of antimalarial action.
| The fight against malaria, i.e. one of the three major infectious diseases and transmitted by mosquitos, is conducted at three levels: i. transmission control (by attacking the mosquito vector or biological processes of vector infection), ii. vaccination (by stimulating the immune system to produce antibodies against molecular parasite surface structures), and iii. antimalarial drugs (by developing and applying small molecules that interfere with vital biochemical pathways). Despite strong efforts on levels i. and ii., small molecule drugs remain an indispensible antimalarial means; however, emergence and spreading of resistant malaria parasites against all currently used drugs pose a growing threat to treatment success. Therefore, novel drug targets need to be identified and exploited. Here, we show that a recently discovered lactic acid transporter, PfFNT, is a novel valid drug target and we provide first compounds that potently block transport and kill malaria parasites. Lactic acid is the metabolic end product of the parasites’ energy generation metabolism and interfering with this biochemical pathway represents a new mode of action against malaria parasites.
| All currently used antimalarial drugs have caused resistance in the parasite [1]. Hence, novel druggable targets are urgently needed to reload and diversify the therapeutic arsenal. Hitting the glycolytic energy generation pathway is a tempting approach as it is crucial for parasite survival [2,3] (Fig 1A). Earlier studies have shown that targeting glycolysis is effective against rapidly proliferating cells, such as human-pathogenic parasites [4,5] and tumors [6]. However, specificity issues derive from the evolutionary conservation of the involved glucose transporters and glycolytic enzymes between the pathogens and the human host. In this regard, the recently discovered Plasmodium falciparum lactate transporter [7,8], PfFNT, represents an elemental exception because the human genome does not encode similar proteins. PfFNT is a member of the microbial formate-nitrite transporter family (FNT) [9] and acts as a high capacity lactate/proton symporter. Human lactate transporters, e.g. of erythrocytes, are members of the monocarboxylate transporter family (MCT) [10] and differ fundamentally from PfFNT in terms of protein structure and transport mechanism [7]. Lactic acid, in dissociation equilibrium with the lactate anion plus a proton, is the metabolic end product of glycolytic glucose breakdown in plasmodia, and swift release from the cytoplasm is vital for maintaining the parasite’s energy flux and pH homeostasis [7,8,11–13] (Fig 1A). Current inhibitors of PfFNT, such as cinnamic acid derivatives [7] or niflumic acid [8], exhibit too low affinity and selectivity for therapeutic use. Nevertheless, addition of such compounds to cultured P. falciparum killed the parasites [14].
Here, we describe the discovery of potent inhibitors of PfFNT from a 400-member antimalarial compound collection, malaria box [16], which are highly effective against the parasite and in a PfFNT lactate transport model in yeast [7]. We identified a fluoroalkylated vinylogous carboxylate structure as the pharmacophore that leads to a potentially irreversible compound interaction. The compounds contain a novel prodrug principle based on intramolecular cyclization that reversibly converts a lipophilic state with good cell accessibility into a polar active form. Selection of a resistant P. falciparum line resulted in a single nucleotide exchange in the PfFNT encoding gene and the corresponding mutation rendered lactate transport based on the yeast-expressed transporter insensitive to the compound. We established the binding mode and generated first compounds that circumvent the resistance mutation. Our results show that PfFNT is a valid novel drug target and provide a chemical basis for the development of a new class of potent antimalarial drugs.
We employed a yeast system that we established earlier for PfFNT expression and inhibitor screening [7]. It is based on a strain [17] lacking the endogenous monocarboxylate transporting proteins Jen1 and Ady2. The assay detects uptake of 14C-labeled substrate via PfFNT over time yielding transport rates. Our criterion for hit identification was complete inhibition of transport in a 1 mM substrate gradient at 10 μM compound concentration. Individual malaria box compounds were added 20 min prior to the assay. The screening yielded two compounds, MMV007839 and MMV000972, that fully blocked PfFNT (Fig 1B; S1 Table).
Both compounds inhibited lactate transport of PfFNT to half-maximal rates at 170 nM (Fig 1C and 1D). The compounds also potently killed cultured P. falciparum 3D7 parasites (Fig 1C and 1D) with IC50 values of 140 nM (MMV007839) and 1.7 μM (MMV000972) determined after 2 days of incubation; the variation in potency possibly derives from different compound uptake efficiency across the various lipid membranes or metabolic conversion. Short period incubations with MMV007839 at IC90 for 1 h and over night reduced growth by 17% and 70%, respectively (S1 Fig), i.e. a time course that is in line with a compound targeting energy generation rather than acting acutely cytotoxic. MMV007839 further inhibited the red blood cell lactate transporter [18], MCT1, yet with 300 times lower efficiency (IC50 = 55 μM; Fig 1E). Since FNTs and MCTs are unrelated regarding their protein structure and transport mechanism, inhibition by MMV007839 suggests that the compound acts as an analog of the common lactate substrate and interacts at a lactate interaction site.
MMV007839 and MMV000972 exhibit the same structural scaffold, which differs only in the aromatic substitution (Fig 1F). Both molecules contain a fluoroalkyl chain attached to a cyclic hemiketal structure and a phenone moiety. In this form, however, similarity to the lactate substrate is not obvious. Further, the benzene ring renders the compounds considerably larger than the small acid substrates of FNT facilitated transport; therefore, MMV007839 and MMV000972 will most likely not pass PfFNT.
We then investigated the binding and dissociation kinetics of MMV007839. Half maximal PfFNT inhibition was reached within minutes at the tested nanomolar to low micromolar compound concentrations (Fig 2A). To measure the off-rate, we pre-incubated PfFNT expressing yeast cells with MMV007839 to obtain half and full inhibition, respectively, washed out the compound, and analyzed re-gain of transport activity in inhibitor-free buffer over time. During the assay period, the cells were kept in the absence of nutrients to restrict new production of PfFNT protein providing constant assay conditions for two hours. Within this timeframe, PfFNT exhibited normal functionality as seen by the 50% inhibition curve (Fig 2B). In view of the rapid on-rates it was quite unexpected that the inhibitory effect remained stable despite the absence of inhibitor. This indicates that dissociation of MMV007839 is very slow or binding may even be irreversible [19] (Fig 2B) further underscoring that the compound is not a transport substrate of PfFNT.
An irreversible inhibitor interacts with its target in a time-dependent fashion and the reaction proceeds towards completion rather than equilibrium [19]. In such a case, IC50 values decrease with elongated incubation times. To test for this, we added MMV007839 to PfFNT expressing yeast cultures 24 h before the assay and kept the cells at 4°C for the last 18 h to minimize growth and new production of PfFNT protein. The treatment did not affect yeast viability; yet, we determined a ten times lower IC50 of 15 nM (Fig 2C), which consolidates very slow off-kinetics or irreversible binding of MMV007839.
We obtained an inhibitory concentration, Ki, of 1.4 μM by plotting the observed on-rate constants, kobs, of PfFNT inhibition (shown in Fig 2A) against the concentration of MMV007839 (Fig 2D) [19]. Comparison to the earlier determined affinity of lactate to PfFNT (Km = 87 mM) [7] indicates that MMV007839 binds with > 60,000 times higher affinity than the physiological substrate.
Next, we aimed at identifying structural elements of the inhibitor compounds that are essential for nanomolar PfFNT blockade and efficient absorption by infected red blood cells. To do this, we systematically varied the MMV007839 scaffold (Fig 3, center). The incubation time of PfFNT expressing yeast cells with the test compounds was kept constant at 20 min in order to prevent time-dependent shifts in the IC50 (curves are shown in S2 Fig).
Altering the hemiketal moiety to a vinylogous lactone [1] (Fig 3, top center) resulted in an inactive compound indicating that the hemiketal structure is indispensible for PfFNT inhibition. Equally, loss of activity was obtained by removal of the fluoroalkyl chain plus replacement of the hemiketal by an ether [2]. Changing the position of the aromatic substituent from para to meta relative to the carbonyl, [3] and [4], led to a reduction in activity by at least two orders of magnitude (Fig 3, top right). Introduction of a methyl branch at the α-position to the carbonyl [5] yielded almost three orders of magnitude lower efficiency than MMV007839. Replacement of the carbonyl by an oxim moiety [6] substantially reduced the inhibitory activity (Fig 3, top left). Together, structural changes to the MMV007839 scaffold result in loss or dramatic reduction of activity.
We thus focused on modification of the fluoroalkyl chain regarding fluorine content and length (Fig 3, bottom left). The most modest alteration, i.e. removal of a single fluorine atom from the pentafluoroethyl chain [9], reduced efficiency by half. Shortening of the fluoroalkyl chain to di- [7] or trifluoromethyl [8] was less well tolerated than elongation to heptafluoropropyl [10]. Even extension to a four-carbon nonafluorobutyl chain [11] maintained inhibitory activity, though in the single-digit micromolar range, showing that the original pentafluoroethyl chain is optimal. Yet, extension by one fluorinated methyl unit [10] retained sub-micromolar antimalarial potency.
Finally, we produced a series of compounds with increasing volume of the aromatic substituent (Fig 3, bottom right). In this direction, the original MMV007839 compound was most tolerant to modification. Removal of the methoxy group (corresponding to MMV000972), or replacement by chlor [12], ethoxy [13], isopropoxy [14], and even benzyloxy [15] still yielded sub-micromolar inhibitors of PfFNT. Comparison of the yeast and P. falciparum data shows that the para-aromatic substituent contributes less to target affinity but mainly improves uptake by infected red blood cells. In this regard, compounds [13] and [14] are particularly striking because they represent improvements over MMV007839 with up to three times higher potency of 50 nM and 100 nM in IC50, respectively, against cultured P. falciparum parasites.
Hemiketals undergo reversible conversion, accordingly, this lipophilic form of MMV007839 linearizes to a polar vinylogous acid (pKa (pred.) = 5.0; Fig 4A) and vice-versa. Correlation NMR shows both compound states in solvent dependent equilibrium, e.g. 65% hemiketal (characterized by two protons at the α-carbon “i”) and 35% acid (single “i” proton) in CDCl3 (Fig 4A).
Considering the PfFNT transport mechanism, i.e. attraction of the lactate anion, subsequent proton transfer, and passage of the neutral lactic acid via lipophilic constriction sites [7], we suspected the linear form of MMV007839 to represent the active inhibitor. In this form, the inhibitor would mimic two consecutive lactate substrate molecules: one in its charged lactate anion form (deprotonated vinylogous acid moiety) and one in its neutral lactic acid form (fluoroalkyl chain). MMV007839 may, thus, interact with PfFNT in a mechanism-based type by binding simultaneously to the polar, pre-transport lactate attraction site and to the lipophilic transport path.
To test this, we synthesized an MMV007839 variant, BH-296, lacking the phenolic hydroxyl group, which prevents formation of a cyclic hemiketal (Fig 4B). In yeast, BH-296 was equally active in blocking PfFNT as MMV007839 (Fig 4B, black curve) showing that the linear, vinylogous acid is indeed the active form. When tested in P. falciparum culture, BH-296 was 25 times less efficient (Fig 4B, red curve) clearly hinting at poor absorption due to compound polarity. We, thus, conclude that in MMV007839 the cyclic hemiketal represents an internal prodrug facilitating penetration of consecutive lipid membranes. We synthesized an alternative, ethyl ester prodrug form of BH-296 (Fig 4B, blue). Yet, in P. falciparum culture the compound was no improvement over the free vinylogous acid of BH-296 (Fig 4B, blue curve). Either activation of the ester prodrug occurred before entering the parasite, i.e. in the medium or the red blood cell cytosol, or the prodrug was too stable for sufficient release of BH-296. This further demonstrates the advantage of the reversible, internal prodrug principle.
A consequent next step towards identification of the minimal requirements for PfFNT inhibition by a small molecule was to eliminate the phenol ring altogether, yielding a compound of only six carbons in length (Fig 4C). Considering the very small size and limited capability for interaction or shielding from the solvent, the compound was remarkably efficient in blocking PfFNT (IC50 = 1.9 μM). Hence, we refer to this molecule as the pharmacophore (Fig 4C). It should be noted that in terms of shape, volume, and charge distribution this structure very well resembles a lined-up assembly of a neutral lactic acid molecule followed by a lactate anion.
Constant exposure of P. falciparum cultures to 3 × IC50 concentrations of MMV007839 gave rise to selection of resistant parasites in less than three weeks. DNA sequencing of the PfFNT gene revealed a single nucleotide exchange in the resistant parasites, resulting in the replacement of Gly107 by serine on the protein level (Fig 5A). Gly107 is situated in one of the two most highly conserved and functionally relevant FNT regions [20], i.e. the L2 loop that interrupts transmembrane span 2; an analogous protein structure, L5, is present in transmembrane span 5 (Fig 5B). When inspecting 71 representative bacterial and protozoal FNT protein sequences [8], we found glycine and serine but no other amino acid residue at the position corresponding to PfFNT Gly107 (Fig 5B). In a PfFNT structure model [7], the mutation site is located at the cytoplasmic lactate entry site of PfFNT suggesting that the larger sidechain of serine interferes with MMV007839 binding (Fig 5C).
We determined the potency of MMV007839 on the selected, resistant parasites and found a 250 fold shift in IC50 to 35 μM compared to the non-resistant parental parasite line (Fig 5D, red curve). To test whether resistance against MMV007839 is directly connected to the identified PfFNT G107S mutation, we expressed the mutant protein in yeast. Treatment of PfFNT G107S with MMV007839 yielded a similar shift in IC50 to 21 μM (Fig 5D, black curve). We extended the data set and confirmed the binding site of MMV007839 to the cytoplasmic FNT substrate entry by expressing and testing the lactate transporter from a related parasite, Babesia bovis [22] (BbFNT), in yeast. BbFNT naturally carries a serine at the position corresponding to PfFNT Gly107 (S3 Fig). Accordingly, wildtype BbFNT should exhibit resistance against MMV007839, whereas exchange of the serine by glycine (BbFNT S93G) should significantly increase inhibition by MMV007839. Indeed, we found an IC50 of 150 μM for wildtype BbFNT and a 14 times higher efficiency with the BbFNT S93G mutant (IC50 = 11 μM; Fig 5E). Taken together, the data confirm PfFNT as the site of action of MMV007839 in malaria parasites. As a consequence, our findings validate PfFNT as a potent novel and druggable antimalarial target.
The PfFNT G107S mutant exhibits a slightly reduced lactate transport rate (Fig 6A and 6B). Nevertheless, the mutation remained stable in the cultured parasites after removal of the MMV007839 selection pressure and the resistant parasites exhibited the same fitness as the 3D7 wildtype strain (S4 Fig). We further tested for the mutational flexibility at the Gly107 site by replacing serine with the isosteric amino acid residues alanine and cysteine, and with the slightly larger valine. Only alanine and cysteine retained some low transport activity (20 and 10%, respectively) despite even higher expression levels than PfFNT wild-type (S5 Fig) indicating that amino acid exchanges of G107 with residues larger than serine are unlikely to occur in the parasites (Fig 6A and 6B).
We hypothesized that the low selectivity of MMV007839 for PfFNT G107S (Fig 6C) may derive from collision of the serine sidechain with the benzene ring of MMV007839 carrying a hydroxyl group in the active vinylogous acid form. Therefore, we tested our previously generated compound BH-296 which lacks the hydroxyl for inhibition of PfFNT G107S. The IC50 obtained with BH-296 in the yeast system was 2.3 μM (Fig 6D), i.e. 16 times lower than with wildtype PfFNT. However, BH-296 was one order of magnitude more efficient in inhibiting PfFNT G107S than MMV007839 (Fig 6C). This improvement was somewhat less pronounced in resistant P. falciparum culture (IC50 = 8.6 μM; Fig 6D) due to the earlier observed uptake issues of the polar BH-296 compound. Still, BH-296 represents a major improvement in selectivity for the resistance mutation, because the resistant parasites were only 2.5 fold less sensitive to BH-296 than the non-resistant parental parasites, which is in clear contrast to MMV007839 for which the sensitivity change in the resistant parasites was 250 fold.
Importantly, treatment of the MMV007839-resistant parasite culture with 3 × IC50 concentrations of BH-296 did not give rise to new resistance, which is in agreement with the observed limited mutational flexibility of PfFNT G107 (Fig 6A and 6B).
We figured that a less voluminous and more flexible inhibitor scaffold would further increase binding to the resistant PfFNT G107S mutant. As a proof of principle we found that the minimal pharmacophore compound has a four times better relative efficiency than BH-296 in blocking PfFNT G107S (IC50 = 8.1 μM; Fig 6C and 6E). This molecule can, thus, serve as a core structure for an expansion library of drug-like compounds that address the G107S resistance mutation.
The malaria box [16] is derived from more than 5 million compounds of which about 20,000 exhibited < 4 μM potency against parasite growth; 400 of these hits were selected for chemical diversity of the scaffold. Further, a safety factor of > 10 for cytotoxicity against a human cell line (HEK-293) was specified [23]. Our discovery of potent PfFNT inhibitors from the malaria box is an example of the power of phenotypic screening [1,24], particularly in the search for anti-infective compounds and shows the relevance of PfFNT as a target. In situations where lipid bilayer penetration is critical, such as multi-membrane shielded intracellular plasmodia [25] or mycobacteria with an exceptionally tight membrane [26,27], the phenotypic screening approach delivers compounds, which exhibit both, potent activity on the target protein and excellent absorption.
This way, we were able to identify a novel reversible prodrug principle that rapidly interconverts the antimalarial MMV007839 compound between a lipophilic transport form and a polar active form (Fig 4A). In combination with the observed very low or possibly even absent off-kinetics, the active form is permanently eliminated from the equilibrium, driving the reaction towards binding to the PfFNT target. Reversibility renders the prodrug non-consumable, permitting re-use of the principle for passage of consecutive membranes. Reversible hemiketals should be applicable to other compounds to improve absorption.
The active form of MMV007839 contains similar structural elements as niflumic acid (S6 Fig), i.e. the former most efficient PfFNT inhibitor [8] with an IC50 > 100 μM. Despite a different scaffold, both compounds carry a fluoroalkyl chain (pentafluoroethyl vs. trifluoromethyl) as well as an acidic group (vinylogous vs. standard carboxyl). Hence, both molecules may interact similarly with PfFNT at the cytoplasmic lactate entry. Niflumic acid is more voluminous, probably hindering interaction, whereas the linear MMV007839 mimics the lactate substrate better and rapidly gains deep access to the protein core (Fig 6F). Tight, lipophilic interaction of the fluoroalkyl chain in combination with electrostatic interaction of the vinylogous acid anion and efficient shielding from the solvent can explain the extraordinary long residence time of MMV007839 on PfFNT. Long target residence times correlate with a drug’s potency and selectivity, and have the potential to ameliorate off-target-based toxicities [28]. The Ki value of 1.4 μM for MMV007839 must be regarded as the lowest estimate of PfFNT affinity because Ki determination in the yeast system includes diffusion or transport of MMV007839 across the membrane to allow for binding to the cytoplasmic side of PfFNT, which results in slower apparent association kinetics. Hence, the true Ki of MMV007839 can be expected to reside in the nanomolar range.
The effect of the PfFNT G107S resistance mutation on MMV007839 inhibition is severe. However, the extraordinary affinity of the small pharmacophore renders it a promising starting point for the development of potent, substrate analog-type inhibitors that circumvent the mutation. This endeavor is worthwhile in view of the basically absent mutational flexibility at the PfFNT G107 site, since mutations to larger residues than serine abolish the physiological lactate transport function.
Our data suggest the following binding mode of MMV007839 to PfFNT (Fig 6F): MMV007839 binds in its linear vinylogous acid form to the cytoplasmic transporter entry site with the fluoroalkyl extending into the lipophilic protein center and the aromatic substituent pointing out of the too narrow transport path. The spatial restrictions in the core allow for one more fluoromethyl unit, whereas elongation of the aromatic substituent is unproblematic. The G107S resistance mutation reduces the diameter at the binding site, leading to a collision mainly with the aromatic phenol group. Mutations resulting in larger amino acid residues than serine are not tolerated because this would affect passage of the physiological substrate lactate. Insertion of a less voluminous and more flexible alkyl chain into the inhibitor scaffold circumvents the serine sidechain (Fig 6G). A hydroxyl moiety is required in the scaffold to utilize the internal hemiketal prodrug principle for better absorption. The substrate-analogous pharmacophore can be extended by suitable ligands (“R” position in Fig 6G) to increase inhibitor affinity by shielding from the solvent, and for optimization of the pharmacokinetic properties.
In conclusion, our findings validate PfFNT as a novel antimalarial target of glycolytic energy generation and pH homeostasis in malaria parasites. Plasmodial lactate/proton co-transport represents a unique and therapeutically exploitable mechanism in that it depends on a protein that has no structural counterpart in the human host. With the identification of the pharmacophore, its binding mode, and the reversible prodrug principle we provide the basis for a medicinal chemistry approach towards the establishment of a new class of antimalarial drugs.
Codon-optimized PfFNT in the yeast expression vector pDR196 has been described [7]. Open reading frame DNA of rat MCT1 was kindly provided by H. Becker, Kaiserslautern, Germany; Spe I and Sal I restriction sites were introduced by PCR (fw: gagaga ACT AGT ATG CCA CCT GCG ATT GGC GGG CCA GTG / rev: gagaga GTC GAC GAC TGG GCT CTC CTC CTC CGC GGG GTC) for cloning into pDR196. BbFNT DNA (NCBI# XP_001608703.1) was synthesized (GenScript) and cloned into pDR196 via Spe I and Xho I. Point mutations were introduced into PfFNT and BbFNT using the QuikChange protocol (Stratagene) and primers with respective nucleotide exchanges (Life Technologies; for primers see S2 Table). All generated constructs encode an N-terminal hemagglutinin epitope plus a C-terminal 10 × His tag and were sequenced for verification. W303-1A jen1Δ ady2Δ (MATa, can1-100, ade2-loc, his3-11-15, leu2-3,-112, trp1-1-1, ura3-1, jen1::kanMX4, ady2::hphMX4) yeast cells, kindly provided by M. Casal [17], were transformed using the lithium acetate/single stranded carrier DNA/polyethylenglycol procedure [29]. Transformed yeast was grown at 30°C in selective media (SD) with adenine, histidine, leucine, tryptophan, and 2% (wt/V) glucose in the absence of uracil, and controlled for FNT and MCT1 expression by Western blot (S6 Fig). The proteins were detected using a mouse monoclonal anti-hemagglutinin antibody (Roche), a horseradish peroxidase-conjugated secondary antibody (Jackson Immuno Research), and the ECL Plus system (GE Healthcare) for documentation (Lumi-Imager F1, Roche).
The assays were carried out as described earlier [7]. Briefly, yeast cultures were harvested at an OD600 of 0.8, resuspended in 50 mM HEPES/Tris, pH 6.8 ± 0.1, to an OD600 of 50 (± 10%), and kept on ice. Transport and inhibition was tested at 18°C in 1.5 ml reaction tubes using 80 μl yeast suspension supplemented with 1 μl of the inhibitor solution in DMSO. The final DMSO concentration was 1.25% and was also added to uninhibited control yeast. Transport was initiated after 20 min by adding 20 μl of substrate solution to yield a final concentration of 1 mM substrate and 0.04 μCi (FNTs) or 0.08 μCi (MCT1) [1-14C]-l-lactate or [1-14C]-formate (malaria box screening). The specific activity of the radiolabels was 55 mCi mmol–1 (Hartmann Analytic). The reaction was stopped by abrupt dilution with 1 ml ice-cold water, rapid transfer of the suspension onto a vacuum filtration unit fitted with a GF/C filter membrane (Whatman), and washing with 7 ml water. The filter membranes were transferred to scintillation vials containing 3 ml of scintillation cocktail (Quicksafe A, Zinsser Analytic) and analyzed using a Packard TriCarb liquid scintillation counter (Perkin Elmer Inc.). For screening of the malaria box, two replicates of each of the 400 compounds were assayed at 10 μM for 30 s; a compound was considered a hit when transport was fully blocked. Binding kinetics were analyzed by pre-incubating PfFNT expressing yeast cells with 0.17 μM, 0.4 μM, 1 μM, 4 μM, and 10 μM of MMV007839 for defined time points between 30 s and 20 min. For dissociation of MMV007839 from PfFNT, yeast was incubated with 1 μM and 0.17 μM inhibitor until maximal or half-maximal inhibition was reached. Then, MMV007839 was quantitatively removed by two consecutive washing steps with inhibitor-free buffer, the cells were kept at 18°C, and l-Lactate transport was monitored up to 2 h after removal of the inhibitor.
P. falciparum parasites strain 3D7 were cultured in 5% human 0+ erythrocytes according to standard conditions [30] in RPMI 1640 medium containing 0.5% albumax. For the selection of resistant parasites, culture flasks containing 50 ml of parasite culture (starting parasitemia of 2% rings) were subjected to 3 × IC50 of the respective drug and the parasites were fed daily with medium containing the drug until parasites disappeared. Thereafter medium was changed every 48 h under continued drug pressure until parasites were once more detected. DNA was isolated using the QIAamp DNA Mini Kit (Qiagen). Sequencing of PCR amplified PfFNT was carried out by SeqLab (Göttingen).
IC50 values were determined using serial 1:2 drug dilutions and a control without drug but containing DMSO in 2 ml P. falciparum culture volumes in 2 × 12 well dishes. Cultures were fed 24 h later and fresh drug was added. After another 24 h the parasitemia was determined using a LSR II FACS (BD Biosciences).
The malaria box was obtained from Medicines for Malaria Ventures (www.mmv.org). MMV007839 and MMV000972 and compounds [2]–[4]; [6]–[9], and [11] were from Vitas-M Laboratory; [1] was from Chembridge and the pharmacophore (5,5,6,6,6-pentafluoro-4-hydroxyhex-3-en-2-one) from Manchester Organics (for CAS numbers see S3 Table). [12]–[15] were synthesized by Claisen-type condensation of the corresponding 4-substituted 2-hydroxyacetophenones with ethyl pentafluoropropanoate in anhydrous THF in the presence of lithium hydride [31]. For [5] the phenone component was 2-hydroxy-4-methoxypropiophenone, and for [10] the fluoroalkyl component was ethyl heptafluorobutanoate. For BH-296, 4-methoxyacetophenone was used as the phenone component. BH-296 was further alkylated with ethyl p-toluenesulfonate in the presence of caesium carbonate in DMF to yield the ethyl ester prodrug [32]. All synthesized compounds were purified by re-crystallization or liquid chromatography and verified by mass spectrometry (LC-MS; Bruker Amazon SL) and nuclear magnetic resonance (Bruker Avance III 300; for 1H-NMR data, see S4 Table). For correlation NMR of MMV007839 in CDCl3, a 1H / 13C heteronuclear single quantum correlation spectrum (HSQC) was generated.
For IC50 determinations, at least triplicate time points in the initial linear phase of the transport curves (Fig 6a [PfFNT], S7 Fig [BbFNT, MCT1]) were used: 30 s (PfFNT, BbFNT), 60 s (BbFNT S93G), 120 s (PfFNT G107S), and 180 s (MCT1). For the necessary accuracy, data points required technical replicates depending on the degree of inhibition: n = 3 for inhibitory concentrations leading to less than 10% activity, n = 6 in the intermediate range around the IC50, and n = 9 for remaining transport activities above 80%. All IC50 values were controlled in at least three independent experiments. A dashed line indicating the inhibitory effect of MMV007839 on PfFNT in yeast is shown as an averaged reference in various figures. The respective data points (110 measurements) were obtained from control experiments throughout the study with MMV007839 compound from three independent sources, i.e. the malaria box, a commercial vendor (Vitas-M Laboratory), and from chemical synthesis in our own laboratory. This curve and the one shown in Fig 1C yielded identical IC50 values and error margins. The Ki value is derived from five independently determined inhibition rates with 0.17, 0.4, 1.0, 4.0, and 10.0 μM MMV007839 compound. Sigmoidal Hill, linear, and exponential curve fittings were done using SigmaPlot (Systat Software). Error bars denote S.E.M.
|
10.1371/journal.pntd.0005472 | Trypanocidal Effect of Isotretinoin through the Inhibition of Polyamine and Amino Acid Transporters in Trypanosoma cruzi | Polyamines are essential compounds to all living organisms and in the specific case of Trypanosoma cruzi, the causative agent of Chagas disease, they are exclusively obtained through transport processes since this parasite is auxotrophic for polyamines. Previous works reported that retinol acetate inhibits Leishmania growth and decreases its intracellular polyamine concentration. The present work describes a combined strategy of drug repositioning by virtual screening followed by in vitro assays to find drugs able to inhibit TcPAT12, the only polyamine transporter described in T. cruzi. After a screening of 3000 FDA-approved drugs, 7 retinoids with medical use were retrieved and used for molecular docking assays with TcPAT12. From the docked molecules, isotretinoin, a well-known drug used for acne treatment, showed the best interaction score with TcPAT12 and was selected for further in vitro studies. Isotretinoin inhibited the polyamine transport, as well as other amino acid transporters from the same protein family (TcAAAP), with calculated IC50 values in the range of 4.6–10.3 μM. It also showed a strong inhibition of trypomastigote burst from infected cells, with calculated IC50 of 130 nM (SI = 920) being significantly less effective on the epimastigote stage (IC50 = 30.6 μM). The effect of isotretinoin on the parasites plasma membrane permeability and on mammalian cell viability was tested, and no change was observed. Autophagosomes and apoptotic bodies were detected as part of the mechanisms of isotretinoin-induced death indicating that the inhibition of transporters by isotretinoin causes nutrient starvation that triggers autophagic and apoptotic processes. In conclusion, isotretinoin is a promising trypanocidal drug since it is a multi-target inhibitor of essential metabolites transporters, in addition to being an FDA-approved drug largely used in humans, which could reduce significantly the requirements for its possible application in the treatment of Chagas disease.
| Polyamines are polycationic compounds essential for the regulation of cell growth and differentiation. In contrast with other protozoa, Trypanosoma cruzi, the etiological agent of Chagas disease, is auxotrophic for polyamines; therefore the intracellular availability of these molecules depends exclusively on transport processes. It was previously demonstrated that the lack of polyamines in T. cruzi leads to its death, making the polyamine transporter an excellent therapeutic target for Chagas disease. In this work, the polyamine permease TcPAT12 was selected as a target for drug screening using 3000 FDA-approved compounds and computational simulation techniques. Using two combined virtual screening methods, isotretinoin, a well-known and safe drug used for acne treatment, bound to substrate recognition residues of TcPAT12 and was chosen for further in vitro studies. Isotretinoin inhibited not only the polyamine transport but also all tested amino acid transporters from the same protein family as TcPAT12. Interestingly, isotretinoin showed a high trypanocidal effect on trypomastigotes, with an IC50 in the nanomolar range. Autophagy and apoptosis were proposed as mechanisms of parasites death induced by isotretinoin. These results suggest that isotretinoin is a promising trypanocidal drug, being a multi-target inhibitor of essential metabolites transporters.
| Chagas disease is a major health and economic problem in the Americas and its causative agent is the hemoflagellate Trypanosoma cruzi [1]. According to the World Health Organization, about 8 million people worldwide are infected with the parasite, and 10,000 people per year die from complications linked to Chagas disease, mostly in Latin America where the disease is endemic [2]. In addition, the chronicity of the pathology implies great health expenditures due to the disability associated with the chronic state of this infection, being heart failure the main disabling condition [3]. Only two drugs are approved for treating Chagas disease, the nitroimidazole benznidazole and the nitrofuran nifurtimox, which were discovered half a century ago and have very limited efficacy with severe side effects [4,5]. This highlights the need for the development of new therapeutic alternatives and the identification of novel drug targets.
Since transport of nutrients from the extracellular medium is inexpensive in terms of energy economy compared to their metabolic synthesis, the uptake is a very common and desirable strategy for parasitic organisms. T. cruzi is exposed to different environments along its life cycle, alternating between the gut of the insect vector, the bloodstream of the mammalian hosts, and within different cell types [6], and the availability of nutrients in these dissimilar milieus determines the need for complex metabolic adaptations. The first and probably the only multigenic family of amino acid transporters in T. cruzi (TcAAAP) was identified by our group [7]. One interesting feature of these permeases is the absence of orthologs in mammalian genomes. Few members of this family have been characterized in trypanosomatids, including polyamines, arginine, proline and lysine permeases [8,9,10,11,12]. This T. cruzi transporter family comprises at least 36 genes coding for proteins with lengths of 400–500 amino acids and 10–12 predicted transmembrane α-helical spanners. Another remarkable feature of these proteins is the variability of the N-terminal domain (about 90 amino acids with only 5% of consensus positions), in contrast to the central and C-terminal domains, which have a very similar sequence [7,9]. In Leishmania spp. it was demonstrated that only the variable N-terminal domain is involved in the substrate specificity [13]. On the contrary, mutagenesis analysis in T. cruzi locates the substrate recognition site of the polyamine transporter outside the N-terminal variable region [14]. The high sequence similarity could be an advantage for the development of multi-target inhibitors against the TcAAAP transporter family. The supply of essential polyamines in T. cruzi is exclusively achieved through transport processes, a clear case of metabolic-transport replacement in the evolutionary adaptation to parasitism [15]. Interestingly, TcPAT12 (also known as TcPOT1) is the only polyamine transporter present in T. cruzi. Recent studies have shown that the trypanocidal drug pentamidine blocks TcPAT12 in this parasite [16,17]. All this evidence highlights that T. cruzi nutrient transporters are promising targets for drug development. One interesting approach is the use of molecular docking to identify pharmacological active compounds among drugs already used for other therapeutic indications (called “drug repositioning” or “drug repurposing) [18]. For example, the discovery of polyamine analogs, by computational simulation, with inhibitory effects on the proliferation of T. cruzi has been recently reported [19].
Retinol (vitamin A, all-trans-retinol) and its derivatives play an essential role in metabolic functioning of the retina, the growth and differentiation of epithelial tissue, bone growth, reproduction, and immune response. Dietary retinol is derived from a variety of carotenoids found in plants, liver, egg yolks, and the fat component of dairy products. This compound activates retinoic acid receptors (RARs), inducing cell differentiation and apoptosis of some cancer cell types and inhibiting carcinogenesis [20,21,22,23,24]. Isotretinoin (13-cis-retinoic acid) is a retinol derivative used in the treatment of severe acne and some types of cancer [25,26]. The usage dose is 0.5–1 mg.kg-1 [27] and its most common side effects are skin xerosis, especially on exposed skin, cheilitis, telogen effluvium, inflammatory bowel disease and myalgia [28]. Despite its exact mechanism of action remains unknown, several studies have shown that this drug induces apoptosis in sebaceous gland cells. Isotretinoin has a low affinity for RARs and retinoid X receptors (RXR), but it may be intracellularly converted to metabolites that act as agonists of these nuclear receptors [29]. Previous data reported that butylated hydroxyanisole, retinoic acid and retinol acetate dramatically inhibit the growth of Leishmania donovani promastigotes, and retinol acetate also decreases by half the intracellular polyamine levels [30]. Furthermore, isotretinoin alters the life cycle of the protozoan parasite Opalina ranarum in frogs, inhibiting the induction of cyst formation [31]. Considering the effects of some retinoids in protozoan organisms, in this work we evaluated by virtual screening and in vitro assays different retinoids used in medical practice. We demonstrated that isotretinoin has trypanocidal effect through the specific inhibition of permeases from TcAAAP family. As a repositioned drug, isotretinoin has many advantages over developing new drugs because of its oral bioavailability, low cost and current use for treating other diseases.
Computational approaches for the identification of putative TcPAT12 inhibitors started with a ligand-based virtual similarity screening search followed by molecular docking, which is a receptor-based technique. Retinol acetate was used as the reference compound for similarity searching in a database that comprises a total of 2924 worldwide commercially available drugs and nutraceuticals approved by U.S. Food and Drug Administration (FDA). This screening was performed using LiSiCA v1.0 (Ligand Similarity using Clique Algorithm) software [32], and similarities were expressed using the Tanimoto coefficient [33]. The structural data of compounds retrieved from similarity screening, as well as putrescine and spermidine, TcPAT12 natural ligands. Selected molecules for molecular docking were obtained from a subset of the ZINC database (http://zinc.docking.org/). The compounds analyzed and their corresponding ZINC IDs were: acitretin (3798734), alitretinoin (12661824), etretinate (3830820), isotretinoin (3792789), putrescine (1532552), retinal (4228262), retinol (3831417), retinoic acid (12358651), retinol acetate (26892410), and spermidine (1532612). Further preparation of the PDBQT files (Protein Data Bank, partial charge (Q), and atom type (T)) was performed using AutoDock Tools v1.5.6 [34]. Three-dimensional structure of TcPAT12 (GenBank ID: AY526253) was obtained by homology modeling, with as template the Escherichia coli amino acid antiporter (AdiC; PDB ID: 3L1L; about 30% amino acid identity with TcPAT12) using the Swiss-Model server (http://swissmodel.expasy.org/) [35]. The model was refined using the software Modeller v7 and the previously reported model of the TcPAT12 [14,36]. The obtained structure was evaluated by Ramachandran plotting using Chimera v1.8 [37,38]. From this model, residues Asn245, Tyr148 and Tyr400 were taken as flexible using AutoDock Tools 1.5.6. The grid parameter file was generated with Autogrid 4.2.6 so as to surround the flexible residues, with a grid map of 40 points in each dimension, a spacing of 0.0375 nm, and centered on position X = -2.794, Y = 9.659, and Z = 22.928. An additional docking assay was performed using a grid covering the whole transporter molecule, without defining specific flexible residues and using the same spacing and automatic centering. AutoDock 4.2.6 was used for calculation of optimal energy conformations for the ligands interacting with the protein active site, running the Lamarckian Genetic Algorithm 100 times for each ligand, with a population size of 300, and 2.7x104 as the maximum number of generations. For each ligand, bound conformations were clustered and two criteria for selection of the preferred binding conformation were followed: taking the lowest free binding energy conformation of all the poses, and from the most populated cluster [39]. A diagram of the virtual screening workflow is shown in Fig 1.
Epimastigotes of T. cruzi Y strain (5x106 cells.mL-1) were cultured at 28°C in plastic flasks (25 cm2), containing 5 mL of BHT (brain-heart infusion-tryptose) medium supplemented with 10% fetal calf serum, 100 U.mL-1 penicillin, 100 μg.mL-1 streptomycin and 20 μg.mL-1hemin [40]. CHO-K1 cells (Chinese Hamster Ovary) were cultured in RPMI medium supplemented with 10% heat-inactivated Fetal Calf Serum (FCS), 0.15% (w/v) NaCO3, 100 U/mL penicillin and 100 U/mL streptomycin at 37°C in 5% CO2. Trypomastigotes were obtained from the extracellular medium of CHO-K1 infected cells as previously described [41].
Aliquots of epimastigote cultures (107 parasites) were centrifuged at 8,000 xg for 30 s and washed once with phosphate-buffered saline (PBS). Cells were resuspended in 0.1 mL PBS and then 0.1 mL of the transport mixture containing the corresponding radiolabeled substrate was added: [3H]-putrescine, [3H]-proline, [3H]-lysine, [3H]-amino acids mixture (Ala, Arg, Asp, Glu, Gly, His, Ile, Leu, Lys, Phe, Pro, Ser, Thr, Tyr and Val), [3H]-thymidine or [14C]-glucose (PerkinElmer's NEN Radiochemicals; 0.4 μCi). Parasites were pre-incubated for 15 min with concentrations of isotretinoin between 0–50 μM for all molecules, except for putrescine in which case 0–100 μM were used. Following incubation at 28°C, the transport reaction was stopped by adding 1 mL of ice-cold PBS. Cells were centrifuged as indicated above and washed twice with ice-cold PBS. Cell pellets were resuspended in 0.2 mL of water and counted for radioactivity in UltimaGold XR liquid scintillation cocktail (Packard Instrument Co., Meridien CT, USA) [42,43]. Non-specific binding and carry over were evaluated by a standard transport assay supplemented with a 100-fold molar excess of the corresponding substrate.
Exponentially growing T. cruzi epimastigotes were cultured as described above, in 24-wells plate at a start density of 107 cells.mL-1 in BHT medium. Parasites growth was evaluated at different concentrations of isotretinoin, in the range of 0–300 μM, and parasite proliferation was determined after 72 h. Inhibition of trypomastigote bursting from infected cells was performed using CHO-K1 cells (5x104 per well) infected with trypomastigotes (2.5×106 per well) for 4 h. After this period, the infected cells were washed twice with PBS, the RPMI medium was replaced, and the cells were kept in culture in the presence of different concentrations of isotretinoin (0–30 μM) for 24 h. After infection, plates were incubated at 33°C and parasites were collected from the extracellular medium on the sixth day. Cells were counted using a Neubauer chamber using a blinded design or by viability assays using “Cell Titer 96 Aqueous One Solution Cell Proliferation Assay (MTS)” (Promega, Madison, WI, USA) according to the manufacturer instructions.
In order to test if isotretinoin exerts cell permeabilization, epimastigote cells (5x108) were washed twice and resuspended in PBS. Aliquots of 100 μl containing 108 parasites were mixed with 100 μl of the same buffer containing increasing amounts of isotretinoin (0, 5, 25 and 100 μM). After 30 min of incubation at room temperature in the presence of isotretinoin, the tubes were centrifuged at 16100 xg for 2 min. Supernatants were kept on ice and pellets were resuspended in the same buffer. Permeabilization of epimastigotes with digitonin was used as a positive control; cells were washed twice and resuspended in 50 mM Tris-HCl buffer, pH 7.5, containing 0.25 M sucrose and 10 μM E64. Aliquots of 100 μl containing 108 parasites were mixed with 100 μl of the same buffer containing 0.3 mg.mL-1 of digitonin. After 2.5 min of incubation at room temperature, the tubes were centrifuged at 16100 ×g for 2 min. Supernatants were transferred to new tubes and pellets were resuspended in the same buffer. All supernatant and pellet fractions were analyzed by Western blot. Briefly, samples were run on 15% SDS-polyacrylamide gels (PAGE) and transferred onto a PVDF membrane. The membranes were blocked and incubated with primary rabbit antibodies anti-glutamate dehydrogenase (1:5000 dilution) followed by incubation with peroxidase-conjugated anti-rabbit (1:5000 dilution). The peroxidase-labeled proteins were revealed using Super Signal West Pico Chemiluminescent substrate following the manufacturer instructions (Pierce, Waltham, MA, USA).
For apoptosis analysis by TUNEL (Terminal deoxynucleotidyl transferase dUTP nick end labeling), parasites (107) were treated with the corresponding concentrations of isotretinoin and, after letting the cells settle for 20 min onto poly-L-lysine coated coverslips, were fixed for 20 min with 4% paraformaldehyde in PBS and permeabilized with 0.1% Triton X-100. Assays were performed using the “In situ cell death detection Kit” (Roche) according to the manufacturer instructions. Positive and negative controls were made using DNAse I and untreated parasites, respectively. Slides were mounted using Vectashield with DAPI (Vector Laboratories) and cells were observed under an Olympus BX60 fluorescence microscope. Images were recorded with an Olympus XM10 camera. To detect phosphatidylserine, annexin V binding on the external surface of the plasma membrane of treated and untreated parasites was evaluated using the “Annexin V: FITC Apoptosis Detection Kit” (Sigma-Aldrich) according to the manufacturer’s protocol. Co-staining of the parasites with propidium iodide was performed to evaluate the integrity of plasma membrane during the treatments. Fluorescence was detected in FACSCalibur equipment (Becton Dickinson & Co., NJ, USA). Data was analyzed using Cyflogic software. [44]. Autophagy was evaluated using monodansylcadaverine (MDC) labeling [45]. Briefly, after isotretinoin treatment, parasites were incubated with 0.05 mM MDC in PBS at 37°C for 15 min and washed twice in PBS. MDC stain was evaluated using a fluorescence microscope Olympus BX60 and images were captured with an Olympus XM10 digital camera. To evaluate the formation of autophagic structures by indirect immunoflurescence microscopy, epimastigote samples were collected, washed twice with PBS, and settled for 20 min at room temperature onto poly L-lysine coated coverslips. Parasites were then fixed at room temperature for 20 min with 4% formaldehyde in PBS, permeabilized with cold methanol for 5 min, and rehydrated in PBS for 15 min. The samples were blocked with 1% BSA in PBS for 10 min and incubated with the primary antibody in blocking buffer (rabbit anti-Atg8.1 polyclonal, 1:250 dilution) for 2 h. The antibody was kindly provided by Dra. Vanina E. Alvarez from the “Instituto de Investigaciones Biotecnológicas” (IIB-INTECH). After three washes, the parasites were incubated with anti-rabbit antibodies tagged with FITC, at a dilution of 1:500 for 30 min, washed and mounted using Vectashield with DAPI (Vector Laboratories). Cells were observed in an Olympus BX60 fluorescence microscope and recorded with an Olympus XM10 camera. To detect apoptotic bodies, cultures were stained with acridine orange and ethidium bromide. The morphology of death and surviving cells were observed by fluorescence microscopy. Ethidium bromide only enters into non-viable cells and stains chromatin and apoptotic bodies with an orange color. Acridine orange penetrates in viable cells and turns green when it intercalates with DNA [46].
All the experiments were made at least in triplicates and results presented here are representative of three independent assays. IC50 values were obtained by non-linear regression of dose-response logistic functions, using GraphPad Prism 6.01.
Two consecutive virtual screening techniques were applied to find out putative TcPAT12 inhibitors. As mentioned, previous data demonstrated that retinol acetate has toxic effects on Leishmania parasites by diminishing the intracellular polyamine concentrations [30]. According to these results, this molecule was used as a reference compound for virtual screening. In order to get more ligands to test besides retinol acetate, the first approach was a ligand-based virtual screening using a database containing 2924 FDA approved drugs and nutraceuticals. Seven therapeutic retinoids [47] were obtained from the first step of virtual screening and were used to construct a similarity matrix (S1 Fig) and a dendrogram based on the Tanimoto coefficient, using the Unweighted Pair Group Method with Arithmetic mean (UPGMA) algorithm [33,48]. The similarity graphic discriminates between clusters containing the different generations of retinoids which are structurally unrelated to the natural substrates of TcPAT12 (Fig 2).
The second step was a receptor-based strategy using molecular docking simulations. The three-dimensional structure of TcPAT12 was modeled using as a template the Escherichia coli amino acid antiporter, and refined with experimentally validated data about the putrescine binding site [14]. Since evaluation methods for homology models quality were made based on the data available on the PDB, they are biased towards globular proteins, and cannot be used for membrane proteins [49]. For that reason, the quality of the TcPAT12 model was evaluated by checking torsion angles of the peptide backbone in a Ramachandran Plot, one of the most powerful tools used to determinate the quality of a model coordinates [37,50] (S2 Fig). Results showed that the obtained model has only 4.3% of the torsions in the outlier regions of the plot, and none of those residues were involved in the active site of the transporter. Given that 91% of the experimental structures deposited in the PDB have 10% or less residues in the outlier region, and only 76.5% possess less than 5% of outliers, the generated model for TcPAT12 can be considered to have a reasonable quality [51]. The ability of the selected retinoids to interact with TcPAT12 substrate binding residues was tested by a computer-assisted simulation with AutoDock 4.0, using the natural substrates of TcPAT12 as binding parameter references (Table 1). For each compound, two criteria were used to analyze docking results: the lowest global binding score, and the lowest binding score from the most populated cluster. Isotretinoin had the lowest binding energy values for both sorting criteria. These results, together with its availability and market price, point it out as the best candidate for further analysis. According to docking models, isotretinoin binds within the hydrophobic channel of the transporter, in the previously reported putrescine-binding pocket, interacting with residues Asn245, Tyr148 and Tyr400 (Fig 3A and 3B).
Isotretinoin possess a ligand efficiency of -0.37 kcal.mol-1 for its interaction with the polyamine binding site of TcPAT12, and the cluster with the lowest global free binding score (-10.78 kcal.mol-1) was also the one with more conformations, with 35 of the 100 generated poses. Interestingly, when docking a simulation between TcPAT12 and isotretinoin was performed over the whole transporter molecule, without limiting the region to be tested, similar results were obtained; isotretinoin also bound residues Asn245, Tyr148, and Tyr400. Both results suggest that isotretinoin binds more stably in that region of TcPAT12 than its natural ligands. All these data are summarized in Table 1.
In order to validate the results obtained by virtual screening, the ability of isotretinoin to inhibit putrescine uptake through TcPAT12 was evaluated. Putrescine concentration was fixed in 100 μM, about 10-fold its Km value [52], and transport assays were performed in the presence of different concentrations of isotretinoin in the range of 0–100 μM. Results shown in Fig 4A confirmed that low concentrations of this drug produce a significant inhibition of putrescine transport. The calculated isotretinoin concentration that inhibited 50% of the putrescine transport (IC50) was 4.6 μM. To evaluate if the putrescine transport could be inhibited by other retinoids that scored promising ΔG values in the docking assays, the experiments were repeated with acitretin (ΔG = -6.70 kcal.mol-1), a drug used for psoriasis treatment, and the isotretinoin precursor retinol (ΔG = -8.69 kcal.mol-1). The IC50 of acitretin was 6.8 μM while retinol had no effect on putrescine transport in the tested concentrations.
The amino acid and polyamine transporters of the TcAAAP family are very similar in terms of amino acid sequences [7]. For this reason, the inhibitory effect of isotretinoin on other transporters from the same family, and also over structurally unrelated permeases, was evaluated. IC50 values were calculated with the same criteria used for putrescine transport; about 10-fold the Km concentration of each compound, in the presence of isotretinoin from 0 to 50 μM. The assayed substrates of TcAAAP transporters were proline, lysine, and an amino acid mix [9,12], while thymidine and glucose incorporation was evaluated for effect of isotretinoin on unrelated permeases [53]. Calculated IC50 values for proline, lysine and the amino acid mix were 10.3 μM, 5.1 μM and 5.8 μM, respectively. On the other hand, isotretinoin produced no significant inhibition on thymidine and glucose transport demonstrating its specificity for members of the TcAAAP family.
With the aim of analyzing if the inhibition of putrescine uptake could affect the parasites viability, isotretinoin toxicity over trypomastigotes and epimastigotes was assessed. The effect of isotretinoin in trypomastigotes, the mammal infective form of T. cruzi, was evaluated using a model of in vitro infection in CHO-K1 cells. Infected cells were exposed to isotretinoin for 24 h in the concentration range of 0–30 μM. At very low concentrations, isotretinoin inhibited the trypomastigotes burst after six days of infection, with a calculated IC50 of 130 nM (Fig 4B). Remarkably, this IC50 value is significantly lower than those obtained for the drugs currently used as a treatment for Chagas disease [12,16]. The calculated selectivity index of isotretinoin against trypomastigotes over human macrophages was about 920. In addition, the infection index (infected cells x total amastigotes per cell) was calculated for infected cells treated with isotretinoin from 65 to 260 nM. For control cells it was 6.44 (±1.55) and for cells treated with 65, 130 and 260 nM were 6.49 (±1.98), 4.13 (±1.16) and 3.61 (±0.94), respectively. Epimastigotes, the insect stage of T. cruzi, were also treated with different concentrations of isotretinoin (0–300 μM) for 72 h. As Fig 4C shows, isotretinoin was also effective as growth inhibitor of cultured epimastigotes, but at concentrations 230-fold higher than in trypomastigotes (IC50 = 30.6 μM).
To evaluate the cytotoxicity of isotretinoin on CHO-K1 cells and peripheral blood monocytes, cells were exposed to isotretinoin for 24 h in a concentration range from 0 to 30 μM, and no effect was observed at any of these drug concentrations.
In order to test if the observed inhibition of parasites growth was mediated by a programmed cell death mechanism, apoptosis analysis in epimastigote cells was performed. The first approach was to evaluate the exposition of phosphatidylserine (annexin V) and propidium iodide exclusion by flow cytometry. As S3 Fig shows both cell death markers were negative in epimastigotes exposed to isotretinoin from 30 to 120 μM for 72 h. The second technique used was the TUNEL assay. Epimastigote cells treated with 30 μM isotretinoin for 72 h presented a TUNEL negative staining (Fig 5). In order to evaluate apoptosis using a short-time isotretinoin treatment, the IC50 for epimastigotes was calculated at 6 h post-treatment, with a value of 214 μM. At 200 μM isotretinoin, the percentage of TUNEL positive cells decreased to 64.9% (± 0.03) showing a complete change in cell morphology, partially or fully rounded cells. Negative and positive controls were also assayed, using untreated cells and cells treated with DNAse I, respectively. Under these conditions the exposition of phosphatidylserine and propidium iodide were also evaluated. Flow cytometry analysis showed that 21.3% of the parasites treated with 200 μM drug were positive for annexin V and all the cells population was negative to propidium iodide. These results suggest that parasites entered only in apoptosis, no necrosis process was observed (S3 Fig, lower panel). In addition, to assess whether an autophagic component is involved in cell death induced by isotretinoin treatment, parasites were evaluated using MDC, a fluorescent probe that accumulates in autophagic vacuoles [45]. Parasites treated for 6 h with 200 μM isotretinoin presented rounded structures stained by MDC, (Fig 6A). To validate the formation of autophagic structures, subcellular localization of TcAtg8.1 protein, an autophagosomal membrane marker [54], was evaluated by indirect immunofluorescence microscopy. Autophagic vacuoles were detected in parasites also treated for 6 h with 200 μM isotretinoin, indicating that autophagic processes were triggered by this drug (Fig 6B). Finally, when parasites under the same treatment conditions were stained with ethidium bromide and acridine orange, apoptotic bodies were detected (Fig 6A).
To determine if the trypanocidal effect of isotretinoin was due to an increased permeability of the plasma membrane, permeabilization experiments with this drug were carried out using the non-ionic detergent digitonin as a positive control (S4 Fig). The pattern of extraction of glutamate dehydrogenase, which localizes in the parasites cytosol, was used as a membrane stability marker [55]. At isotretinoin concentrations up to 100 μM, Western Blot analysis showed that glutamate dehydrogenase was completely absent in the parasites supernatant demonstrating that the structure of the plasma membrane remained unaltered at these drug concentrations. Positive control experiments using digitonin 0.3 mg.mL-1 showed that the cytosolic marker had been totally extracted. The plasma membrane integrity was also evaluated by propidium iodide exclusion using flow cytometry. Results of treatments up to 120 μM for 72 h and 200 μM for 6 h, suggest that the plasma membrane remains unaltered after isotretinoin exposure (S3 Fig).
Two of the most promising alternatives related to the Chagas disease therapy were the use of benznidazole in the chronic phase of the disease (BENEFIT) and the implementation of posaconazole as a novel trypanocidal drug. Unfortunately, after a recent evaluation of their effectiveness, none of these alternatives was successful [56,57] highlighting the urgent need for the development of new therapeutic substitutes for conventional treatments. In this context, drug repositioning is a rapid way to obtain compounds with new desired biological activity from drugs already approved for human use, smoothing the path for quickly reaching the counters [58]. Within in silico strategies for drugs identification, the combination of different virtual screening techniques significantly enhances the possibility to succeed in subsequent in vitro and in vivo studies [18]. In this work the strategy used was a combination of a ligand-based virtual screening by similarity followed by a receptor-based technique (molecular docking). The T. cruzi polyamine permease TcPAT12 is a promising therapeutic target for rational drug design since this parasite is the only trypanosomatid unable to synthesize polyamines de novo, which makes it dependent on transport processes [59]. Besides, as reported so far, this is the only polyamine transporter present in T. cruzi.
Isotretinoin was selected on the basis of its low predicted free binding energy and for being a very common, low-cost compound. Isotretinoin is a retinol derivative, naturally found in small quantities in the human body and mainly used in the treatment of severe acne [60]. Another advantage is its market price of about USD 300–400 per kilogram which would make it easily accessible for developing countries. The calculated lowest free binding energy of isotretinoin to TcPAT12 substrate recognition site was -10.78 kcal.mol-1. This value is similar to the obtained using the specific human polyamine transport blocker AMXT-1501 (-14.01 kcal.mol-1) [61] and lower than those of the natural substrates, putrescine (-3.31 kcal.mol-1) and spermidine (-3.08 kcal.mol-1). These data suggest that the stability of the isotretinoin—transporter complex is higher than those formed with its natural substrates probably because of the greater number of atoms (23) capable of engage in molecular interactions. When three of the molecules predicted to bind to the transporter were tested, isotretinoin and acitretin produced a strong inhibition of putrescine transport, while retinol had no effect. This is due to the predictability achievable by AutoDock simulations, which because of the completely theoretical nature of their scoring function, when compared with experimental models of other ligand-membrane protein interactions, have demonstrated 62% chances of identifying active compounds, while 44% chances of misidentifying an inactive compound as an active one [62]. Isotretinoin inhibition of transport correlated with its trypanocidal activity in epimastigotes. The IC50 calculated for trypomastigotes was in the nanomolar order and about 230-fold higher than the observed in epimastigotes, and also had a value similar to that of a new proteasome inhibitor tested by Novartis for treating T. cruzi infection in mice [63]. When compared, the effect of the drug in trypomastigotes was almost three orders of magnitude higher than its effect in human macrophages (selectivity index). These results are important since only the mammalian T. cruzi stages are relevant from a therapeutic perspective. In addition, the concentration at which isotretinoin acts in vitro is one order of magnitude lower than those reported for the drugs currently used for the treatment of Chagas disease; benznidazole and nifurtimox [12,16]. Isotretinoin is a good candidate for the treatment of Chagas disease since it does not require to be chemically modified. This feature is relevant since after any chemical modification it will be considered as a new drug and not as a repositioned one, with the consequent expensive trials required for its approval. Once the effect of isotretinoin over TcPAT12 was validated, inhibition of other amino acid transporters from the same family was tested. Remarkably, isotretinoin inhibited all the assessed transporters from TcAAAP family and no effect was observed over structurally unrelated proteins such as hexoses and nucleosides transporters. This specificity could be explained by the high structural similarity between all members of polyamines and amino acids transporters of the TcAAAP family.
Autophagy is a mechanism by which cells under starvation digest their own components to provide amino acids that may function as an energy source [44] and this process was reported in trypanosomatids more than 10 years ago [64]. Considering that isotretinoin inhibited polyamine and also amino acid transporters, the consequent nutrient starvation would initiate autophagic process that might not recover the cells and thus the programmed cell death by apoptosis might be triggered. This is particularly relevant in the case of T. cruzi since epimastigotes use amino acids as the main carbon and energy sources alternative to glucose, as well as the use of amino acids in stage differentiation, host cells invasion, stress resistance, cell energy management, among others [12,65,66,67]. Isotretinoin acts as a multi-target inhibitor of transporter proteins from the TcAAAP family, improving its trypanocidal potential as well as diminishing the possibility of generating drug resistance in the parasite. In addition, isotretinoin probably acts on the transporters in the external side of the plasma membrane, avoiding one of the most common problems of drugs, such as the way of entry into the cells.
Summarizing, isotretinoin is a promising trypanocidal drug because it has activity in the nanomolar range of concentrations, it is a multi-target inhibitor of essential metabolites transporters, it is already approved by the FDA and also it is a drug largely used in humans, which significantly reduces the requirements for its application in therapy for Chagas disease.
|
10.1371/journal.pcbi.1000464 | Recognizing Sequences of Sequences | The brain's decoding of fast sensory streams is currently impossible to emulate, even approximately, with artificial agents. For example, robust speech recognition is relatively easy for humans but exceptionally difficult for artificial speech recognition systems. In this paper, we propose that recognition can be simplified with an internal model of how sensory input is generated, when formulated in a Bayesian framework. We show that a plausible candidate for an internal or generative model is a hierarchy of ‘stable heteroclinic channels’. This model describes continuous dynamics in the environment as a hierarchy of sequences, where slower sequences cause faster sequences. Under this model, online recognition corresponds to the dynamic decoding of causal sequences, giving a representation of the environment with predictive power on several timescales. We illustrate the ensuing decoding or recognition scheme using synthetic sequences of syllables, where syllables are sequences of phonemes and phonemes are sequences of sound-wave modulations. By presenting anomalous stimuli, we find that the resulting recognition dynamics disclose inference at multiple time scales and are reminiscent of neuronal dynamics seen in the real brain.
| Despite tremendous advances in neuroscience, we cannot yet build machines that recognize the world as effortlessly as we do. One reason might be that there are computational approaches to recognition that have not yet been exploited. Here, we demonstrate that the ability to recognize temporal sequences might play an important part. We show that an artificial decoding device can extract natural speech sounds from sound waves if speech is generated as dynamic and transient sequences of sequences. In principle, this means that artificial recognition can be implemented robustly and online using dynamic systems theory and Bayesian inference.
| Many aspects of our sensory environment can be described as dynamic sequences. For example, in the auditory domain, speech and music are sequences of sound-waves [1],[2], where speech can be described as a sequence of phonemes. Similarly, in the visual domain, speaking generates sequences of facial cues with biological motion [3],[4]. These auditory and visual sequences have an important characteristic: the transitions between the elements are continuous; i.e., it is often impossible to identify a temporal boundary between two consecutive elements. For example, phonemes (speech sounds) in a syllable are not discrete entities that follow each other like beads on a string but rather show graded transitions to the next phoneme. These transitions make artificial speech recognition notoriously difficult [5]. Similarly, in the visual domain, when we observe someone speaking, it is extremely difficult to determine exactly where the movements related to a phoneme start or finish. These dynamic sequences, with brief transitions periods between elements, are an inherent part of our environment, because sensory input is often generated by the fluent and continuous movements of other people, or indeed oneself.
Dynamic sequences are generated on various time-scales. For example, in speech, formants form phonemes and phonemes form syllables. Sequences, which exist at different time-scales, are often structured hierarchically, where sequence elements on one time-scale constrain the expression of sequences on a finer time-scale; e.g. a syllable comprises a specific sequence of phonemes. This functional hierarchy of time-scales may be reflected in the hierarchical, anatomical organisation of the brain [6]. For example, in avian brains, there is anatomical and functional evidence that birdsong is generated and perceived by a hierarchical system, where low levels represent transient acoustic details and high levels encode song structure at slower time-scales [7],[8]. An equivalent temporal hierarchy might also exist in the human brain for representing auditory information, such as speech [1], [9]–[12].
Here we ask the following question: How does the brain recognize the dynamic and ambiguous causes of noisy sensory input? Based on experimental and theoretical evidence [13]–[18] we assume the brain is a recognition system that uses an internal model of its environment. The structure of this model is critical: On one hand, the form of the model must capture the essential architecture of the process generating sensory data. On the other hand, it must also support robust inference. We propose that a candidate that fulfils both criteria is a model based on a hierarchy of stable heteroclinic channels (SHCs). SHCs have been introduced recently as a model of neuronal dynamics per se [19]. Here, we use SHCs as the basis of neuronal recognition, using an established Bayesian scheme for modelling perception [20]. This brings together two recent developments in computational approaches to perception: Namely, winnerless competition in stable heteroclinic channels and the hypothesis that the brain performs Bayesian inference. This is important because it connects a dynamic systems perspective on neuronal dynamics [19],[21],[22] with the large body of work on the brain as an inference machine [13]–[18].
To demonstrate this we generate artificial speech input (sequences of syllables) and describe a system that can recognize these syllables, online from incoming sound waves. We show that the resulting recognition dynamics display functional characteristics that are reminiscent of psychophysical and neuronal responses.
In this section, we describe an online recognition scheme for continuous sequences with hierarchical structure. This scheme rests on the concept of stable heteroclinic channels (SHCs) [23], which are combined with an online Bayesian inversion scheme [20]. We now describe these elements and how they are brought together. Note that all variables and their meaning are also listed in Table 1 and 2.
SHCs are attractors formed by artificial neuronal networks, which prescribe sequences of transient dynamics [22]–[25]. The key aspect of these dynamical systems is that their equations of motion describe a manifold with a series of saddle points. At each saddle point, trajectories are attracted from nearly all directions but are expelled in the direction of another saddle point. If the saddle points are linked up to form a chain, the neuronal state follows a trajectory that passes through all these points, thereby forming a sequence. These sequences are exhibited robustly, even in the presence of high levels of noise. In addition, the dynamics of the SHCs are itinerant due to dynamical instability in the equations of motion and noise on the states. This noise also induces a variation in the exact times that sequence elements are visited. This can be exploited during recognition, where the SHC places prior constraints on the sequence that elements (repelling fixed-points) are visited but does not constrain the exact timing of these visits.
The combination of these two features, robustness of sequence order but flexibility in sequence timing, makes the SHC a good candidate for the neuronal encoding of trajectories [19],[26]. Rabinovich et al. have used SHCs to explain how spatiotemporal neuronal dynamics observed in odour perception, or motor control of a marine mollusc, can be expressed in terms of a dynamic system [22],[27].
Varona et al. used Lotka-Volterra-type dynamics to model a network of six neurons in a marine mollusc [27]: With particular lateral inhibition between pairs of neurons and input to each neuron, the network displayed sequences of activity. Following a specific order, each neuron became active for a short time and became inactive again, while the next neuron became active, and so on. Stable heteroclinic channels rest on a particular form of attractor manifold that supports itinerant dynamics. This itinerancy can result from deterministic chaos in the absence of noise, which implies the presence of heteroclinic cycles. When noise is added, itinerancy can be assured, even if the original system has stable fixed-points. However, our motivation for considering stochastic differential equations is to construct a probabilistic model, where assumptions about the distribution of noise provide a formal generative model of sensory dynamics.
As reviewed in [22], Lotka-Volterra dynamics can be derived from simple neural mass models of mean membrane potential and mean firing rate [21]. Here, we use a different neural mass model, where the state-vector x can take positive or negative values:(1)where the motion of a hidden-state vector (e.g., mean membrane potentials) x is a nonlinear function of itself with scalar parameters , , and a connectivity matrix . The hidden state-vector enters a nonlinear function S to generate outcomes (e.g., neuronal firing rates) y. Each element determines the strength of lateral inhibition from state j to i. Both the state and observation equations above include additive normally distributed noise vectors w and z. When choosing specific parameter values (see below), the states display stereotyped sequences of activity [28]. Rabinovich et al. [19] termed these dynamics ‘stable heteroclinic channels’ (SHCs). If the channel forms a ring, once a state is attracted to a saddle point, it will remain in the SHC.
SHCs represent a form of itinerant dynamics [26],[29],[30] and may represent a substrate for neuronal computations [31]. Remarkably, the formation of SHCs seems to depend largely on the lateral inhibition matrix and not on the type of neuronal model; see Ivanchenko et al. [32] for an example using a complex two-compartment spiking neuron model.
In this paper, we propose to use SHCs not as a model for neuronal dynamics per se but as a generative model of how sensory input is generated. This means that we interpret x as hidden states in the environment, which generate sensory input y. The neuronal response to sampling sensory input y are described by recognition dynamics, which decode or deconvolve the causes x from that input. These recognition dynamics are described below. This re-interpretation of Eq. 1 is easy to motivate: sensory input is usually generated by our own body and other organisms. This means input is often generated by neuronal dynamics of the sort described in Eq. 1.
A SHC can generate repetitive, stereotyped sequences. For example, in a system with four saddle points, an SHC forces trajectories through the saddle points in a sequence, e.g. ‘1-2-3-4-1-2-3-4-1…’. In contrast, a SHC cannot generate ‘1-2-3-4-3-4-2-1…’, because the sequence is not repetitive. However, to model sensory input, for example speech, one must be able to recombine basic sequence-elements like phonemes in ever-changing sequences. One solution would be to represent each possible sequence of phonemes (e.g. each syllable) with a specific SHC. A more plausible and parsimonious solution is to construct a hierarchy of SHCs, which can encode sequences generated by SHCs whose attractor topology (e.g. the channels linking the saddle points) is changed by a supraordinate SHC. This can be achieved by making the connectivity matrix at a subordinate level a function of the output states of the supra-ordinate level. This enables the hierarchy to generate sequences of sequences to any hierarchical depth required.
Following a recent account of how macroscopic cortical anatomy might relate to time-scales in our environment [6], we can construct a hierarchy by setting the rate constant of the j-th level to a rate that is slower than its subordinate level, . As a result, the states of subordinate levels change faster than the states of the level above. This means the control parameters at any level change more slowly than its states, ; because the slow change in the attractor manifold is controlled by the supraordinate states:(2)where the superscript indexes level j (level 1 being the lowest level), are ‘hidden states’, and are outputs to the subordinate level, which we will call ‘causal states’. As before, at the first level, is the sensory stream. In this paper, we consider hierarchies with relative time-scales of around four. This means that the time spent in the vicinity of a saddle point at a supraordinate level is long enough for the subordinate level to go through several saddle points. As before, all levels are subject to noise on the motion of the hidden states and the causal states . At the highest level, the control parameters, are constant over time. At all other levels, the causal states of the supraordinate level, , enter the subordinate level by changing the control parameters, the connectivity matrix :(3)Here, is a linear mixture of ‘template’ control matrices , weighted by the causal states at level . Each of these templates is chosen to generate a SHC. Below, we will show examples of how these templates can be constructed to generate various sequential phenomena. The key point about this construction is that states from the supraordinate level select which template controls the dynamics of the lower level. By induction, the states at each level follow a SHC because the states at the supraordinate level follow a SHC. This means only one state is active at any time and only one template is selected for the lower level. An exception to this is the transition from one state to another, which leads to a transient superposition of two SHC-inducing templates (see below). Effectively, the transition transient at a specific level gives rise to brief spells of non-SHC dynamics at the subordinate levels (see results). These transition periods are characterized by dissipative dynamics, due to the largely inhibitory connectivity matrices, inhibition controlled by parameter (Eq. 2) and the saturating nonlinearity S.
In summary, a hierarchy of SHCs generates the sensory stream at the lowest (fastest) level, which forms a sequence of sequences expressed in terms of first-level states. In these models, the lower level follows a SHC, i.e. the states follow an itinerant trajectory through a sequence of saddle points. This SHC will change whenever the supraordinate level, which follows itself a SHC, moves from one saddle point to another. Effectively, we have constructed a system that can generate a stable pattern of transients like an oscillator; however, as shown below, the pattern can have deep or hierarchical structure. Next, we describe how the causes can be recognized or deconvolved from sensory input y.
We have described how SHCs can, in principle, generate sequences of sequences that, we assume, are observed by an agent as its input y. To recognise the causes of the sensory stream the agent must infer the hidden states online, i.e. the system does not look into the future but recognizes the current states and of the environment, at all levels of the hierarchy, by the fusion of current sensory input and internal dynamics elicited by past input. An online recognition scheme can be derived from the ‘free-energy principle’, which states that an agent will minimize its surprise about its sensory input, under a model it entertains about the environment; or, equivalently maximise the evidence for that model [18]. This requires the agent to have a dynamic model, which relates environmental states to sensory input. In this context, recognition is the Bayesian inversion of a generative model. This inversion corresponds to mapping sensory input to the posterior or conditional distribution of hidden states. In general, Bayesian accounts of perception rest on a generative model. Given such a model, one can use the ensuing recognition schemes in artificial perception and furthermore compare simulated recognition dynamics (in response to sensory input), with evoked responses in the brain. The generative model in this paper is dynamical and based on the nonlinear equations 1 and 2. More precisely, these stochastic differential equations play the role of empirical priors on the dynamics of hidden states causing sensory data.
In the following, we review briefly, the Bayesian model inversion described in [20] for stochastic, hierarchical systems and apply it, in the next section, to hierarchical SHCs.
Given some sensory data vector y, the general inference problem is to compute the model evidence or marginal likelihood of y, given a model m:(4)where the generative model is defined in terms of a likelihood and prior on hidden states. In Equation 4, the state vector subsumes the hidden and causal states at all levels of a hierarchy (Eq. 2). The model evidence can be estimated by converting this difficult integration problem (Eq. 4) into an easier optimization problem by optimising a free-energy bound on the log-evidence [33]. This bound is constructed using Jensen's inequality and is a function of an arbitrary recognition density, :(5)The free-energy comprises an energy term and an entropy term and is defined uniquely, given a generative model . The free-energy is an upper bound on the surprise or negative log-evidence, because the Kullback-Leibler divergence , between the recognition and conditional density, is always positive. Minimising the free-energy minimises the divergence, rendering the recognition density an approximate conditional density. When using this approach, one usually employs a parameterized fixed-form recognition density, [20]. Inference corresponds to optimising the free-energy with respect to the sufficient statistics, of the recognition density:(6)The optimal statistics are sufficient to describe the approximate posterior density; i.e. the agent's belief about (or representation of) the trajectory of the hidden and causal states. We refer the interested reader to Friston et al. [34] for technical details about this variational Bayesian treatment of dynamical systems. Intuitively, this scheme can be thought of as augmented gradient descent on a free-energy bound on the model's log-evidence. Critically, it outperforms conventional Bayesian filtering (e.g., Extended Kalman filtering) and eschews the computation of probability transition matrices. This means it can be implemented in a simple and neuronally plausible fashion [20].
In short, this recognition scheme operates online and recognizes current states of the environment by combining current sensory input with internal recognition dynamics, elicited by past input.
A recognition system that minimizes its free-energy efficiently will come to represent the environmental dynamics in terms of the sufficient statistics of recognition density; e.g. the conditional expectations and variances of . We assume that the conditional moments are encoded by neuronal activity; i.e., Equation 6 prescribes neuronal recognition dynamics. These dynamics implement Bayesian inversion of the generative model, under the approximations entailed by the form of the recognition density. Neuronally, Equation 6 can be implemented using a message passing scheme, which, in the context of hierarchical models, involves passing prediction errors up and passing predictions down, from one level to the next. These prediction errors are the difference between the causal states (Equation 2);(7)at any level j, and their prediction from the level above, evaluated at the conditional expectations [18],[35]. In addition, there are prediction errors that mediate dynamical priors on the motion of hidden states within each level (Equation 2);(8)This means that neuronal populations encode two types of dynamics: the conditional expectations of states of the world and the prediction errors. The dynamics of the first are given by Equation 6, which can be formulated as a function of prediction error. These dynamics effectively suppress or explain away prediction error; see [34] for details.
This inversion scheme is a generic recognition process that receives dynamic sensory input and can, given an appropriate generative model, rapidly identify and track environmental states that are generating current input. More precisely, the recognition dynamics resemble the environmental (hidden) states they track (to which they are indirectly coupled), but differ from the latter because they are driven by a gradient descent on free-energy; Eq. 6 (i.e. minimize prediction errors: Eqs. 7 and 8). This is important, because we want to use SHCs as a generative model, not as a model of neuronal encoding per se. This means that the neuronal dynamics will only recapitulate the dynamics entitled by SHCs in the environment, if the recognition scheme can suppress prediction errors efficiently in the face of sensory noise and potential beliefs about the world.
We are now in a position to formulate hierarchies of SHCs as generative models, use them to generate sensory input and simulate recognition of the causal states generating that input. In terms of low-level speech processing, this means that any given phoneme will predict the next phoneme. At the same time, as phonemes are recognized, there is also a prediction about which syllable is the most likely context for generating these phonemes. This prediction arises due to the learnt regularities in speech. In turn, the most likely syllable predicts the next phoneme. This means that speech recognition can be described as a dynamic process, on multiple time-scales, with recurrently evolving representations and predictions, all driven by the sensory input.
In the auditory system, higher cortical levels appear to represent features that are expressed at slower temporal scales [36]. Wang et al. [37] present evidence from single-neuron recordings that there is a ‘slowing down’ of representational trajectories from human auditory sensory thalamus (a ‘relay’ to the primary auditory cortex), the medial geniculate body (MGB) to primary auditory cortex (AI). In humans, it has been found that the sensory thalamus responds preferentially to faster temporal modulations of sensory signals, whereas primary cortex prefers slower modulations [10]. These findings indicate that neuronal populations, at lower levels of the auditory system (e.g. MGB), represent faster environmental trajectories than higher levels (e.g., A1). Specifically, the,MGB responds preferentially to temporal modulations of ∼20 Hz (∼50 ms), whereas AI prefers modulations at ∼6 Hz (∼150 ms) [10]. Such a temporal hierarchy would be optimal for speech recognition, in which information over longer time-scales provides predictions for processing at shorter time scales. In accord with this conjecture, optimal encoding of fast (rapidly modulated) dynamics by top-down predictions has been found to be critical for communication [1],[12],[38].
We model this ‘slowing down’ with a hierarchical generative model based on SHCs. This model generates sequences of syllables, where each syllable is a sequence of phonemes. Phonemes are the smallest speech sounds that distinguishes meaning and a syllable is a unit of organization for a sequence of phonemes. Each phoneme prescribes a sequence of sound-wave modulations which correspond to sensory data. We generated data in this fashion and simulated online recognition (see Figure 1). By recognizing speech-like phoneme-sequences, we provide a proof-of-principle that a hierarchical system can use sensory streams to infer sequences. This not only models the slowing down of representations in the auditory system [10],[12],[37],[38], but may point to computational approaches to speech recognition. In summary, the recognition dynamics following Equation 6 are coupled to a generative model based on SHCs via sensory input. The systems generating and recognising states in Fig. 1 are both dynamic systems, where a non-autonomous recognition system is coupled to an autonomous system generating speech.
All our simulations used hierarchies with two levels (Figure 2). The first (phonemic) level produces a sequence of phonemes, and the second (syllabic) level encodes sequences of syllables. We used Equation 2 to produce phoneme sequences, where the generating parameters are listed in Table 3. The template matrices (Equation 3) were produced in the following way: We first specified the sequence each template should induce; e.g., sequence 1-2-3 for three neuronal populations. We then set elements on the main diagonal to 1, the elements (2,1), (3,2), (1,3) to value 0.5, and all other elements to 5 [28]. More generally for sequence (9)Note that SHC hierarchies can be used to create a variety of different behaviours, using different connectivity matrices. Here we explore only a subset of possible sequential dynamics.
When generating sensory data y, we added noise and to both the hidden and causal states. At the first and second levels, this was normally distributed zero-mean noise with log-precisions of ten and sixteen, respectively. These noise levels were chosen to introduce noisy dynamics but not to the extent that the recognition became difficult to visualise. We repeated all the simulations reported below with higher noise levels and found that the findings remained qualitatively the same (results not shown). Synthetic stimuli were generated by taking a linear mixture of sound waves extracted from sound files, in which a single speaker pronounced each of four vowel-phonemes: [a], [e], [i], [o]. These extracts W were sampled at 22050 Hz and about 14 ms long. The mixture was weighted by the causal states of the phonemic level; . This resulted in a concatenated sound wave file w. When this sound file is played, one perceives a sequence of vowels with smooth, overlapping transitions (audio file S1). These transitions are driven by the SHCs guiding the expression of the phonemes and syllables at both levels of the generative hierarchy.
For computational simplicity, we circumvented a detailed generative model of the acoustic level. For simulated recognition, the acoustic input (the sound wave) was transformed to phonemic input by inverting the linear mixing described above every seven ms of simulated time (one time bin). This means that our recognition scheme at the acoustic level assumes forward processing only (Fig. 1). However, in principle, given an appropriate generative model [39],[40], one could invert a full acoustic model, using forward and backward message passing between the acoustic and phonemic levels.
In this section, we illustrate that the recognition scheme described above can reliably decode syllabic and phonemic structure from sensory input online, if it has the correct generative model. We will also describe how recognition fails when the generative model does not have a form that provides veridical predictions of the sensorium, e.g., when agents are not conspecific or we hear a foreign language. These simulations relate to empirical studies of brain responses evoked by unpredicted linguistic stimuli. We conclude with a more subtle violation that we deal with in everyday audition; namely the recognition of speech presented at different speeds.
To create synthetic stimuli we generated syllable sequences consisting of four phonemes or states; [a], [e], [i], and [o], over 11.25 seconds (800 time points), using a two-level SHC model (Fig. 2). To simulate word-like stimuli, we imposed silence at the beginning and the end by windowing the phoneme sequence (Fig. 3A, top left). At the syllabic level, we used three syllables or states to form the second-level sequence (1–2–3)(2); where the numbers denote the sequence and the superscript indicates the sequence level. The three causal states of the syllabic level entered the phonemic level as control parameters to induce their template matrices as in Equation 3. This means that each of the three syllable states at the second level causes a phoneme sequence at the first: , , and , see Fig. 2 and listen to the audio file S1. In Fig. 3A we show the causal and hidden states, at both levels, generated by this model. The remaining parameters, for both levels, are listed in Table 3. Note that the rate constant of the syllabic level is four times slower than at the phonemic level. As expected, the phoneme sequence at the first level changes as a function of the active syllable at the second level. The transients caused by transitions between syllables manifest at the first level as temporary changes in the amplitude or duration of the active phoneme.
We then simulated recognition of these sequences. Fig. 3B shows that our recognition model successfully tracks the true states at both levels. Note the recognition dynamics rapidly ‘lock onto’ the causal states from the onset of the first phoneme of the first syllable (time point 50). Interestingly, the system did not recognize the first syllable (true: syllable 3 (red line), recognized: syllable 2 (green line) between time points 50 to 80 (see red arrow in Fig. 3B), but corrected itself fairly quickly, when the sensory stream indicated a new phoneme that could only be explained by the third syllable. This initial transient at the syllabic level shows that recognition dynamics can show small but revealing deviations from the true state dynamics. In principle, these deviations could be used to test whether the real auditory system uses a recognition algorithm similar to the one proposed; in particular, the simulated recognition dynamics could be used to explain empirical neurophysiological responses.
What happens if the stimuli deviate from learned expectations (e.g. violation of phonotactic rules)? In other words, what happens if we presented known phonemes that form unknown syllables? This question is interesting for two reasons. First, our artificial recognition scheme should do what we expect real brains to do when listening to a foreign language: they should be able to recognize the phonemes but should not derive high-order ‘meaning’ from them; i.e. should not recognize any syllable. Secondly, there are well-characterised brain responses to phonotactic violations, e.g. [41]–[43]. These are usually event-related responses that contain specific waveform components late in peristimulus time, such as the N400. The N400 is an event-related potential (ERP) component typically elicited by unexpected linguistic stimuli. It is characterized as a negative deflection (topologically distributed over central-parietal sites on the scalp), peaking approximately 400 ms after the presentation of an unexpected stimulus.
To model phonotactic violations, we generated data with the two-level model presented above. However, we used syllables, i.e. sequences of phonemes, that the recognition scheme was not informed about and consequently could not recognise (it has three syllables in its repertoire: , , and ). Thus the recognition scheme knows all four phonemes but is unable to predict the sequences heard. Fig. 4A shows that the recognition system cannot track the syllables; the recognized syllables are very different from the true syllable dynamics. At the phonemic level, the prediction error deviates from zero whenever a new (unexpected) phoneme is encountered (Fig. 4B). The prediction error at the syllabic level is sometimes spike-like and can reach high amplitudes, relative to the typical amplitudes of the true states (see Fig. 4A and B). This means that the prediction error signals violation of phonotactic rules. In Fig. 4C, we zoom in onto time points 440 to 470 to show how the prediction error evolves when evidence of a phonotactic violation emerges: At the phoneme level, prediction error builds up because an unexpected phoneme appears. After time point 450, the prediction error grows quickly, up to the point that the system resolves the prediction error. This is done by ‘switching’ to a new syllable, which can explain the transition to the emerging phoneme. The switching creates a large amplitude prediction error at time point 460. In other words, in face of emerging evidence that its current representation of syllables and phonemes cannot explain sensory input, the system switches rapidly to a new syllable representation, giving rise to a new prediction error. It may be that these prediction errors are related to electrophysiological responses to violations of phonotactic rules, [44],[45]. This is because the largest contributors to non-invasive electromagnetic signals are thought to be superficial pyramidal cells. In biological implementations of the recognition scheme used here [20], these cells encode prediction error.
In summary, these simulations show that a recognition system cannot represent trajectories or sequences that are not part of its generative model. In these circumstances, recognition experiences intermittent high-amplitude prediction errors because the internal predictions do not match the sensory input. There is a clear formal analogy between the expression of prediction error in these simulations and mismatch or prediction violation responses observed empirically. The literature that examines event-related brain potentials (ERPs) and novelty processing “reveals that the orienting response engendered by deviant or unexpected events consists of a characteristic ERP pattern, comprised sequentially of the mismatch negativity (MMN) and the novelty P3 or P3a” [46].
Human speech recognition is robust to the speed of speech [47],[48]. How do our brains recognize speech at different rates? There are two possible mechanisms in our model that can deal with ‘speaker speed’ parameters online. First, one could make the rate constants and free parameters and optimise them during inversion. Adjusting to different speaker parameters is probably an essential faculty, because people speak at different speeds [49]. The second mechanism is that the recognition itself might be robust to deviations from the expected rate of phonemic transitions; i.e., even though the recognition uses the rate parameters appropriate for much slower speech, it still can recognize fast speech. This might explain why human listeners can understand speech at rates that they have never experienced previously [47]. In the following, we show that our scheme has this robustness.
To simulate speed differences we used the same two-level model as in the simulations above with for the generation of phonemes, but with for recognition so that the stimulus stream was 50% faster than expected. As can be seen in Fig. 5A, the recognition can successfully track the syllables. This was because the second level supported the adaption to the fast sensory input by changing its recognition dynamics in responses to prediction error (see Fig. 5B: note the amplitude difference in Fig. 5A between the true and recognized ). The prediction errors at both levels, and , are shown in Fig. 5C. In particular, the second-level error displayed spike-like corrections around second-level transitions. These are small in amplitude compared to both the amplitude of the hidden states and the prediction errors of the previous simulation (Fig. 4B). These results show that the system can track the true syllables veridically, where the prediction error accommodates the effects caused by speed differences. This robustness to variations in the speed of phoneme transitions might be a feature shared with the auditory system [50].
We have shown that stable heteroclinic channels (SHCs) can be used as generative models for online recognition. In particular, we have provided proof-of-concept that sensory input generated by these hierarchies can be deconvolved to disclose the hidden states causing that input. This is a non-trivial observation because nonlinear, hierarchical and stochastic dynamical systems are difficult to invert online [51],[52]. However, we found that the inversion of models based on SHCs is relatively simple. Furthermore, the implicit recognition scheme appears robust to noise and deviations from true parameters. This suggests that SHCs may be a candidate for neuronal models that contend with the same problem of de-convolving causes from sensory consequences. Moreover, hierarchical SHCs seem, in principle, an appropriate description of natural sequential input, which is usually generated by our own body or other organisms, and can be described as a mixture of transients and discrete events.
The general picture of recognition that emerges is as follows: Sensory input is generated by a hierarchy of dynamic systems in the environment. We couple this dynamic system, via sensory sampling, to our recognition system implementing the inversion dynamics (Fig. 1). The recognition system minimizes a proxy for surprise or model evidence; the negative free-energy (Eq. 6). To do this, the states of the recognition system move on manifolds, defined through the free-energy by the generative model. Here, we use a hierarchy of SHCs as generative model so that the manifold changes continuously at various time-scales. The inferred SHC states never reach a fixed point, but are perpetually following a trajectory through state-space, in the attempt to mirror the generative dynamics of the environment. When sensory input is unexpected (see second simulation, Fig. 4), the system uses the prediction error to change its representation quickly, at all levels, such that it best explains the sensory stream.
In a previous paper [6], we have shown that one can use chaotic attractors (i.e., a hierarchy of Lorenz attractors) to model auditory perception. However, SHCs may provide a more plausible model of sensory dynamics: First, they show structure over extended temporal scales, much like real sensory streams. This may reflect the fact that the processes generating sensory data are themselves (usually) neuronal dynamics showing winnerless competition. Secondly, many chaotic systems like the Lorenz attractor have only few states and cannot be extended to high dimensions in a straightforward fashion. This was no problem in our previous model, where we modelled a series of simple chirps, with varying amplitude and frequency [6]. However, it would be difficult to generate sequences of distinct states that populate a high dimensional state-space; e.g. phonemes in speech. In contrast, stable heteroclinic channels can be formulated easily in high dimensional state spaces.
In this paper, we used a generative model which was formally identical to the process actually generating sensory input. We did this for simplicity; however, any generative model that could predict sensory input would be sufficient. In one sense, there is no true model because it is impossible to disambiguate between models that have different forms but make the same predictions. This is a common issue in ill-posed inverse problems, where there are an infinite number of models that could explain the same data. In this context the best model is usually identified as the most parsimonious. Furthermore, we are not suggesting that all aspects of perception can be framed in terms of the inversion of SHCs; we only consider recognition of those sensory data that are generated by mechanisms that are formally similar to the itinerant but structured dynamics of SHCs.
The proof-of-concept presented above makes the SHC hierarchy a potential candidate for speech recognition models. The recognition dynamics we simulated can outpace the dynamics they are trying to recognise. In all our simulations, after some initial transient, the recognition started tracking the veridical states early in the sequence. For example, the scheme can identify the correct syllable before all of its phonemes have been heard. We only simulated two levels, but this feature of fast recognition on exposure to brief parts of the sequence may hold for many more levels. Such rapid recognition of potentially long sequences is seen in real systems; e.g., we can infer that someone is making a cup of tea from observing a particular movement, like getting a teabag out of a kitchen cupboard. The reason why recognition can be fast is that the generative model is nonlinear (through the top-down control of attractor manifolds). With nonlinearities, slow time-scales in hierarchical sequences can be recognized rapidly because they disclose themselves in short unique sequences in the sensory input. Furthermore, we demonstrated another requirement for efficient communication: recognition signals, via prediction error, when unrecognised syllables cannot be decoded with its phonotactic model. This is important, because, an agent can decide online whether its decoding of the message is successful or not. Following the free-energy principle, this would oblige the agent to act on its environment, so that future prediction error is minimized [18]. For example, the prediction error could prompt an action (‘repeat, please’) and initiate learning of new phonotactic rules.
Another aspect of SHC-based models is that they can recombine sensory primitives like phonemes in a large number of ways. This means that neuronal networks implementing SHC dynamics, based on a few primitives at the first level, can encode a large number of sequences. This feature is critical for encoding words in a language; e.g., every language contains many more words than phonemes [53]. The number of sequences that a SHC system can encode is(10)where N is the number of elements [22]. This would mean, in theory, that the number of states that can be encoded with a sequence, given a few dozens primitives, is nearly endless. It is unlikely that this full capacity is exploited in communication. Rather, for efficient communication, it might be useful to restrict the number of admissible sequences to make them identifiable early in the sequence.
We did not equip the recognition model with a model of the silent periods at the beginning and end of a word (Fig. 3A). It is interesting to see how recognition resolves this: to approximate silence, the system held hidden phoneme states very negative by driving the states away from the SHC attractor and tolerating the violation of top-down predictions. However, the tolerance is limited as can be seen by the slightly positive inferred hidden states (Fig. 3B). Such behaviour is beneficial for recognition because the agent, within bounds, can deviate from internal predictions. A built-in error tolerance which is sensitive to the kind of errors it should endure to make recognition robust is important in an uncertain world. Robustness to errors would be impossible with an inversion scheme based on a deterministic model, which assumes that the sensory input follows a deterministic trajectory without any noise on the environmental causes. With such a recognition system, the agent could not deal with (unexpected) silence, because the SHC-based inversion dynamics would attract the state-trajectory without any means of resolving the resulting prediction error between the zero (silent) sensory input and the internal predictions. Recognition schemes based on stochastic systems can deviate adaptively from prior predictions, with a tolerance related to the variance of the stochastic innovations. Optimising this second-order parameter then becomes critical for recognition (see [20]).
There is emerging evidence in several areas of neuroscience that temporal hierarchies play a critical role in brain function [6]. The three areas where this is most evident are auditory processing [12], [37], [54]–[56], cognitive control [57]–[59], and motor control [60]. Our conclusions are based on a generic recognition scheme [20] and are therefore a consequence of our specific generative model, a temporal hierarchy of SHCs. This hierarchy of time-scales agrees well with the temporal anatomy of the hierarchical auditory system, where populations close to the periphery encode the fast acoustics, while higher areas form slower representations [9],[10],[37],[38],[61],[62]. In particular, our model is consistent with findings that phonological (high) levels have strong expectations about the relevance of acoustic (low) dynamics [38].
Neurobiological treatments of the present framework suppose that superficial pyramidal cell populations encode prediction error; it is these cells that contribute most to evoked responses as observed in magneto/electroencephalography (M/EEG) [63]. There is an analogy between the expression of prediction error in our simulations and mismatch or prediction violation responses observed empirically. In our simulations, prediction error due to a deviation from expectations is resolved by all levels (Fig. 4B). This might be an explanation for prominent responses to prediction violations to be spatially distributed, e.g., the mismatch negativity, the P300, and the N400 all seem to involve various brain sources in temporal and frontal regions [45], [46], [64]–[66]. Inference on predictable auditory streams has been studied and modelled in several ways, in an attempt to explain the rapid recognition of words in the context of sentences, e.g., [38], [67]–[70]. Our simulations show how, in principle, these accounts might be implemented in terms of neuronal population dynamics.
Learning, storing, inferring and executing sequences is a key topic in experimental [71]–[79], and theoretical neurosciences [80]–[82]; and robotics [83]–[86]. An early approach to modelling sequence processing focussed on feed-forward architectures. However, it was realised quickly that these networks could not store long sequences, because new input overwrote the internal representation of past states. The solution was to introduce explicit memory into recurrent networks, in various forms; e.g. as contextual nodes or ‘short-term memory’ [87],[88]. Although framed in different terms, these approaches can be seen as an approximation to temporal hierarchies, where different units encode representations at different time-scales.
A central issue in modelling perception is how sequences are not just recalled but used as predictions for incoming sensory input. This requires the ‘dynamic fusion’ of bottom-up sensory input and top-down predictions, Several authors e.g., [83], [89]–[92] use recurrent networks to implement this fusion. Exact Bayesian schemes based on discrete hierarchical hidden Markov models, specified as a temporal hierarchy, have been used to implement memory and recognition [93]. Here, we have used the free-energy principle (i.e. variational Bayesian inference on continuous hierarchical dynamical systems) to show how the ensuing recognition process leads naturally to a scheme which can deal with fast sequential inputs.
In conclusion, we have described a scheme for inferring the causes of sensory sequences with hierarchical structure. The key features of this scheme are: (i) the ability to describe natural sensory input as hierarchical and dynamic sequences, (ii) modeling this input using generative models, (iii) using dynamic systems theory to create plausible models, and (iv) online Bayesian inversion of the resulting models. This scheme is theoretically principled but is also accountable to the empirical evidence available from the auditory system; furthermore, the ensuing recognition dynamics are reminiscent of real brain responses.
|
10.1371/journal.pgen.1001181 | Facioscapulohumeral Dystrophy: Incomplete Suppression of a Retrotransposed Gene | Each unit of the D4Z4 macrosatellite repeat contains a retrotransposed gene encoding the DUX4 double-homeobox transcription factor. Facioscapulohumeral dystrophy (FSHD) is caused by deletion of a subset of the D4Z4 units in the subtelomeric region of chromosome 4. Although it has been reported that the deletion of D4Z4 units induces the pathological expression of DUX4 mRNA, the association of DUX4 mRNA expression with FSHD has not been rigorously investigated, nor has any human tissue been identified that normally expresses DUX4 mRNA or protein. We show that FSHD muscle expresses a different splice form of DUX4 mRNA compared to control muscle. Control muscle produces low amounts of a splice form of DUX4 encoding only the amino-terminal portion of DUX4. FSHD muscle produces low amounts of a DUX4 mRNA that encodes the full-length DUX4 protein. The low abundance of full-length DUX4 mRNA in FSHD muscle cells represents a small subset of nuclei producing a relatively high abundance of DUX4 mRNA and protein. In contrast to control skeletal muscle and most other somatic tissues, full-length DUX4 transcript and protein is expressed at relatively abundant levels in human testis, most likely in the germ-line cells. Induced pluripotent (iPS) cells also express full-length DUX4 and differentiation of control iPS cells to embryoid bodies suppresses expression of full-length DUX4, whereas expression of full-length DUX4 persists in differentiated FSHD iPS cells. Together, these findings indicate that full-length DUX4 is normally expressed at specific developmental stages and is suppressed in most somatic tissues. The contraction of the D4Z4 repeat in FSHD results in a less efficient suppression of the full-length DUX4 mRNA in skeletal muscle cells. Therefore, FSHD represents the first human disease to be associated with the incomplete developmental silencing of a retrogene array normally expressed early in development.
| Facioscapulohumeral muscular dystrophy is caused by the deletion of a subset of D4Z4 macrosatellite repeats on chromosome 4. Each repeat contains a retrogene encoding the double-homeobox factor DUX4. We show that this retrogene is normally expressed in human testis, most likely the germ-line cells, and pluripotent stem cells. DUX4 expression is epigenetically suppressed in differentiated tissues and the residual DUX4 transcripts are spliced to remove the carboxyterminal domain that has been associated with cell toxicity. In FSHD individuals, the expression of the full-length DUX4 transcript is not completely suppressed in skeletal muscle, and possibly other differentiated tissues, and results in a small percentage of cells expressing relatively abundant amounts of the full-length DUX4 mRNA and protein. We therefore propose that FSHD is caused by the inefficient developmental suppression of the DUX4 retrogene and that the residual expression of the full-length DUX4 in skeletal muscle is sufficient to cause the disease. Therefore, FSHD represents the first human disease to be associated with the incomplete developmental silencing of a retrogene array that is normally expressed early in development.
| Facioscapulohumeral dystrophy (FSHD) is an autosomal dominant muscular dystrophy caused by the deletion of a subset of D4Z4 macrosatellite repeat units in the subtelomeric region of 4q on the 4A161 haplotype (FSHD1; OMIM 158900) [1]. The unaffected population has 11–100 D4Z4 repeat units, whereas FSHD1 is associated with 1–10 units [2]. The retention of at least a portion of the D4Z4 macrosatellite in FSHD1 and the demonstration that the smaller repeat arrays have diminished markings of heterochromatin [3] support the hypothesis that repeat contraction results in diminished heterochromatin-mediated repression of a D4Z4 transcript, or a transcript from the adjacent subtelomeric region. The hypothesis that derepression of a regional transcript causes FSHD is further supported by individuals with the same clinical phenotype and decreased D4Z4 heterochromatin markings but without a contraction of the D4Z4 macrosatellite in the pathogenic range (FSHD2) [4], [5].
The D4Z4 repeat unit contains a conserved open reading frame for the DUX4 retrogene, which Clapp et al suggest originated from the retrotransposition of the DUXC mRNA [6], a gene present in many mammals but lost in the primate lineage. Dixit et al [7] demonstrated that DUX4 transcripts were present in cultured FSHD muscle cells and mapped a polyadenylation site to the region telomeric to the last repeat, a region referred to as pLAM. Lemmers et al [8] recently demonstrated that the region necessary for a contracted D4Z4 array to be pathogenic maps to this polyadenylation site, which is intact on the permissive 4A chromosome but not on the non-permissive chromosomes 4B or 10, indicating that stabilization of the DUX4 mRNA is necessary to develop FSHD on a contracted allele. Our prior study [9] demonstrated bidirectional transcription of the D4Z4 region associated with the generation of small RNAs, and we suggested that these D4Z4-associated small RNAs might contribute to the epigenetic silencing of D4Z4. We also identified alternatively spliced transcripts from the DUX4 retrogene that terminate at the previously described [7] polyadenylation site in the pLAM region. However, we identified DUX4 mRNA transcripts in both FSHD and wild-type muscle cells, as well as similar amounts of D4Z4-generated small RNAs.
Together these studies implicate a stabilized DUX4 mRNA transcript from the contracted D4Z4 array as the cause of FSHD. However, several important questions remain to be addressed:
In this study, we address each of these important questions. Together, our data substantiate a developmental model for FSHD: full-length DUX4 mRNA is normally expressed early in development and is suppressed during cellular differentiation, whereas FSHD is associated with the failure to maintain complete suppression of full-length DUX4 expression in differentiated skeletal muscle cells. Occasional escape from repression results in the expression of relatively large amounts of DUX4 protein in a small number of skeletal muscle nuclei.
A recent study [8] demonstrated that the sequence polymorphisms of the 4A161 haplotype necessary for FSHD include the region of the poly-adenylation signal for the DUX4 mRNA and showed that this correlated with the detection of DUX4 mRNA in three FSHD muscle cultures compared to controls. Our previous study of RNA transcripts from D4Z4 repeat units identified a full-length mRNA transcript that contains the entire DUX4 open reading frame and has one or two introns spliced in the 3-prime UTR (GenBank HQ266760 and HQ266761), and a second mRNA transcript utilizing a cryptic splice donor in the DUX4 ORF that maintains the amino-terminal double-homeobox domains and removes the carboxyterminal end of DUX4 (GenBank HQ266762) (Figure 1A and 1B). We will refer to these two transcripts as DUX4-fl (full length) and DUX4-s (shorter ORF), respectively (see [9] for splice junction sequences). The PCR approach in the Lemmers et al study [8] would not have detected the DUX4-s mRNA.
We used oligo-dT primed cDNA and a PCR strategy that would detect both DUX4-fl and DUX4-s (see Figure 1B) to determine the presence of polyadenylated DUX4 mRNAs in quadriceps muscle needle biopsies from ten FSHD and fifteen control individuals (Table 1 and Figure 1C). In general we used two cycles of PCR with nested primers to increase specificity and to detect low abundance transcripts. DUX4-fl was detected in five of the ten FSHD samples, based on primers amplifying DUX4-fl and primers amplifying the 3-prime region of DUX4-fl (DUX4-fl3′) that is contained in DUX4-fl but not in DUX4-s (see Figure 1B). The sequenced products matched the FSHD-permissive 4A161 haplotype polymorphisms and the variation in size of the PCR product reflected alternative splicing of only the second intron in the UTR or both the first and second UTR introns (see Figure 1B). In contrast, none of the fifteen control samples expressed mRNA that amplified with primers to DUX4-fl or DUX4-fl3′, including seven biopsies from individuals with at least one 4A161 chromosome. Instead, DUX4-s was detected in all control samples with 4A161 and in some of the FSHD samples. We did not detect DUX4 transcripts using these primers in six control biopsies that do not contain the 4A chromosome. These data indicate that the 4A D4Z4 region is actively transcribed and produces alternatively spliced and polyadenylated DUX4 mRNA in both FSHD and unaffected individuals. However, the full-length DUX4 mRNA was only detected in the FSHD muscle biopsies, whereas DUX4-s was detected in muscle from controls and some FSHD individuals.
The expression of DUX4-fl mRNA in FSHD muscle biopsies could be a primary consequence of the D4Z4 contraction or a secondary response to the inflammation associated with muscle degeneration and/or regeneration. Therefore, we extended our analysis to myoblast cultures derived from four control and six FSHD individuals, including one individual with FSHD2. As seen in the muscle biopsies, the control muscle cells contained no detectable amounts of DUX4-fl mRNA, whereas muscle cells derived from both FSHD1 and FSHD2 samples expressed DUX4-fl transcripts as well as the DUX4fl-3′ (Table 2 and Figure 1D). All control and a subset of the FSHD samples expressed DUX4-s. These data are consistent with observations made in the muscle biopsies and indicate that both FSHD and control muscle cells actively transcribe DUX4. Unaffected cells produce DUX4-s from a splice donor site in the DUX4 ORF, whereas FSHD cells produce DUX4-fl with an alternative splice donor site after the translation termination codon of the DUX4 ORF.
In both control and FSHD cells the DUX4 mRNA transcripts, either DUX4-fl or DUX4-s, were only detected after nested PCR amplifications, indicating very low abundance of DUX4 mRNA in the FSHD and control biopsies and cells. We used the 9A12 mouse monoclonal anti-DUX4 antibody [7] and also produced mouse and rabbit monoclonal antibodies to the amino-terminal and carboxyterminal portion of the DUX4 protein [10], but were unable to detect DUX4 protein in western analysis of FSHD muscle cultures, consistent with the very low amounts of DUX4 mRNA.
Low transcript abundance could reflect a small number of transcripts in every cell or a large number of transcripts in a small subset of cells in the population. We assessed the presence of DUX4-fl mRNA in samplings of 100, 600, and 10,000 FSHD cultured muscle cells. DUX4-fl mRNA was present in five-out-of-ten pools of 600 cells (Figure 2A) and three-out-of-20 pools of 100 cells (data not shown), as well as in the single pool of 10,000 cells. This frequency of positive pools indicates that approximately one-out-of-1000 cells is expressing a relatively abundant amount of DUX4-fl mRNA at any given time. Immunostaining of cultured FSHD and control cultured muscle cells with four independent anti-DUX4 monoclonal antibodies showed that approximately one-out-of-1000 nuclei co-stained with an antibody to the amino-terminus and an antibody to the carboxy-terminus of DUX4 (Figure 2B), whereas no nuclei in the control cultures showed double-positive staining.
Both the mRNA analysis and the immunodetection indicate that approximately 0.1% of FSHD muscle nuclei express DUX4 mRNA and protein. This could represent transient bursts of expression or stochastic activation of expression that leads to cell death, or both. Forced expression of DUX4 has been shown to induce apoptosis in muscle cells [9], [11], [12]. When DUX4 is expressed in control human muscle cells by lenti-viral delivery, the DUX4 protein is distributed relatively homogeneously during the first 24 hrs and then aggregates in nuclear foci at 48 hrs when the cells are undergoing apoptosis (Figure 2C, panels c and d). These DUX4 nuclear foci associated with apoptosis are present in the nuclei of FSHD muscle cultures (compare panel d in Figure 2C with panels a–f in Figure 2B). Expression of DUX4-s in control human muscle cells does not induce apoptosis and does not accumulate in nuclear foci at 48 hrs (Figure 2, panel e). Therefore, the data indicates that FSHD muscle cells that express endogenous full-length DUX4 also exhibit the nuclear foci that are characteristic of DUX4-induced apoptosis.
Although there is no known function of DUX4 in human biology, the open reading frame has been conserved [6]. DUX4 is a retrogene thought to be derived from DUXC [6], or a DUXC-related gene, but also similar to the DUXA family mouse Duxbl gene [13]. Therefore, if DUX4 has a biological function it is likely to be similar to DUXC or Duxbl. Duxbl is expressed in mouse germ-line cells and we reasoned that because retrotranspositions entering the primate lineage must have occurred in the germ-line, then the parental gene to DUX4, either Duxbl or DUXC, must be expressed in the germ-line. Indeed, we detect the canine DUXC mRNA in canine testis but not in canine skeletal muscle (data not shown). Therefore, if DUX4 has a biological function similar to DUXC or Duxbl, we would anticipate DUX4 expression in the human germ-line.
We obtained RNA from different adult human tissues and identified DUX4-fl in testis (Figure 3A), whereas DUX4-s was present in a subset of differentiated tissues. DUX4-fl was detected in six additional testis samples, whereas only DUX4-s was detected in donor-matched skeletal muscle (Figure 3B and 3C). Quantitative PCR (qPCR) showed that human testis samples expressed almost 100-fold higher amounts of DUX4 mRNA compared to FSHD muscle biopsies, and almost 15-fold higher amounts compared to cultured FSHD muscle cells (Figure 3D). Western analysis using three different DUX4 antibodies identified a protein of the correct mobility in protein lysates from testes but not in other cells or tissues that do not express DUX4-fl mRNA, including control muscle cells (Figure 3E and data not shown). Furthermore, immunoprecipitation of testis proteins with rabbit anti-DUX4 antibodies followed by western with a mouse monoclonal antibody to DUX4 detected the same protein (Figure 3F). Western analysis of protein extracts from three additional human testis samples identified a similar band (data not shown). Immunostaining identified DUX4-expressing cells near the periphery of the seminiferous tubule that have the large round nucleus characteristic of spermatogonia or primary spermatocytes (Figure 4A–4C), and additional more differentiated appearing cells in the seminiferous tubules were also stained following antigen retrieval (Figure 4E). The large numbers and nuclear morphology of the cells staining with DUX4 in the seminiferous tubules, together with expression of DUX4 in the human germ-cell cell tumor lines SuSa and 833K [14] (data not shown), leads us to conclude that DUX4 is expressed in the germ-line lineage. Further studies will be necessary to determine more precisely the timing and cell stages of DUX4 expression in the in the testis and to ascertain whether it has a biological function.
The relatively high abundance of DUX4 mRNA and protein in human testes suggests a possible role for this protein in normal development. However, we have previously demonstrated that the alleles of chromosome 4 and 10 that are non-permissive for FSHD contain polymorphisms that inhibit polyadenylation of the DUX4 transcript, and, therefore, only the 4A allele would be predicted to make a DUX4 mRNA [8]. We do not have haplotype information on the testis donors and it is possible that some might lack the 4A haplotype entirely. To determine whether only the 4A haplotype produced stable DUX4 mRNA in human testes, we sequenced mRNAs from the seven testis samples in a region with informative polymorphisms regarding transcripts from 4A, 4B, and 10. All testis mRNA had transcripts from both chromosomes 4 and 10 in approximately equal amounts (Table 3) based on the informative polymorphisms (Table 4). Some samples had 4A and 4B haplotypes.
3-prime RACE analysis on testis mRNA demonstrated that the chromosome 10 transcripts used alternative 3-prime exons with a polyadenylation signal in exon 7 that is approximately 6.5 kb further telomeric than the previously identified 4A polyadenylation site in the pLAM region (GenBank HQ266763) (Figure 5). Some 4A transcripts also use the exon 7 polyadenylation site (Genbank HQ266764 and HQ266765), but the exon 3 polyadenylation site associated with the permissive allele is preferred (data not shown). The 4B transcripts do not use either the exon 3 or exon 7 polyadenylation sites since the 4B haplotype lacks these regions, however, we have not yet identified the full 3-prime sequence of the DUX4 mRNA from the 4B chromosome. Re-analysis of the muscle cell line, muscle biopsy, and somatic tissue transcripts did not identify any DUX4 mRNA utilizing the exon 7 polyadenylation site from either chromosome 10 or 4, including a control sample with a contraction to 9 copies of D4Z4 on chromosome 10 (biopsy 2318 in Table 1, data not shown). We conclude that chromosome 10 DUX4 transcripts in the testes use a distal exon 7 polyadenylation signal, whereas this region is not used in somatic tissues, even when the chromosome 10 D4Z4 array has contracted to ten repeats. Therefore, polyadenylated DUX4 mRNA from chromosomes 4 and 10 are present in the testis, but only chromosome 4A produces polyadenylated transcripts in somatic tissues.
The expression of DUX4-fl mRNA in unaffected human testes and the expression of DUX4-s in some unaffected somatic tissues, including skeletal muscle, suggested a developmental regulation of splice site usage in the DUX4 transcript. To directly determine whether the transition between DUX4-fl and DUX4-s expression is developmentally regulated, we generated induced pluripotent stem (iPS) cells from FSHD and control fibroblasts by expression of SOX2, OCT4, and KLF4 transcription factors from Moloney murine leukemia virus vectors [15]. Stem-cell clones had normal karyotypes, exhibited the expected cellular and colony morphology, contained tissue non-specific alkaline phosphatase activity, and expressed embryonic antigens (Figure 6A). RT-PCR demonstrated expression of stem cell markers NANOG, HTERT, cMYC, and endogenous transcripts from OCT4, SOX2, and KLF4 (Figure 6B). Pluripotency was demonstrated by the ability to form teratomas containing tissues derived from ectoderm, endoderm, and mesoderm (See Figure 6A). We used these characterized iPS cells to determine the expression of DUX4-fl and DUX4-s in the parental fibroblasts, undifferentiated iPS cells, and in the iPS cells after differentiation into embryoid bodies.
DUX4-s, but not DUX4-fl, was detected in control fibroblasts. In contrast, iPS cells derived from the control fibroblasts expressed DUX4-fl, whereas differentiation of these cells to embryoid bodies resulted in a switch to the expression of DUX4-s and loss of DUX4-fl transcripts (Table 2 and Figure 6C). In contrast, DUX4-fl was detected in FSHD fibroblasts and the iPS cells and embryoid bodies derived from FSHD fibroblasts. As expected, DUX4-fl3′ was detected in samples expressing DUX4-fl. (The relative amounts of DUX4-fl in a subset of iPS cells is shown in Figure 3D and a band migrating at the size of DUX4 was detected on a western with an anti-DUX4 antibody (data not shown)). DUX4-fl was detected in some human ES cell lines, but at much lower levels compared to the iPS cells (data not shown).
All of the splice donor and acceptor sites in the multiple alternative splicing events in the 3-prime UTR have consensus splice donor and acceptor sequences. In contrast, the splice donor in the ORF that produces DUX4-s is a non-canonical donor sequence and would normally not be favored for splicing. Recent studies have indicated that repressive chromatin modifications can favor splice donor usage [16] and we tested whether the degree of H3K9me3 correlated with the usage of the DUX4-s splice site. Chromatin immunoprecipitation showed that the control fibroblasts and embryoid bodies with DUX4-s expression had relatively higher levels of trimethylation of lysine 9 in histone H3 (H3K9me3), a repressive chromatin modification, compared to the control iPS cells, which express DUX4-fl (Figure 6D). The FSHD cells maintained relatively low levels of H3K9me3 in both iPS and differentiated cells. These findings are consistent with previous studies showing decreased H3K9me3 at the D4Z4 region in FSHD1 and FSHD2 [3] and suggest a correlation between the relatively higher levels of repressive chromatin modifications and the use of the cryptic splice donor to produce DUX4-s.
We note that prior studies reported the presence of polyadenylated DUX4 transcripts in a small number of samples of cultured FSHD muscle cells but not in control muscle cells [7], [8]. Our study both confirms and significantly extends these prior studies by (a) including a larger number of FSHD muscle cell cultures, (b) assaying controls that have a permissive 4A chromosome and non-permissive 4B chromosomes, (c) extending the analysis to mRNA from primary muscle biopsies of FSHD and haplotype-matched controls, (d) identifying the DUX4-s splice form of the DUX4 mRNA in control cells and showing that the qualitative difference between control and affected muscle is splice-site usage and not production of DUX4 mRNA; (e) demonstrating that the very low abundance of DUX4 mRNA in FSHD muscle represents a small percentage of nuclei with relatively high abundance mRNA and protein; (f) demonstrating that relatively high amounts of the DUX4 mRNA are expressed in the human testes and pluripotent cells and that developmental regulation is achieved by a combination of chromatin-associated splice-site usage and polyadenylation site usage.
Together our data provide the basis for a specific model of FSHD pathophysiology: (1) full-length DUX4 is produced from the last D4Z4 unit in early stem cells; (2) in differentiated tissues, the D4Z4 array is associated with increased repressive H3K9me3 and DUX4 expression is repressed; (3) in the residual transcripts that escape repression, an alternative first-intron splice donor is utilized to produce DUX4-s instead of DUX4-fl; (4) contraction of the D4Z4 arrays impedes the conversion to repressive chromatin and the transition from DUX4-fl to DUX4-s, resulting in expression of the full-length DUX4 in skeletal muscle and possibly other tissues; and (5) the very low levels of full-length DUX4 expression in FSHD muscle reflects relatively high amounts of expression in a small sub-population of cells. Several groups have shown that expression of full-length DUX4 in muscle cells can induce pathologic features of apoptosis and expression of PITX1 [7], [11], [17], [18]. In contrast, expression of DUX4c, a DUX4-like protein that lacks the carboxyterminal portion of DUX4, does not induce apoptosis [18]. Therefore, it is reasonable to believe that expression of DUX4-fl might induce muscle cell damage in FSHD, whereas DUX4-s expression would not be harmful to the cells. Indeed, FSHD muscle cells expressing the endogenous DUX4 have nuclear foci of DUX4 characteristic of the foci that appear during early stages of apoptosis when DUX4 is exogenously expressed in human skeletal muscle cells (see Figure 2), suggesting, but not yet proving, that these DUX4 expressing cells might be initiating a process of nuclear death.
The observed association of decreased H3K9me3 of D4Z4 with detectable levels of DUX4-fl mRNA suggests a specific mechanism of regulating DUX4 splicing. Previously [9], we demonstrated bidirectional transcription of the D4Z4 repeats with the generation of small si/mi/pi-like RNA fragments and suggested that the small RNAs generated from D4Z4 might function to suppress DUX4 expression in a developmental context, a suppression mechanism observed for other retrogenes [19], [20], [21]. A recent publication demonstrated that the small RNAs mediating heterochromatin formation also regulate splice-donor usage, either by targeting the nascent transcripts or by altering the rate of polymerase progression through condensed chromatin [16], [22]. Therefore, the repressive chromatin associated with D4Z4 in differentiated cells might facilitate the usage of the non-canonical splice donor to generate DUX4s, either through siRNAs from the region or through the impediment of polymerase progression, whereas the more permissive chromatin in FSHD and pluripotent cells might favor polymerase progression through to the consensus splice donor and generate DUX4-fl.
A recent study by Lemmers et al [8] identifies sequence variants on 4A necessary to produce polyadenylated DUX4 mRNA transcripts in somatic tissues. Our results are consistent with these findings since we have not been able to identify polyadenylated transcripts from non-permissive alleles in somatic tissues. In contrast, we do find alternative distal polyadenylation usage for DUX4 mRNA from non-permissive alleles in the testis. Developmentally regulated polyadenylation site usage has been described for other genes [23] and appears to be one additional mechanism of silencing expression of the DUX4 retrogene in somatic cells.
Our finding that the wild-type chromosomes 4 and 10 express a full-length DUX4 mRNA in human testes, most likely in the germ-line, and that the protein is relatively abundant suggests that DUX4 might have a normal role in development. This is supported by the expression of canine DUXC in germ-line tissue (L. Geng, unpublished data). In addition, a DUX4-like gene in the mouse, Duxbl, is expressed in mouse germ-line cells in both spermatogenesis and oogenesis, as well as in early phases of skeletal muscle development [13]. Similar to DUX4, Duxbl has developmentally regulated splicing to produce a full-length protein and a protein truncated after the double homoeodomains and studying the roles of Duxbl in germ-line and muscle development in mouse will likely inform our understanding of DUX4. We should note that our study describes the expression of human DUX4 in testes but we believe it is likely to be expressed in oogenesis as well. Limited access to appropriate tissue has limited our ability to carefully examine expression in cells of the ovary.
Generating new genes through retrotransposition is a common mechanism of mammalian evolution [24], particularly for genes with a role in germ cell development. Recently an FGF4 retrogene was identified as causing the short-legged phenotype in many dog breeds [25], indicating that retrogenes can direct dramatic phenotypic evolution in a population. Our study demonstrates that the expression of the DUX4 retrogene is developmentally regulated and might have a role in germ-line development, and, if similar to Duxbl, possibly in aspects of early embryonic muscle development. Maintaining the DUX4 retrogene in the primate lineage suggests some selective advantage compared to maintaining the parental gene itself. Based on current knowledge, this could be due to a function in germ-line development, or to a modulation of muscle mass in primate face and upper extremity. In this regard, it is interesting to speculate that a normal function of the DUX4 retrogene might be to a regulate the development of facial and upper-extremity muscle mass in the primates, and that FSHD represents a hypermorphic phenotype secondary to inefficient developmental suppression. Alternatively, the persistent expression of full-length DUX4 might induce a neomorphic phenotype unrelated to an evolutionarily selected role of DUX4. In either case, our findings substantiate a comprehensive developmental model of FSHD and demonstrate that FSHD represents the first human disease to be associated with the incomplete developmental silencing of a retrogene array that is expressed in pluripotent stem cells and in normal development.
This study used pre-existing and de-identified human tissue samples from tissue repositories and commercial sources and was approved by the Fred Hutchinson Cancer Research Center and the University of Washingtion Institutional Review Boards. Animal studies were approved by the University of Washington Institutional Animal Care and Use Committee and followed the Assessment and Accreditation of Laboratory Animal Care guidelines.
Muscle biopsy samples were collected from the vastus lateralis muscle of clinically affected and control individuals using standardized needle muscle biopsy protocol and cell cultures were derived from biopsies as described on the Fields Center website: http://www.urmc.rochester.edu/fields-center/protocols/documents/PreparingPrimaryMyoblastCultures.pdf. The sex, age, and severity score for the FSHD muscle biopsies were: F1998 (M, 43, 2); F0519 (M, 43, 4); F0515 (F, 48, 2); F0509 (M, 47, 2); F0531 (F, 47, 2); F2306 (F, 46, ND); F2331 (F, 56, 4); F2316 (F, 34, 5); F2319 (M, 52, ND); F2315 (F, 40, 3). Pathologic grading scale is 0–12 (from normal to severe) based on a score of 0–3 for each of four parameters: muscle fiber size/shape; degree of central nucleation; presence of necrotic/regenerating fibers or inflammation; and degree of fibrosis. Controls were selected in the same age range and sex representation. Muscle cell culture MB216 and muscle biopsy F2316 are from the same individual, otherwise the muscle cultures were derived from other individuals. RNA and protein lysates from human tissues were purchased from BioChain (Hayward, CA) and Origene (Rockville, MD).
Total RNA was isolated from muscle biopsies and cultured cells using Trizol (Invitrogen) and then treated with DNase I for 15 minutes using conditions recommended by Invitrogen with the addition of RNaseOUT (Invitrogen) to the reaction. DNase reaction components were removed using the RNeasy (Qiagen) system and RNA eluted by two sequential applications of 30 µl of RNase-free water. Volume was reduced by speed vac and 1.5–2 µg of RNA used for first strand cDNA synthesis. RNA from adult human tissues was purchased from Biochain and had been DNase-treated by the supplier. First strand synthesis was performed using Invitrogen SuperScript III reverse transcriptase and Oligo dT primers according to manufacturer's instructions at 55° for 1 hour followed by digestion with RNase H for 20 minutes at 37°. Finally, the reactions were cleaned using the Qiaquick (Qiagen) pcr purification system and eluted with 50 µl of water. Primary pcr reactions were performed with 10% Invitrogen PCRx enhancer solution and Platinum Taq polymerase using 10–20% of the first strand reaction as template in a total reaction volume of 20 µl in thin wall MicroAmp (Applied Biosystems) reaction tubes. Nested pcr reactions used 1 µl of the primary reaction as template. Primers for Dux4-fl and -s detection in biopsy and cultured cell samples were 14A forward and 174 reverse, nested with15A (or 16A) forward and 175 reverse. Primers for 3′ detection were 182 forward and 183 reverse nested with1A forward and 184 reverse. All primer sequences are listed in Table 5.
Dux4-fl and -s in adult human tissues were detected using 14A forward and 183 reverse, then nested with 15A forward and 184 reverse primers. Pcr cycling conditions were as follows for both primary and nested pcr: 94° 5 minutes denaturation, 35 cycles of 94° for 30″, 62° for 30″ and 68° for 2.5 minutes or 1 minute depending on expected length of product. A single final extension of 7 minutes at 68° was included. Pcr products were examined on 2% NuSieve GTG (Lonza) agarose gels in TBE.
To assess for stochastic expression of DUX4 in affected muscle cells, FSHD primary myoblasts were trypsinized and collected at confluence or after differentiation for 96 hr. Cells were counted and split into pools of 100-cell, 600-cell, or 10,000-cell aliquots. RNA was extracted from individual aliquots using Dynabeads mRNA DIRECT Kit (Invitrogen) following manufacturer's instructions. Bound polyadenylated mRNA was used directly for reverse transcription reaction with SuperScript III using on-bead oligo dT as primer. Synthesis was carried out at 52°C for 1 hr, terminated at 70°C for 15 min, followed by 15 min of RNase H treatment. 2 uL of cDNA product was used for nested DUX4-fl3′ PCR as described above.
Pcr reactions were performed on RT reactions generated as described above and using nested primer sets to sequences in exons 1 and 2 that are common to alleles on chromosomes 4 as well as 10. Transcripts were detected using primers 1A and 187 followed by nesting with 138S and 188 (Table 5). Diagnostic polymorphisms (underlined) in the 5′ end of exon 2 were used to assign allele origins of transcripts:
For quantitative PCR, 1 ug of DNase'd RNA was used for first strand cDNA synthesis. Reverse transcription was performed as above, except at 52°C for the synthesis reaction followed by 15 minutes of RNase H treatment and the Qiaquick purification eluted in 30 µl of water. One round of PCR reactions were performed using the same reagents as above and 2 uL of purified cDNA template. Primers for full length detection were 92 forward and 116 reverse (Table 5). PCR cycling conditions were as follows: 95°C 5 min denaturation, 36 cycles of 95°C for 30″, 62°C for 30″ and 68°C for 1 min, and final extension of 5 min at 68°C. Sequence of the product matched DUX4. A standard curve for DUX4 template copies was generated from PCR reactions using the same primers and cycling conditions but with known dilutions of a plasmid containing full length DUX4 cDNA in water. Test sample PCR reactions and standard PCR reactions were run in triplicate and examined on the same 1% agarose/TBE gels stained with SYBR Gold (Invitrogen) for 40 min per manufacturer instructions. Fluorescence was detected with Typhoon Trio Multi-mode Imager (GE Healthcare): excitation laser 488 nm; emission filter 520DP 40, PMT 500 V, 100 µm resolution. Histogram analysis was performed to ensure no signals were saturated. Gel band intensities were quantified with ImageQuant TL v2005 (GE Healthcare) software. Estimates for the copies of DUX4 full length template in the test samples were interpolated from the line of best fit of the dilutional standards, with the lowest visible dilutional signal setting the detection limit. The interpolated number was doubled to adjust for the single-stranded cDNA input in contrast to the double-stranded plasmid standard input. This resulted in an estimated copy number of DUX4 full-length per ug of total RNA. Final copy number estimates per cell were calculated based on assumptions of 100% efficient reverse transcription and 3.3 pg of total RNA per cell.
To assess for the full coding region of DUX4, three rounds of PCR were performed on cDNA, totaling 36 cycles. Conditions for each round were as follows: 95°C for 5′, 3 cycles of 95°C for 30″ and 68°C for 1′33″, 3 cycles of 95°C for 30″ and 65°C for 30″ and 68°C for 1′33″, 6 cycles of 95°C for 30″ and 62°C for 30″ and 68°C for 1′33″. 3 uL of primary PCR was used in the secondary PCR, and 3 uL of secondary PCR were used in the tertiary PCR. Primers for successive rounds of pcr (133, 134, 135, 136, 137, and 138G) are listed in Table 5.
3′ RACE was performed on total RNA using Invitrogen Gene Racer kit essentially as described. Prior to pcr with gene specific primers and the GeneRacer 3′ primers the RT reaction was cleaned using Qiaquick (Qiagen) spin columns as described above. Gene specific forward primers were 182 and 1A (nesting). Pcr products were gel purified, cloned into TOPO 4.0 (Invitrogen) and sequenced.
iPS cells were generated by forced expression of human OCT4, SOX2, and KLF4 using the retroviral vectors essentially as previously described (1). MLV vectors (pMXs-hOCT4, pMXs-hSOX2, and pMXs-hKLF4) were purchased from Addgene (www.addgene.com, Cambridge, MA) and vector preparations were generated by transient transfection of Phoenix-GP cells (2) with pCI-VSV-G and vector plasmids (1∶1 ratio), replacing the culture medium 16 and 48 hours later, harvesting and filtering (0.45 um pore size) conditioned medium after a 16 hour exposure to cells, and concentrating 50 to 100-fold by centrifugation (3). Transduction with MLV vectors was performed with polybrene (4ug/ml concentration) (Sigma-Aldrich Corp., St. Louis, MO) added to the medium. iPS cell colonies were identified by their characteristic morphology, cloned by microdissection, and expanded on irradiated mouse embryo fibroblasts (6000 rads) for further characterization. Typically, 5×104 fibroblasts cultured in DMEM plus 10% FBS were seeded to a 9.4 cm2 well on day minus 1, the medium was replaced with medium containing vectors and polybrene on day 0, and changed again to medium with DMEM plus 10% FBS on day 1. Cells were detached with trypsin and seeded to five 55 cm2 dishes on day 2 and medium changed on day 4. On day 6 cells are again detached with trypsin and 5×105 cells seeded to 55 cm2 dishes containing 7×105 irradiated mouse embryo fibroblasts (6000 rads) in human ES cell culture medium (see below). Medium is replaced every other day and colonies with typical morphology of iPS cells appear between day 20 and day 30 post infection. Colonies are mechanically dissected using drawn Pasteur pipettes and seeded to mouse embryo fibroblast feeder layers for culture and passaged every 2–3 days using 2 u/ml dispase.
iPS cells and Human ES cells were grown in a solution of DMEM∶F12 (1∶1) with 3.151 g/L glucose, supplemented with L-Glutamine (Invitrogen), non-essential amino acids (10 mM (100×) liquid, Invitrogen, # 11140-076), sodium pyruvate (100 mM (100×), liquid, # 11360-070), 20% knockout serum replacer (# 10828010) (Invitrogen, Carlsbad, CA), 1mM beta-mercapto-ethanol (Sigma, St. Louis MO), and 5 ng/ml basic fibroblast growth factor (Peprotech, #AF-100-18B ). Cells were generally cultured in 0.1% gelatin coated dishes containing irradiated mouse embryo fibroblasts at a density of 1.3×104 cells/cm2. When cells were used as a source of RNA, DNA, or protein, they were cultured on matrigel (1∶60 dilution, BD biosciences, #356234) coated dishes in medium conditioned by exposure to confluent layers of mouse embryo fibroblasts over a 3 day period. Cells were passaged a minimum of 4 times under these conditions before DNA, RNA, or protein was harvested.
iPS cells were evaluated for the presence of tissue non-specific alkaline phosphatase activity by fixing colonies in phosphate buffered saline solution containing 0.5% gluteraldehyde, and washing ×3 in PBS. A staining buffer containing 100 mM Tris pH 8.5, 100 mM NaCL, 50 mM MgCl2, 0.1 mg/ml 5-Bromo-4-chloro-3-indolyl phosphate (xphos) and 1 mg/ml p-Nitro-Blue tetrazolium chloride (NBT) (Sigma-Aldrich, St. Louis, MO, USA) was used to detect tissue non-specific alkaline phosphatase activity. Stage Specific embryonic antigen 4 (SSEA4) was detected using mouse monoclonal MC-813-70 and goat anti-mouse FITC conjugated secondary. TRA-1-60 was detected using mouse monoclonal TRA-1-60 (Millipore, Billerica, MA), and goat anti-mouse FITC conjugated secondary (Millipore, Billerica, MA). Human NANOG was detected with a goat polyclonal flurophore (Northern Lights) conjugated antibody (NL493, R & D systems, Minneapolis, MN). Human OCT4 was detected with a rabbit polyclonal (Abcam, Cambridge, MA) and goat anti rabbit secondary conjugated with the Alexa 488 flurophore (Invitrogen, Carlsbad, CA). Cell karyotypes were determined by the University of Washington Cytogenetics laboratory.
Induced pluripotent stem cells were detached from culture dishes with dispase (2 units/ml working concentration), 2×106 cells resuspended in F12∶DMEM (1∶1 mixture) medium without supplements, and injected into the femoral muscle of SCID-Beige mice (CB17.B6-PrkdcscidLystbg/Crl Charles River, Stock # 250). Mice were maintained under biosafety containment level 2 conditions and palpable tumor masses developed approximately 6 weeks later. When a tumor mass was palpable the mice were sacrificed and tumor tissue fixed for several days in phosphate buffered saline solution containing 4% formaldehyde, and imbedded in paraffin. Sections of the tumor (5 micron thickness) were placed on slides and stained with hematoxylin and eosin using standard protocols.
Human iPS were prepared for embryoid body formation by expanding cell numbers on mouse irradiated feeder layers, detaching colonies with dispase, triturating with a Pasteur pipette, and seeding colony fragments to dense layers of mouse embryo fibroblast feeders (5×104 irradiated mef/cm2) prior to EB formation. Four days later densely grown colonies from a 55 cm2 dish were treated with dispase and gently detached by pipetting or scraping. Colony fragments were washed several times and seeded (1∶1) to Ultra Low Attachment 55 cm2 culture dishes (Corning, Corning, NY) in DMEM supplemented with 20% Fetal Bovine Serum. Every three days, EB's were allowed to gravity settle and the medium was gently removed and replaced. RNA and chromatin was harvested three weeks later for analysis.
iPS cells were grown without MEF feeders for preparation of RNA to be used in gene expression analysis. Cells were seeded to matrigel coated dishes and filtered conditioned medium from mouse embryo fibroblasts was used for culture. RNA was purified from cells using standard techniques and treated with DNAse to remove residual genomic DNA from the cells. cDNA synthesis was primed with oligo dT and reverse transcriptase. In all cases a tube was processed in parallel without the addition of reverse transcriptase to serve as a control for possible DNA contamination. The presence of RNA transcripts were detected using 28 thermal cycles with the primer pairs for OCT4, SOX2, hTERT, NANOG, KLF4, cMYC, and GAPDH indicated in Table 5. RNA was replaced with water as a negative control for the reaction.
The Chromatin Immunoprecipitation (ChIP) analysis of repressive histone modifications at the 5′-region of DUX4 was performed on primary fibroblasts, induced pluripotent stem (iPS) cells and corresponding embryoid bodies (EB) derived from unaffected individuals and FSHD patients, following a previously described protocol [3], [26]. Briefly, cells were cross-linked with formaldehyde at 1.42% final concentration for 15 min at room temperature, quenched, and sonicated to generate 500–100 bp DNA fragments. 25 µg aliquots (representing approximately 500,000 cells) of chromatin were used for each immunoprecipitation with anti-Histone H3K9me3 antibodies (Abcam) and nonimmune IgG fraction used as a mock control. After reverse cross-linking and DNA purification, the IP products were analyzed by real time PCR. The 5′-region of the DUX4 gene was analyzed using the 4q-specific D4Z4 primers, 4qHox or Q-PCR, that detect internal D4Z4 units including the last repeat unit [3]. The real-time PCR signals obtained for IP antibodies were normalized to mock control IgG and to input to account for the number of D4Z4 repeats. Data are presented as mean ± stdev and represent the results of at least three independent immunoprecipitations followed by real-time PCR analysis done in triplicates.
We generated monoclonal antibodies to the amino- and carboxy-terminus of DUX4 for this study. The full characterization of these antibodies will be published separately [10]. Briefly, the N-terminal 159 amino acids and the C-terminal 76 amino acids of DUX4 were fused to glutathione-s-transferase tags, respectively, and injected into the animals as immunogens. Mouse monoclonals were produced at the Antibody Development core facility at the Fred Hutchinson Cancer Research Center and will be commercially available. Rabbit monoclonals were produced in collaboration with and will be available through Epitomics (Burlingame, CA). Hybridoma clones were screened for specificity by ELISA, western blot and immunofluorescence in C2C12 myoblasts transfected with DUX4. The C-terminal antibodies P4H2, P2B1 and E5-5 are specific to DUX4 and do not recognize DUX4c, whereas the N-terminal antibodies P2G4 and E14-3 recognize both DUX4 and DUX4c.
For western blotting, protein lysates were prepared by resuspension in standard Laemmli buffer and sonicated briefly. Equivalent amounts of test samples were loaded onto 4–12% gradient gel and transferred to nitrocellulose membrane, which were then blocked with 5% non-fat dry milk in PBS 0.1% Tween-20. Custom monoclonal antibodies (Epitomics, Burlington, CA) raised against DUX4 were used to probe the blots and detected by ECL reagent (Pierce, Rockford, IL). Membranes were stripped and reprobed with anti-α-tubulin antibody (Sigma-Aldrich, St Louis, MO) for loading control. Immunoprecipitation was performed on samples resuspended in PBS with protease inhibitor cocktail (Roche) by incubating overnight at 4°C with pooled anti-DUX4 rabbit monoclonal antibodies bound to protein A- and G-coupled Dynabeads (Invitrogen, Carlsbad, CA). Samples were eluted directly into Laemmli buffer and analyzed on western blot as described. For immunofluorescence, cells were fixed in 2% paraformaldehyde for 7 min and permeabilized in 1% Triton X-100 in PBS for 10 min at room temperature. Cells were probed with pairs of rabbit and mouse primary antibodies raised against N- or C-terminus of DUX4 diluted in PBS overnight at 4°C. Double labeling was detected with Alexa Fluor 488 goat anti-mouse IgG and Alexa Fluor 568 goat anti-rabbit IgG (Invitrogen) at 1∶500 in PBS for 1 hr and counterstained with DAPI.
Immunohistochemistry was performed by the FHCRC Experimental Histopathology Shared Resource. Six-micron sections of OCT embedded frozen de-identified human testes tissue were sectioned and fixed for 10 minutes in 10% neutral buffer formalin. The slides were rehydrated in TBS-T wash buffer, permeablized with 0.1% triton X-100 for 10 minutes, and then endogenous peroxidase activity was blocked with 0.3% hydrogen peroxide (Dako, Carpinteria, CA) for 8 minutes. Five minute incubation in 50% acetone and 50% methanol was used for antigen retrieval on a subset of slides. Protein block containing 0.25% casein and 0.1% Tween 20 was applied for 10 minutes. Slides were incubated over night at 4 degrees C with a 1∶5 dilution of either clone E5-5 or P2B1 in a 0.3 M NaCl antibody diluent containing 1% BSA. Staining was developed using Mach2 HRP-labeled polymers (Biocare Medical, Concord, CA). The staining was visualized with 3,3′-diaminobenzidine (DAB, Dako, Cupertino, CA) for 8 minutes, and the sections were counter-stained with hematoxylin (Dako) for 2 minutes. Concentration matched isotype control slides were run for each tissue sample (Jackson ImmunoResearch).
|
10.1371/journal.ppat.1000915 | EBV Promotes Human CD8+ NKT Cell Development | The reports on the origin of human CD8+ Vα24+ T-cell receptor (TCR) natural killer T (NKT) cells are controversial. The underlying mechanism that controls human CD4 versus CD8 NKT cell development is not well-characterized. In the present study, we have studied total 177 eligible patients and subjects including 128 healthy latent Epstein-Barr-virus(EBV)-infected subjects, 17 newly-onset acute infectious mononucleosis patients, 16 newly-diagnosed EBV-associated Hodgkin lymphoma patients, and 16 EBV-negative normal control subjects. We have established human-thymus/liver-SCID chimera, reaggregated thymic organ culture, and fetal thymic organ culture. We here show that the average frequency of total and CD8+ NKT cells in PBMCs from 128 healthy latent EBV-infected subjects is significantly higher than in 17 acute EBV infectious mononucleosis patients, 16 EBV-associated Hodgkin lymphoma patients, and 16 EBV-negative normal control subjects. However, the frequency of total and CD8+ NKT cells is remarkably increased in the acute EBV infectious mononucleosis patients at year 1 post-onset. EBV-challenge promotes CD8+ NKT cell development in the thymus of human-thymus/liver-SCID chimeras. The frequency of total (3% of thymic cells) and CD8+ NKT cells (∼25% of NKT cells) is significantly increased in EBV-challenged chimeras, compared to those in the unchallenged chimeras (<0.01% of thymic cells, CD8+ NKT cells undetectable, respectively). The EBV-induced increase in thymic NKT cells is also reflected in the periphery, where there is an increase in total and CD8+ NKT cells in liver and peripheral blood in EBV-challenged chimeras. EBV-induced thymic CD8+ NKT cells display an activated memory phenotype (CD69+CD45ROhiCD161+CD62Llo). After EBV-challenge, a proportion of NKT precursors diverges from DP thymocytes, develops and differentiates into mature CD8+ NKT cells in thymus in EBV-challenged human-thymus/liver-SCID chimeras or reaggregated thymic organ cultures. Thymic antigen-presenting EBV-infected dendritic cells are required for this process. IL-7, produced mainly by thymic dendritic cells, is a major and essential factor for CD8+ NKT cell differentiation in EBV-challenged human-thymus/liver-SCID chimeras and fetal thymic organ cultures. Additionally, these EBV-induced CD8+ NKT cells produce remarkably more perforin than that in counterpart CD4+ NKT cells, and predominately express CD8αα homodimer in their co-receptor. Thus, upon interaction with certain viruses, CD8 lineage-specific NKT cells are developed, differentiated and matured intrathymically, a finding with potential therapeutic importance against viral infections and tumors.
| We show that the average frequency of total and CD8+ NKT cells in PBMCs from 128 healthy latent EBV-infected subjects is significantly higher than in 17 patients with acute lytic EBV infection, 16 EBV-associated HL patients, and 16 EBV-negative normal subjects. The frequency of total and CD8+ NKT cells is remarkably increased in the lytic EBV-infected patients at year 1 post-onset. EBV-challenge promotes total and CD8+ NKT cell development in the thymus and liver of human-thymus/liver-SCID chimeras, compared to those in the unchallenged chimeras. After EBV-challenge, a proportion of NKT precursors diverges from DP thymocytes, develops and differentiates into mature CD8+ NKT cells in thymus in EBV-challenged human-thymus/liver-SCID chimeras or reaggregated thymic organ cultures. Thymic EBV-infected dendritic cells are required for this process. IL-7 is an essential factor for CD8+ NKT cell differentiation. EBV-induced CD8+ NKT cells produce remarkably more perforin, and predominately express CD8αα homodimer. CD8 lineage-specific NKT cells are developed and differentiated intrathymically upon EBV-exposure, a finding with potential therapeutic importance against viral infections and tumors.
| NKT cells are unconventional T cells that bridge the innate and adaptive immune systems [1]–[4]. Unlike conventional T cells, which recognize MHC-molecule-presented peptide antigens via their αβTCR, NKT cells recognize CD1d-presented glycolipids. Two subsets of functionally distinct CD1d-dependent NKT cells have been identified based on whether the cells express the semi-invariant Vα24-Jα18 TCR (Vα14-Jα18 in mice) [1], [2], [5]–[12] and whether they recognize the exogenous NKT cell ligand α-GalCer. Other NKT-like cells have been reported based on their CD1d-independence and CD161 (NK1.1 in mouse) or CD56 expression [12]–[16], or other semi-invariant Vα7.2-Jα33/Vβ2,13 TCR expression (Vα19/Vβ6,8 in mouse) [12].
In mice, conventional αβT cell development in the thymus proceeds through three major stages, i.e. CD4−CD8− (DN), CD4+CD8+ (DP), and CD4+CD8− or CD4−CD8+ (SP) [17]. The developing αβT cells undergo positive and negative selection based on TCR affinity of MHC expressed on antigen presenting cells. By contrast, the semi-invariant αβTCR DP NKT precursors interact with the CD1d-ligand complex either on cortical thymocytes to undergo positive selection [1]–[2], or on thymic dendritic cells (DCs) to undergo negative selection [18]. Positively selected DP NKT cell precursors mature by down-regulating CD8 to reach a CD4+CD44lo stage [1]–[2]. Unlike conventional T cells, which emigrate from the thymus as naïve cells, CD44lo NKT cells remain in the postnatal thymus and undergo a linear differentiation program including the expression of the terminal differentiation marker NK1.1 [19], [20]. However, a proportion of the immature NKT cells remains NK1.1− and leaves the thymus [19], [20]. The final NKT-differentiation step takes place in both thymus and periphery [21], [22]. Peripheral NKT cells reside preferentially in the liver [23], [24], but are also present in the spleen, lymph nodes, bone marrow, lung, and gut [1], [2]. Human NKT cells have not been detected in engrafted fetal thymus tissue in a hu-thy/liv-SCID model, leading to a presumption that the development of peripheral NKT cells is thymus independent [25]. In later studies, it was proposed that the human thymus has little or no role in generating peripheral NKT cells after birth. This hypothesis is based on the inverse correlation between NKT cell frequency in fetal thymus and gestational age, and on the lack of a clear NKT cell population in postnatal thymus but their definite presence in adult blood [26]–[28]. However, reports on the origin of human CD8+ Vα24+TCR NKT cells are still controversial.
In mice, it is believed that there are essentially no CD8+ NKT cells [1], [2], [29]. However, recent report shows that IL-15 expands CD8ααNK1.1+ cells [30]. In humans, the existence of CD8+ NKT cells in thymus and periphery is an area of controversy. CD8+ αβTCR NKT cells expressing CD8αα homodimer are reported in human PBMC [31]. While there are several reports questioning the existence of these cells [26], [32], it is widely believed that CD8 is expressed on a minor proportion of human NKT cells, and that the CD8 marker is usually acquired after egress from the thymus [27], [28], [33]–[37]. The finding of a limited correlation between human thymic CD4+ NKT cells and peripheral CD8+ NKT cells has raised the question of what is the origin of CD8+ NKT cells [27], [28]. As accumulation of findings on NKT cell development, the underlying mechanisms that control CD4-CD8 differentiation of human NKT cells are becoming better characterized.
We studied 177 eligible patients and subjects including 128 healthy latent EBV-infected subjects [EBV+(La)], 17 newly-onset acute infectious mononucleosis patients [EBV+(IMa)], 16 newly-diagnosed EBV-associated Hodgkin lymphoma patients [EBV+(HL)], and 16 EBV-negative normal control subjects (NS) (Table S1). None of the individuals had received treatment with anti-virals, antibiotics, or corticosteroids before entry into this study. The race of all individuals was Han as determined and registered by the physicians in this study. None of the individuals had other complicating clinical infectious symptoms when the study samples were taken. The average frequency of total NKT cells in PBMCs from the 128 EBV+(La) subjects (1.5±0.5%) was significantly higher than that from 16 EBV-negative NS subjects (0.18±0.2%), 17 new-onset EBV+(IMa) patients (0.15±0.1%) and 16 newly-diagnosed EBV+(HL) patients (0.1±0.1%) (Figure 1B). The frequency of total NKT cells in the EBV+(IMa) patients dramatically increased at year 1 post-onset [EBV+(IMy), 1.6±0.6%] (Figure 1B). The frequency of the CD8+ subset of NKT cells in PBMCs from the EBV+(La) subjects (17±4%) was remarkably higher than from EBV-negative NS subjects (2.1±0.3%), new-onset EBV+(IMa) patients (1.9±0.4%) and EBV+(HL) patients (1.1±0.2%) (Figure 1B). The frequency of CD8+ NKT cells in the EBV+(IMa) patients was significantly increased at year 1 post-onset [EBV+(IMy), 20±5%] (Figure 1B). However, the average frequencies of total T cells and the ratios of CD4+ versus CD8+ T cells in PBMCs among the EBV+(La), EBV+(HL), EBV+(IMy) and NS subjects were not significantly different (Figure 1C), except for a slight and temporary increase in the frequency of total T cells in the EBV+(IMa) patients (Figure 1C, some data not shown). These observations clearly indicate that the EBV status affects the frequency of NKT cells, particularly, the appearance of CD8+ NKT cells in PBMC.
To investigate the mechanism of total and CD8-lineage differentiation of human NKT cells in the context of EBV, we established hu-thy/liv-SCID chimeras. The chimeras were challenged i.t. with EBV, a dsDNA virus, or with human T-cell leukaemia virus type 1 (HTLV-1), a retrovirus. The EBV-challenge efficiently promoted the generation of total NKT cells, whereas HTLV-1-challenge had no effect, but instead promoted a significant increase in the frequency of αβTCR thymocytes and spleen T cells (Figure 2A). The frequency of total NKT cells reached more than 3% of thymic cells and more than 2% of hepatic cells by week 5 post-challenge with EBV. By contrast, the total thymic and hepatic NKT cells were less than 0.01% within 5 weeks post-challenge with HTLV-1, comparable to the frequencies in unchallenged chimeras (Figure 2A). The frequencies of total thymic or hepatic T cells at week 5 were slightly but significantly increased following HTLV-1 infection. There were approximately 30,000–35,000 total NKT cells per million thymic cells, and 20,000–23,000 total NKT cells per million hepatic cells at week 5 in EBV-challenged chimeras (Table S2). The EBV-challenge did not significantly alter the generation of total mainstream αβT cells, whereas HTLV-1-challenge did promote the generation of the T cells, compared with those in unchallenged chimeras (Figure 2B). The frequency of total T cells reached ∼28% of thymic cells and ∼32% of spleen cells at week 5 in the chimeras challenged with EBV or HTLV-1, as well as in the unchallenged chimeras (Figure 2B). Cell phenotyping based on CD4 and CD8 expression (Figure 2C) revealed that EBV-challenge significantly promoted the generation of thymic CD8+ NKT cells and the appearance of hepatic CD8+ NKT cells in the chimeras transplanted i.t. with total thymocytes (NKT cell-depleted) plus thymic stromal cells, a population that includes DC (Figure 2E), compared to unchallenged chimeras (Figure 2D). By contrast, HTLV-1-challenge had no effect on the frequency of CD8+ NKT cells (data not shown). The frequency of CD8+ cells in the chimeras reached more than 25% of thymic NKT cells and more than 23% of hepatic NKT cells at week 5 post-EBV challenge (Figure 2E). The CD8+ cells were essentially undetectable among thymic and hepatic NKT cells at 5 weeks in the unchallenged chimeras (Figure 2D). The different thymic or hepatic NKT cell populations (DN, CD4+, and CD8+) in the HTLV-1-challenged chimeras were comparable to those in the unchallenged chimeras (data not shown). The frequencies of total and CD4/CD8 co-receptor-expressing NKT cells in peripheral blood correlated well with those in thymus and livers in both unchallenged and EBV-challenged hu-thy/liv-SCID chimeras (data not shown). An important role for DCs in the generation of thymic and hepatic CD8+ NKT cells is suggested by the finding that the frequency of these cells was rather low in both unchallenged and EBV-challenged chimeras transplanted i.t. with total fetal thymocytes plus DC-depleted thymic stromal cells (data not shown), an issue explored further below. The different thymic and spleen co-receptor-expressing mainstream T cells (DN, CD4+CD8lo, CD4+, CD8+) in the EBV- or HTLV-1-challenged chimeras were comparable to those in the unchallenged chimeras (Figure 2D, 2E, and some data not shown). The absolute numbers of NKT cells and αβT cells (thymocytes) in the organs from various hu-thy/liv-SCID chimers were shown in the Table S2.
Thymic CD8+ NKT cells from EBV-challenged hu-thy/liv-SCID chimeras displayed an activated memory phenotype (CD69+CD45ROhi), compared with thymic CD4+ NKT cells in same chimeras (Figure S1A). Hepatic CD8+ NKT cells expressed higher amounts of CD62L than thymic CD8+ NKT cells in the EBV-challenged chimeras, probably attributable to their egress from the thymus toward secondary lymphoid organs, whereas these hepatic CD8+ NKT cells expressed similar amounts of CD69 and CD45RO as thymic CD8+ NKT cells (data not shown). CD161, a maturation marker for human NKT cells, was uniformly highly expressed on thymic and hepatic CD8+ NKT cells in EBV-challenged chimeras; the frequency of CD8+ NKT cells in the unchallenged chimeras was too low to evaluate CD161 expression. On CD4+ NKT cells, CD161 expression was independent of EBV challenge and revealed two populations, CD161hi and CD161lo, although the former population expressed much higher levels of CD161 in the EBV treated mice (Figure S1A and some data not shown). In parallel, we also examined the expression of CD69, CD62L and CD45RO on thymic and spleenic CD4+, CD4+CD8lo, CD4+CD8+, and CD8+ T cells in EBV-challenged or unchallenged chimeras. Both CD4+ and CD8+ T cells in thymus and spleen from EBV-challenged hu-thy/liv-SCID chimeras displayed an activated memory phenotype (CD69+CD45ROhi), compared with T cells from unchallenged chimeras (Figure S2B and some data not shown). Spleen CD4+ and CD8+ T cells in EBV-challenged chimeras expressed higher amounts of CD62L (data not shown) than the thymic CD4+ and CD8+ T cells (Figure S1B), which might be attributable to their egress from the thymus to secondary lymph organs.
We established different hu-thy/liv-SCID chimeras by i.t. transplantation with purified human fetal DP thymocytes plus thymic stromal cells (either DC-containing or DC-depleted). In chimeras transplanted DN thymocytes only, the frequency of total NKT cells was very low (<0.01% of thymic or hepatic cells) at week 5 in both unchallenged and EBV-challenged mice (Figure 3B, 3C). The great majority of cells were CD4-expressing in both chimeras. Less than 0.5% of thymic and hepatic CD8+ NKT cells were detected in EBV-challenged chimeras (Figure 3C). Both the frequency of total αβTCR-expressing T cells and the ratio of CD4+ to CD8+ T cells in thymus and spleen from EBV-challenged hu-thy/liv-SCID chimera transplanted with DP thymocytes only (Figure 3B, 3C) were comparable to those in EBV-challenged or unchallenged chimeras transplanted with total thymocytes (Figure 2). We further examined the ontogeny and distribution of NKT cells and T cells in hu-thy/liv-SCID chimeras transplanted i.t. with fetal DP thymocytes plus DC-containing thymic stromal cells. The frequency of total NKT cells was still rather low (∼0.02% of thymic or hepatic cells) at week 5 post-establishment in the unchallenged chimeras (Figure 3D), but was substantially increased (∼3% of thymic or hepatic cells) at week 5 post-establishment in EBV-challenged chimeras (Figure 3E). Up to 25% of thymic or hepatic NKT cells expressed CD8 in EBV-challenged chimeras, whereas the frequencies of thymic or hepatic CD4+ NKT cells were correspondingly lower than those in the unchallenged chimeras (Figure 3D, 3E). The frequency of total mature αβTCR-expressing T cells and the ratio of CD4+ to CD8+ T cells in thymus and spleen in EBV-challenged hu-thy/liv-SCID chimeras were comparable to those in unchallenged chimeras (Figure 3D, 3E). The development of thymic or hepatic CD8+ NKT cells was severely impaired by the DC-deletion. The frequency of thymic and hepatic CD8+ NKT cells was essentially below the level of detection in both unchallenged and EBV-challenged chimera transplanted i.t. with DP thymocytes plus DC-depleted thymic stromal cells (Figure 3F, 3G). We also examined the ontogeny and distribution of NKT cells and T cells in the hu-thy/liv-SCID chimeras transplanted i.t. with DP thymocytes plus purified thymic DC. The frequency of total thymic and hepatic NKT cells and the ratio of CD4+ to CD8+ NKT cells in EBV-challenged or unchallenged chimeras were comparable to those in the counterpart chimeras transplanted with DP thymocytes plus DC-included thymic stromal cells (data not shown). We further established hu-thy/liv-SCID chimeras by transplantation i.t. with fetal DP thymocytes plus i.v. syngeneic fetal BM-derived DCs. The frequency of total NKT cells was substantially increased to ∼2% of thymic or hepatic cells at week 5 post-establishment in the EBV-challenged chimeras (Figure 3I), compared to the unchallenged chimeras (∼0.01% of thymic or hepatic cells) (Figure 3H). Up to 23% of thymic and hepatic NKT cells expressed CD8 in the EBV-challenged chimeras, whereas the levels of thymic and hepatic CD4+ NKT cells were correspondingly lower than those in the unchallenged chimeras (Figure 3H, 3I). The frequency of total mature αβTCR-expressing T cells and the ratio of CD4+ to CD8+ T cells in thymus and spleen from EBV-challenged hu-thy/liv-SCID chimeras were comparable with those in the unchallenged chimeras (Figure 3H, 3I). The absolute numbers of NKT cells and αβT cells (thymocytes) in the organs from various hu-thy/liv-SCID chimers were shown in the Table S2.
We next performed various RTOC of human fetal DP thymocytes reaggregated with syngeneic fetal thymic stromal cells (thymic DC-included), purified thymic DCs, or BM-derived DCs. Various stimuli (EBV-epitopes, infectious EBV or α-GalCer) were applied during the culture. The frequency of total NKT cells was low (<0.01% of RTOC cells) in the EBV- or α-GalCer-challenged RTOC established with only DP thymocytes (Figure 4B). In RTOC of DP thymocytes reaggregated with syngeneic fetal thymic stromal cells (thymic DC-included), purified thymic DCs, or BM-derived DCs, the EBV-challenge significantly promoted the generation of total NKT cells. After 14-day-culture, the frequency of total NKT cells was up to 2.5–2.8% of RTOC cells, whereas treatment with α-GalCer had no such effect (Figure 4B). Addition of HLA-matched or unmatched EBV-epitopes (BMLF1+EBNA1) had no significant effect on the frequency of total NKT cells (<0.01% of RTOC cells) compared to the un-stimulated RTOC (Figure 4B and data not shown). In RTOCs where thymic or BM-derived DCs were present, EBV-challenge substantially promoted the development of CD8+ NKT cells. Up to 25% of NKT cells expressed CD8 in EBV-challenged RTOCs (Figure 4C). In this case, HLA-matched EBV-epitopes moderately and significantly increased the development of CD8+ NKT cells (2.5% of NKT cells), compared with those in un-stimulated RTOCs (Figure 4C). In RTOC of DP thymocytes reaggregated with DC-depleted thymic stromal cells, the EBV-induced increase in CD8+ NKT was almost completely abolished (data not shown). Both the frequency of total mature αβTCR-expressing T cells and the ratio of CD4+ to CD8+ T cells in the different RTOC conditions were comparable (Figure 4B, 4D).
To further confirm that EBV mediated intrathymic CD8-lineage differentiation of human NKT cells, we focused our attention on detecting the actual EBV- or HTLV-1-infection of human progenitor thymocytes, thymic NKT cells and thymic DCs in the virus exposed hu-thy/liv-SCID chimeras. For detection of EBV-infection, five transformation-associated EBV-genes, LMP1, EBNA1, BZLFl, BALF2, and RAZ were examined. By Southern blot and Q-PCR, a high level of viral genes and mRNA transcripts were detected in EBV-exposed human DCs in the chimeras. Since mice are not the natural EBV host and their DCs are well-known to be unsusceptible to EBV, these results indicated that EBV infected only the transplanted human DCs in EBV-exposed hu-thy/liv-SCID chimeras during NKT cell development and differentiation. There was no evidence of EBV viral genes in EBV-exposed human chimeric DP thymocytes or mature CD4+ and CD8+ NKT cells (Figure S2A). For detection of HTLV-1-infection, 2 highly conserved viral X-region DNA sequences, SK43 and SK44, were examined. By Q-PCR assay, high levels of SK43 and SK44 were detected in human chimeric HTLV-1-exposed DP αβTCR-expressing T cells (Figure S2B) as well as in chimeric hepatic T cells (data not shown). There was no detectable HTLV-1 in thymic CD4+ and CD8+ NKT cells or in DCs in the HTLV-1-exposed hu-thy/liv-SCID chimeras, indicating that HTLV-1 virus does not correlate with NKT cell differentiation in EBV-exposed hu-thy/liv-SCID chimeras.
IL-7 and IL-15 were known survival factors for T cells, and enhancers of NKT cell homeostatic proliferation [21], [26], [38], [39]. Both cytokines were used in attempts to differentiate and activate NKT cells from human peripheral and cord blood [35], [40], [41]. DP thymocytes expressed an increased level of IL-7Rα in EBV-challenged hu-thy/liv-SCID chimeras compared to unchallenged chimeras. The thymic DCs produced a considerable amount of IL-7 mRNA in unchallenged chimeras, and the levels increased substantially with EBV-challenge (Figure S3A). The thymic CD8+ NKT cells expressed a very high level of IL-7Rα mRNA in EBV-challenged hu-thy/liv-SCID chimeras compared with other types of thymic NKT cells and αβTCR-expressing T cells in both unchallenged and EBV-challenged chimeras (Figure S3A). The thymic DCs produced a considerable amount of IL-15 mRNA in both unchallenged and EBV-challenged chimeras. The thymic CD4+ and DN NKT cells expressed higher levels of IL-15Rα mRNA in both unchallenged and EBV-challenged hu-thy/liv-SCID chimeras compared with other types of thymic NKT cells and αβTCR-expressing T cells in both chimeras (Figure S3A). In a time course study, the freshly isolated fetal thymic DCs were found to express a low level of IL-7 mRNA. Thymic DCs in unchallenged chimeras expressed comparable levels of IL-7 mRNA at the different time intervals examined, 1, 3, and 5 weeks. By contrast, the thymic DCs rapidly increased the expression of IL-7 mRNA by week 1 post-EBV challenge, and maintained high levels throughout the course of the analysis (Figure S3B). The IL-15 mRNA was uniformly expressed in the thymic DCs of both unchallenged and EBV-challenged chimeras (Figure S3B). Thymic stromal cells (DC-depleted) expressed a uniformly low level of IL-7 and IL-15 mRNA in both unchallenged and EBV-challenged chimeras (data not shown). These observations on cytokine and cytokine receptor mRNA expression were confirmed at the protein level by intracellular flow cytometry (for IL-7 and IL-15) and conventional flow cytometry (for IL-7Rα and IL-15Rα) (data not shown). Thus, the thymic DCs are a major source of IL-7 during the thymus-dependent development of NKT cells.
The frequency of total NKT cells were low (<0.01% of FTOC cells) after 14-days of culture without adding any stimuli in FTOC (Figure 5B). Nearly all of the NKT cells were CD4-positive and CD8+ cells were undetectable in these conditions (Figure 5C). By contrast, in FTOC with added EBV, the frequency of total NKT cells was increased (1.5% of FTOC cells), and of these, 15% of cells expressed CD8, whereas, in FTOC with added HLA-matched EBV-epitopes, neither total nor CD8+ NKT cells were changed (<0.01% of FTOC cells, of which <1% were CD8+) (Figure 5C). Exogenous IL-7 or IL-15 alone slightly but significantly increased the total, but not CD8+ NKT cell differentiation in the FTOCs (<0.01% of FTOC cells) (Figure 5C). HLA-mismatched EBV-epitopes were non-functional in FTOCs (data not shown). In EBV-challenged FTOCs, exogenous IL-7, but not IL-15, could significantly further enhance the total and CD8+ NKT cell differentiation (∼2.5% of FTOC cells, of which 20% were CD8+), compared to the EBV-challenged but non-IL-7-stimulated FTOC and un-stimulated FTOC (Figure 5B, 5C). A mAb against IL-7 completely abolished the effect of IL-7 on the differentiation of CD8+ NKT cells in the EBV-challenged FTOCs, indicating an essential role of the cytokine in the differentiation of CD8+ NKT cells. In FTOCs containing added α-GalCer, IL-7 slightly enhanced total, but not CD8+ NKT cell differentiation (∼1% of FTOC cells, of which <0.8% were CD8+) (Figure 5B, 5C). The mAb against IL-7 completely inhibited the effect of IL-7 on the differentiation of total NKT cells in the FTOCs stimulated with α-GalCer. IL-15 had no such effect on the development of total NKT cells in the FTOCs stimulated with α-GalCer. The frequency of total mature αβTCR-expressing T cells, but not ratios of CD4+ to CD8+ T cells, was enhanced by IL-7 or/and IL-15 in the various FTOCs (Figure 5B, 5C).
Consistent with the above in vitro findings in the FTOCs, administration of exogenous IL-7 further enhanced development of thymic and hepatic total and CD8+ NKT cells in the in vivo EBV-challenged hu-thy/liv-SCID chimeras (Figure 6B), compared to chimeras given exogenous IL-7 but not EBV (data not shown) and to EBV-challenged chimeras not treated with exogenous IL-7 (Figure 2E). The frequency of total NKT cells was significantly increased (∼3.8% of thymic and ∼3.2% hepatic cells) (Figure 6B). About 28% of thymic and hepatic NKT cells expressed CD8 (Figure 6B and Table S2). The administration of mAb against IL-7 (plus exogenous IL-7) completely blocked the function of IL-7 in vivo (Figure 6C and Table S2), indicating an essential role of the cytokine in the EBV-induced development of CD8+ NKT cells. The administration of exogenous IL-7 plus isotype-matched control Ab had no the blocking effect (Figure 6D and Table S2). Thus, EBV-induced increase in CD8+ NKT cell development is IL7-dependent.
To track NKT cells more accurately, we applied 6B11 mAb, which recognized TCR Vα24JαQ junction CDR3-loop, combined with mAb against Vα24TCR for gate of NKT cells by flow cytometry. The outcome of frequencies of total and co-receptor-expressing NKT cells gated by either CD1d tetramers vs. anti-αβTCR mAb or by anti-Vα24 mAb vs. 6B11 mAb were comparable in different human normal subjects or EBV-infected patients (Figure S4A). Comparable sets of data on frequencies of total and co-receptor-expressing NKT cells were also obtained in thymus and liver from EBV-challenged hu-thy/liv-SCID chimeras (Figure S4B), as well as in EBV-exposed RTOCs and FTOCs (data not shown), gated by either CD1d tetramers vs. anti-αβTCR mAb or anti-Vα24 mAb vs. 6B11 mAb by flow cytometry. These data confirmed the observations on EBV-induced development of CD8+ NKT cells, and ruled out possible contamination of activated conventional CD8+ T cells during the flow cytometry analysis.
In our recent study [42], frequencies of CD8+ NKT cells in patients with EBV-associated malignancies were found significantly lower than those in healthy EBV carriers. CD8+ NKT cells in tumor patients were functionally impaired in terms of cytokine production and cytotoxicity. In hu-thy/liv-SCID chimeras, EBV-exposure efficiently augmented the generation of IFN-γ-biased CD8+ NKT cells, which were strongly cytotoxic to EBV-associated tumor-cells. IL-4-biased CD4+ NKT cells were predominately generated in unchallenged chimeras, which were non-cytotoxic. In tumor-transplanted hu-thy/liv-SCID chimeras, adoptive transfer with EBV-induced CD8+ NKT cells remarkably suppressed tumorigenesis by EBV-associated malignancies. CD4+ NKT cells were synergetic with CD8+ NKT cells, leading to a more pronounced T-cell anti-tumor response in the chimeras co-transferred with CD4+ and CD8+ NKT cells. In the present study, we further investigated the perforin expression in NKT cells (Figure 7). CD8+ NKT cells from healthy EBV+ humans, EBV-challenged hu-thy/liv-SCID chimeras, or EBV-exposed RTOCs and FTOCs produced much higher amount of perforin (Figure 7B) than that in counterpart CD4+ NKT cells (Figure 7A), indicating that high production of perforin in EBV-induced CD8+ NKT cells was an additional reason for their high cytotoxicity to EBV-associated tumor cells, besides their biased IFN-γ-production [42]. In the analysis, we applied flow cytometry using the gating of either CD1d tetramers vs. anti-αβTCR mAb and anti-Vα24 mAb vs. 6B11 mAb. Two groups of results were comparable (Figure 7).
Moreover, we further analyzed the CD8α and CD8β expression on NKT cells. Data revealed that CD8αα homodimer was predominately expressed on CD8+ NKT cells in PBMCs from healthy latent EBV-infected subjects and IM patients at year 1 post-onset, as well as from normal control subjects (Figure 8B and 8C), which were consistent with the previous reports [31]. CD8αα homodimer was also expressed in the majority of CD8+ NKT cells in thymus and liver from hu-thy/liv-SCID chimeras challenged i.t. with EBV (Figure 8D and 8E) and in EBV-exposed RTOCs and FTOCs (data not shown). In the analysis, we applied flow cytometry using the gating of either CD1d tetramers vs. anti-αβTCR mAb and anti-Vα24 mAb vs. 6B11 mAb. Two groups of results were comparable (Figure 8).
Taken advantage of the hu-thy/liv-SCID chimeric mouse model [43], [44], we have found that a sizable CD8+ fraction (up to 25%) of total human thymic NKT cells is generated in vivo after EBV-challenge. The development of CD8+ NKT cells is promoted in the thymus at the DP precursor stage, and requires participation of thymic DCs. CD4 versus CD8 lineage commitment is controlled by the EBV-challenge. The findings provide a crucial access point for unraveling the mechanism for NKT cell development and differentiation. This study led to two important insights. First, we provide direct evidence that certain pathogens, EBV in this case, are important contributors to CD8-lineage commitment of NKT cells. Second, we demonstrate that differential CD4 versus CD8 lineage commitment can be controlled not only by some known classical endogenous elements [1], [2], but also by exogenous pathogenic element(s) such as EBV.
The impact of different viral pathogens on NKT cell frequencies has been investigated. In humans, infection by HIV and HTLV-1 results in a decrease in NKT cells [45]–[52]. In mice, LCMV induces long-term loss of NKT cells since induction of apoptosis [53], [54]. The patients with severe immunodeficiency (XLP), lacking of NKT cells, is characterized by an extreme sensitivity to EBV infection [55], [56]. Our previous [42] and current works show that EBV-infection promotes generation of IFN-γ- and perforin-biased CD8+ NKT cells, and IL-4-biased CD4+ NKT cells. A protective role of CD8+ NKT cells synergized with their counterpart CD4+ NKT cells against EBV-associated malignancies has been verified. The beneficial role against the persistence of EBV-infection could be speculated. The observation on induction of dominate populations of CD8αα+ NKT cells by EBV-infection is in agreement with previous report on that CD8αα+ NKT cells control expansion of total and EBV-specific T cells in humans [31], and is supporting the observation in mice [30].
There remain several controversial issues concerning CD8+ NKT cell development. For example, why does the frequency of human CD8+ NKT cells show such a limited correlation between different sites (thymus and blood) and among different ages (fetus, neonate and adults), and why are CD8+ NKT cells the most numerically variable NKT cell subset in humans, particularly under pathophysiological circumstances [1], [2], [26]–[28], [33]–[37]. In the present study, we were able to monitor the intrathymic and extrathymic development of human NKT cells in different organs in hu-thy/liv-SCID chimeric mice. Since all mature NKT cells were depleted from the thymocytes prior to cell transplantation (Figure S6), any co-receptor-expressing human NKT cells detected in the mice should have developed and differentiated post-cell-transplantation. Since all fetal samples have been from non-EBV-infected mothers, the EBV-challenge in this animal model accurately reflects the viral effects on the differentiation of CD8+ NKT cells. Nevertheless, more evidences are needed to rule out the possibility that EBV is capable of influencing NKT cell expansion/differentiation in the periphery.
Given the functional distinction between CD4+ and CD8+ NKT cells [26]–[28], [33]–[37] and their potential therapeutic importance such as in cancer treatment [42], it is crucial to identify the factors that induce the development and differentiation of these cells in order to fully understand the causes of NKT cell subset deficiency and dysfunction, particularly of the CD8+ NKT cells. Some investigators hold the opinion that the immature NKT cells undergo extrathymic differentiation in adult blood [27], [28], [33]–[37]. Our studies have demonstrated that the frequency of total NKT cells and the different subsets in thymus, liver and peripheral blood from unchallenged and EBV-challenged chimeras is highly correlated, clearly indicating an intrathymic developmental and differentiation step for human CD8+ Vα24+NKT cells. Moreover, the hu-thy/liv-SCID chimeras have provided an in vivo model to investigate cell development and differentiation of human NKT cells under both physiological and pathophysiological circumstances.
In mice, IL-15 plays an essential role in the maturation and overall population size of NKT cells in the thymus and periphery [21], [22], [30]. On the other hand, IL-7 is critical for the development of NKT cells, but plays a minor role in regulating their maturation and homeostasis [21]. In humans, IL-7 dominates the CD4+ NKT cell development process in the fetus, neonate, and adult [26], [39], whereas IL-15 has a selective and age-specific role in vitro in the expansion and homeostasis of the DN and CD8+ NKT cell subsets [27]. We show here that IL-7 is a major and essential enhancer of EBV-induced development of thymic CD8+ NKT cells in vivo, in the hu-thy/liv-SCID chimeras, and in vitro in FTOCs. We still need to define the role of IL-7 in the continuous NKT cell division in the periphery of adults, if it indeed exists, for instance, in secondary lymphoid organs where IL-7 is available.
Taking the fact that EBV causes asymptomatic life-long infection in ∼90% of adults worldwide into consideration, the experimental design for developmental studies of NKT cells should pay special attention to the EBV status of the donors of any human samples, since we have shown here that EBV-infection status of the hu-thy/liv-SCID chimeras and the human donors can directly affect the frequency of total and co-receptor-expressing populations of NKT cells. More importantly, the present study has raised several interesting questions, such as how the semi-invariant canonical αβTCR is expressed on DP thymocyte precursor before commitment to the CD4 versus CD8 lineage differentiation of NKT cells, as well as what and how the ligand is presented by thymic DCs to the semi-invariant αβTCR-expressing DP thymocyte precursor causing the preferential CD8 differentiation.
The latent EBV-infected [referred to as EBV+(La)] or normal control subjects (NS) were healthy EBV seropositive or seronegative individuals, respectively. The patients with EBV-associated acute infectious mononucleosis [lytic phase, referred to as EBV+(IMa)] were diagnosed by a monospot test and the detection of capsid-specific serum IgM [56], and followed-up at 1 year [latent phase, referred to as EBV+(IMy)]. The patients with EBV-associated Hodgkin lymphoma (HL) were diagnosed according to the WHO criteria, and staged according to the Ann Arbor classification (Table S1). All EBV+(La) and NS individuals were healthy volunteers. All patients eligible for this study were in- or out-patients in different Hospitals in Hubei Province in China. HLA typing was performed using the Lymphotype Class I system (Biotest) and an Olerup SSP kit (GenoVision). The clinical information of all patients and healthy EBV-infected and normal control subjects is listed in Table S1. All patients were newly-diagnosed and had no previous treatment before entry into this study. All patients provided informed consent according to the institutional guidelines and protocol titled “The study on the frequency and subset distributions of human peripheral NKT cells in normal and EBV-infected subjects” that was approved by The Wuhan University Ethical Committee. The written informed consent from each patients and subjects was obtained.
Human fetal thymic cells, bone marrow (BM) cells, liver and PBMCs were anonymously obtained from voluntarily elective pregnancy terminations (<24-wk-gestation; HLA typing matched HLA-A2 and HLA-DRB1(*03), the most prevalent HLA-types for Eastern and Southern Chinese populations, and mismatched HLA-A11, -B8 and HLA-DQ5). The mothers were excluded if lytic and latent EBV- and HTLV-1-infections were detected by Q-PCR and serologic determination [57]. Thymic cells, BM cells and PBMCs were isolated, aliquoted, cryopreserved and maintained in the vapor phase of liquid nitrogen for further use. Viability of thawed cells was evaluated by Trypan blue dye exclusion before use. Thymic dendritic cells were separated from the thymocytes by adhesion onto plastic culture dishes. For transplantation, NKT cells were positively depleted from thymic cells by MACS beads based on staining with α-GalCer-loaded CD1d tetramers [58]. For functional studies, NKT cells were purified from human PBMCs or chimeric thymic cells by flow cytometry cell sorting or a MACS bead system based on staining with α-GalCer-loaded CD1d tetramers [58], [59]. Synthesized peptides (proteins) were EBV-epitopes, GLCTLVAML (HLA-A2-restricted, derived from the lytic cycle protein BMLF1), AVFDRKSDAK (HLA-A11-restricted, derived from nuclear antigen EBNA3B), RAKFKQLL (HLA-B8-restricted, derived from the lytic cycle protein BZLF1), TSLYNLRRGTAL (HLA-DRB1-restricted, derived from nuclear antigen EBNA1), SDDELPYIDPNM (HLA-DQ5-restricted, derived from nuclear antigen EBNA3C) [60]. Recombinant peptides (proteins) were verified free of pyrogenicity (endotoxin <10 units/ml, no bacterial or fungal contamination) according to the certifications from the manufacturer. The α-GalCer-loaded CD1d tetramers were synthesized as previously described [58]–[60]. For preparation of viral stocks, a highly productive EBV-producer cell line P3HR-1 (American Type Culture Collection, ATCC, Manassas, VA) was treated with 12-O-tetradecanoyl-phorbol-13-acetate (TPA, 30 ng/ml) for 14 days. The virus was then pelleted from the culture supernatant. The residual TPA in the viral suspension for final use had no significant promoting effect on cell proliferation in the in vivo human-thymus-SCID chimeras, based on our preliminary experiments. Recombinant human (rh) IL-7 (Roche) and rhIL-15 were purchased from R&D Systems. All mouse anti-human monoclonal antibodies were purchased from BD PharMingen, San Diego, CA, USA, except mAbs against human Vα24 or Vβ11, which were from Immunotech, Marseille, France.
To establish the human-thymus/liver-SCID (hu-thy/liv-SCID) chimeras, 8-wk-old female SCID mice (NOD/LtSz-prkdcscid/prkdcscid strain, the Jackson Laboratory) were irradiated (300 cGy/mouse) prior to cell-transplantation. Human fetal thymic cells were depleted of immature and mature NKT cells based on their reactivity with α-GalCer-loaded CD1d tetramers. Then, 1×107 thymocytes, thymocytes: thymic stromal cells including dendritic cells = 1∶1 (Figure S5), were transplanted into the thymus of anaesthetized SCID mice [43], [44], [58], [59]. Syngeneic human fetal liver tissue (equivalent to 1×107 fetal liver cells) was simultaneously implanted under the mouse kidney capsule, unless otherwise noted. The chimeras were then intrathymically challenged with EBV (107 pfu) [60] or HTLV-1 (107 pfu), and the challenge was repeated after 6 days. The chimeras were maintained for 4 wks, unless otherwise stated [43], [44]. In some cases, chimeras were established by transplantation with human fetal thymic cells (thymocytes plus thymic stromal cells), but without implantation of fetal liver tissue referred to as human-thymus-SCID (hu-thy-SCID chimera). The mice were housed in a pathogen-free environment in the Animal Research Institute, Wuhan University. The protocol for animal study titled “The study on the frequency and subset distributions of NKT cells in human-thymus/liver-SCID chimeras” was approved by The Wuhan University Ethical Committee in accordance with the current Chinese laws.
FTOC was carried out as described previously [61]. Briefly, fetal thymus tissue was dissected into pieces of ∼2 mm3. Three pieces of tissue were placed into 24-well plates with culture medium containing various stimuli as indicated. On day 7, the cultured thymus fragment was dispersed into a single-cell suspension, and cells were stained and analyzed by flow cytometry. RTOC experiments were performed as previously described [61]. Briefly, thymic stromal cells were prepared by disaggregating fetal thymic lobes. DP thymocytes were obtained by gently grinding freshly fetal thymus lobes. The resulting suspensions were sorted for DP thymocytes using CD4 and CD8 labeling. Reaggregates were formed by mixing together the desired thymic stromal cells and DP thymocytes at 1∶1 cell ratio with other stimuli as indicated. After pelleting the cells by centrifugation, the cell mixture was placed as a standing drop on the upper membrane surface, and incubated for 5–12 days.
The α-GalCer-loaded CD1d tetramer and αβTCR (Immunotech, clone BMA031) was used to define total NKT cells. For tetramer staining, the cells were incubated with the tetramer labeled with fluorochromes at 37°C for 15 min. The appropriate isotype Ab (αβTCR mAb isotype mouse IgG2b) and empty CD1d tetramer conjugated with a fluorochrome was used to establish negative staining gates. The representative experiments for NKT cell gate negative staining were illustrated in Figure 1 and Figure S6. The αβTCR and other relevant mAbs were used to identify the different subsets of T cells. In some cases, mAb against human CDR3 loop of invariant TCR Vα24 (6B11, Immunotech) and mAb against human Vα24 (including isotype controls) were used for gating NKT cells. For analysis of co-receptor-expressing NKT cells, single cell suspensions were stained with mAbs to human CD4 and CD8α (R&D Systems, clone 11830 and 37006, isotype mouse IgG2a and IgG2b), unless otherwise noted. In some cases, NKT cells were stained with mAbs to CD8α and CD8β (Abcam, clone 2ST8.5H7, isotype mouse IgG2a), simultaneously. In intracellular staining for detection of perforin, different cells were resuspended in cold Dulbecco's PBS, and then permeabilized by Cytofix/Cytoperm solution (15 min, 4°C, in the dark; BD Pharmingen) according to the manufacturer's protocol. These permeabilized cells were stained with mAb specific for human perforin (FITC-conjugated G9, mouse IgG2b, BD Pharmingen), or isotype control, and analyzed by flow cytometry. All analyses were performed with a FACSCalibur (BD Biosciences). Four- and five-color analysis was done using CellQuest software.
All Q-PCR reactions were performed as described elsewhere [62]. Briefly, total RNA from purified cells (1×104, purity >99%) or cell lines was prepared by using Quick Prep® total RNA extraction kit (Pharmacia Biotech) according to the manufacturer's instructions. RNA was reverse transcribed by using oligo (dT)12-18 and Superscript II reverse transcriptase (Life Technologies, Grand Island, USA). The real time quantitative PCR was performed in special optical tubes in a 96 well microtiter plate (Applied Biosystems, Foster City, CA) with an ABI PRISM® 7700 Sequence Detector Systems (Applied Biosystems). By using the SYBR® Green PCR Core Reagents Kit, fluorescence signals were generated during each PCR cycle via the 5′ to 3′ endonuclease activity of AmpliTaq Gold to provide real time quantitative PCR information. Primers used in Q-PCR are listed in Table S3.
Statistical analyses were performed using the Student′s t test. Values of p<0.05 were considered statistically significant.
|
10.1371/journal.pcbi.1002047 | Integrative Analysis of Transgenic Alfalfa (Medicago sativa L.) Suggests New Metabolic Control Mechanisms for Monolignol Biosynthesis | The entanglement of lignin polymers with cellulose and hemicellulose in plant cell walls is a major biological barrier to the economically viable production of biofuels from woody biomass. Recent efforts of reducing this recalcitrance with transgenic techniques have been showing promise for ameliorating or even obviating the need for costly pretreatments that are otherwise required to remove lignin from cellulose and hemicelluloses. At the same time, genetic manipulations of lignin biosynthetic enzymes have sometimes yielded unforeseen consequences on lignin composition, thus raising the question of whether the current understanding of the pathway is indeed correct. To address this question systemically, we developed and applied a novel modeling approach that, instead of analyzing the pathway within a single target context, permits a comprehensive, simultaneous investigation of different datasets in wild type and transgenic plants. Specifically, the proposed approach combines static flux-based analysis with a Monte Carlo simulation in which very many randomly chosen sets of parameter values are evaluated against kinetic models of lignin biosynthesis in different stem internodes of wild type and lignin-modified alfalfa plants. In addition to four new postulates that address the reversibility of some key reactions, the modeling effort led to two novel postulates regarding the control of the lignin biosynthetic pathway. The first posits functionally independent pathways toward the synthesis of different lignin monomers, while the second postulate proposes a novel feedforward regulatory mechanism. Subsequent laboratory experiments have identified the signaling molecule salicylic acid as a potential mediator of the postulated control mechanism. Overall, the results demonstrate that mathematical modeling can be a valuable complement to conventional transgenic approaches and that it can provide biological insights that are otherwise difficult to obtain.
| Cellulose-based biofuels presently offer the most environmentally attractive and technologically promising alternative to fossil fuels. To be viable, biofuels must be derived from non-food crops, such as grasses, wood, bark, and plant residues. Techniques for releasing the energy stored in these renewable materials must first untangle a very recalcitrant scaffold of interlinking molecules inside the plant cell walls, which is very costly. Much of the recalcitrance is due to the natural polymer lignin, which hardens the cell walls and is composed of three different building blocks, called monolignols. Modern transgenic techniques have yielded plant lines whose cell walls are easier to break down, but some of these modified plants have exhibited unexplained and undesired features. Here, we present new computational methods for analyzing monolignol biosynthesis in unprecedented detail. The analysis simultaneously accounts for lignin biosynthesis in various transgenic lines and different developmental stages and yields six novel, testable postulates regarding the metabolic control of the pathway. The results suggest new, targeted experiments towards a better understanding of monolignol biosynthesis and issues of recalcitrance reduction. More generally, the results highlight the genuine benefits of using computational methods as companions and complements to experimental studies.
| The complex, interwoven structure of lignin, cellulose, and hemicellulose polymers in plant cell walls is the main cause for the recalcitrance of lignocellulosic feedstocks to microbial and enzymatic deconstruction towards fermentable sugars. This recalcitrance, in turn, accounts for the high cost of biofuel production from renewable sources [1]. In current technologies, the release of polysaccharides from the entanglement with lignin demands a thermo-chemical pretreatment that is expensive and has undesirable side effects during the later fermentation steps. Recent efforts aimed at decreasing the lignin content with transgenic techniques suggest that it might be feasible to reduce or even obviate the need for pretreatment [2], which would permit the inclusion of polymer separation in downstream biomass processing technologies [3] and thereby make the cost of biofuel production competitive with that of fossil fuels.
Reflecting the substantial impact of lignin on forage digestibility [4], pulping efficiency [5] and sugar release from biomass [2], considerable effort has been devoted towards a better understanding of monolignol biosynthesis in situ. Most pertinent genes have been identified in model species with complete sequence information, and knowledge from these genomes is currently being used for homology searches in species where sequencing efforts are ongoing. Such homology investigations are often effective, but caution is necessary, because multiple genes with similar sequences and annotations pointing to the same enzyme may possess distinct expression patterns and substrate preferences [5]. As a consequence, monolignol biosynthesis in vivo might be strikingly different among species and depend not only on gene sequences, but also on the tissue or even cell type of interest. Thus, before genetic modification strategies that had proven effective in some species are implemented in another species, it is prudent to consider and account for contextual differences. As a case in point, alfalfa (Medicago sativa L.), the organism used for our analysis, exhibits substantial differences in cell wall composition among the different internodes of young plants.
In most woody plants, the biochemical pathway of monolignol biosynthesis leads to three building blocks of lignin, which are known as p-hydroxyphenyl (H), guaiacyl (G) and syringyl (S) monolignols (Figure 1). In potential bioenergy crops like poplar and switchgrass, lignin consists principally of G and S units, while H units are present in low to negligible quantities. In other plants, including some alfalfa transgenics, H can be present in significant amounts. Although the generic sequences of metabolic reactions within the monolignol pathway have been identified, it is becoming increasingly clear that critical details of the pathway structure and its regulation are not entirely understood. As a case in point, Chen et al. [6] recently introduced systematic, transgenic alterations in alfalfa (Medicago sativa L.) plants by independently modifying the activities of seven key enzymes of monolignol biosynthesis. While many of the results were easily explained, down-regulation of caffeoyl coenzyme A 3-O-methyltransferase (CCoAOMT) had little effect on S lignin, an observation that is conceptually inconsistent with the commonly accepted pathway structure (Figure 1; black colored arrows). A recent study identified two isoforms of cinnamoyl CoA reductase (CCR), MtCCR1 and MtCCR2, in Medicago truncatula [7]. Furthermore, an earlier finding had suggested that caffeyl aldehyde is one of the preferred substrates for caffeic acid 3-O-methyltransferase (COMT) in alfalfa [8]. Taken together, these findings could imply an alternative route for S lignin synthesis (Figure 1; red colored arrows) upon CCoAOMT down-regulation [8], [9]. However, they cannot explain why only G lignin is decreased because feruloyl-CoA is a common precursor of both G and S lignin.
In dicotyledonous plants like alfalfa, the stem consists of many segments, called internodes. During maturation, all internodes grow asynchronously and thus independently represent different developmental stages. This phenomenon suggests a customized modeling approach: Instead of studying the pathway within a single developmental context, it seems advantageous to launch a systematic investigation that simultaneously encompasses dozens of internodes from seven wild-type or transgenic plants. This comprehensive approach supports the fact that lignin biosynthesis is tightly coordinated by a hierarchy of transcription factors during secondary wall thickening [10]. It also circumvents the potential problem that regulatory mechanisms might escape discovery during an analysis based on singular phenotypic datasets, such as lignin content and monomer composition, if only one internode or one transgenic line is studied at a time. This potential failure to detect regulatory signals is exacerbated in the lignin system by the fact that several enzymes in the pathway catalyze multiple steps, which makes intuitive analyses difficult.
With a comprehensive analysis of several datasets as the target, we propose here a novel modeling approach that integrates the data in a semi-dynamic fashion. First, flux balance analysis (FBA) [11] is applied independently in each individual internode of the wild-type plant. In contrast to microbial systems, where maximization of the growth rate is usually assumed to be the species' overall objective, we use the monolignol production as the objective function for FBA. Second, for every internode of a lignin-modified line, we use the method of minimization of metabolic adjustment (MOMA) [12] to characterize the altered flux distribution in relation to the corresponding FBA solution for the same wild-type internode. Specifically, the relative proportions of the fluxes leading to three lignin monomers are constrained at experimentally-observed values to improve the prediction. Finally, we perform a Monte Carlo-like simulation of randomly parameterized kinetic models in cases where the results arising from the static models may have alternative, kinetics-based explanations.
This combined modeling approach represents, to the best of our knowledge, the first computational study of lignin biosynthesis in angiosperm stem tissues and, more generally, of secondary plant metabolism in angiosperms. As we will discuss later, the model analysis resulted in six postulates concerning the metabolic control of monolignol biosynthesis that had not been considered at all or at least not in detail. These postulates address the reversibility of some enzymatic reactions, shed light on the hypothesis of independent pathways for the synthesis of G and S monolignols, and suggest a novel feedforward regulatory mechanism exerted by a cinnamic acid-derived compound. Of note is the fact that evidence in support of this last postulate has subsequently been obtained in laboratory experiments. By critically evaluating the transgenic data against a revised pathway structure in alfalfa, we hope these postulates will not only serve as guidelines for directing future experiments, but also provide mechanistic insights that will aid the design of combined genetic modification strategies toward the generation of bioenergy crops with reduced recalcitrance.
Accounting for recent experimental observations, we adopted a revised pathway structure of monolignol biosynthesis in alfalfa stems that includes the CCR2-catalyzed reduction of caffeoyl-CoA to caffeyl aldehyde and the subsequent synthesis of coniferyl aldehyde by COMT (Figure 1: black and red colored reactions), as explained earlier.
The pathway of monolignol biosynthesis contains a fairly small number of branch points, and it is known that flux partitioning at these branch points determines the ultimate transport fluxes v6, v15 and v19 and thus the relative amounts of lignin monomers (cf. [13]). The FBA-derived steady-state flux analysis for wild-type plants supports this argument. It suggests that variation in lignin composition from young to mature internodes is accomplished by modulating the flux partitioning at three principal branch points: p-coumaroyl-CoA, coniferyl aldehyde, and coniferyl alcohol. As a paradigm illustration, the proportion of H lignin declines from 7% of the total monomer yields in the first two internodes to 1% in the eighth internode. This decline is singularly achieved through a monotonic decrease in v4 (Figure 2A). A parallel increase in the ratio of S to G lignin—commonly termed the S/G ratio—from 0.09 in the first two internodes to 0.64 in the eighth internode requires a combined effort of flux adjustments at coniferyl aldehyde and coniferyl alcohol (Figure 2B). Since F5H controls the first committed steps (i.e., v16 and v20) towards the synthesis of S lignin, one would expect to see its expression being up-regulated in mature versus young internodes, which has recently been validated by microarray analysis (Table 4 of [14]).
For a systemic analysis of the pathway we used the results of a gene modification study in alfalfa where genes encoding for PAL, C4H, HCT, C3H, CCoAOMT, F5H, and COMT were independently down-regulated. With the exception of F5H-modified lines, which did not permit measurements of the targeted enzyme activity, we applied MOMA to each strain and each internode and predicted the new steady-state flux distribution (see Materials and Methods).
A very interesting result is the fact that no feasible solution exists for four of the six transgenic plants, if the revised metabolic map is correct (Figure 1; black and red colored arrows). For example, if C4H activity is down-regulated to 45% of its wild-type level, it is analytically impossible to derive a set of fluxes that satisfies the mass balance at cinnamic acid as well as the observed lignin composition, if the supply of phenylalanine is constant. To remedy this situation, it seems to be necessary to add to the pathway structure three “overflow” fluxes counteracting the potential accumulation of the intermediate metabolites cinnamic acid, p-coumaryl aldehyde, and 5-hydroxyconiferyl alcohol (blue arrows v22, v23, v24 in Figure 1). This proposed amendment is at least partially supported by observations. First, salicylic acid (SA), an essential signaling molecule for systemic acquired resistance against pathogen attack, can be formed from cinnamic acid [15], [16], [17], although it may also originate from the shikimate pathway via isochorismate [18]. Second, the biosynthesis of all flavonoids begins with the condensation of p-coumaroyl-CoA and three molecules of malonyl-CoA by the enzyme chalcone synthase [19]. And third, incorporation of 5-hydroxyconiferyl alcohol into lignin polymer is found in the COMT-deficient alfalfa [20]. Thus, we included these additional effluxes, and the expanded system (Figure 1; v1 to v24) permitted feasible solutions in all cases tested.
In wild-type plants, the FBA-derived steady-state values of the three added fluxes are minimized to prevent lignin precursors from being channeled into peripheral pathways producing SA or flavonoids. In the transgenic plants, these auxiliary fluxes are no longer restricted to small values and thus can be raised to substantial levels to facilitate the re-distribution of fluxes. However, the assumption that the peripheral fluxes are minimized in wild-type plants must be handled with caution: although the phenylpropanoid pathway in cells undergoing secondary wall thickening may evolve towards maximizing the synthesis of lignin precursors, this is apparently not the case when biosynthesis of flavonoid-derived products, which may function as floral pigments or as anti-microbial agents, becomes the plant's top priority.
The MOMA analysis revealed flux distributions for all transgenic lines and their individual internodes. Figure 3 shows the developmental evolution of flux patterns in CCoAOMT-deficient plants; similar plots for other transgenic plants are given in Figures S1, S2, S3, S4, and S5. Of note is that all computed fluxes exhibit strong and essentially monotonic trends: for each transgenic line, the flux partitioning at important branch points follows clear trends throughout the internodes rather than jumping in value from one internode to the next. This result is surprising and encouraging, because MOMA simply assumes that the fluxes undergo a minimal re-distribution when the pathway system is perturbed. Because these perturbations occur independently for each internode, there is no mathematical guarantee that individual fluxes would follow any smooth trend from internode to internode. In other words, the collective results, while fitting into the context of a gradual change in lignification pattern during stem development, are by no means “automatic,” because no external constraints or conditions were imposed or enforced on the transition from one internode to the next. The computed trends are summarized in Table 1.
The following paragraphs are structured as follows. First, we re-evaluate the gene knock-down data in a systematic way across different stages of growth and formulate four postulates that actually do not require a full model analysis, but emerge from the “logic” of the pathway. Second, we discuss two postulates regarding novel mechanisms of metabolic regulation that result from our comprehensive model analysis. Third, we present new experimental results that directly support one of the model-based postulates.
The total lignin production is driven by the availability of phenylalanine rather than by enzymatic limitations. This conclusion results from the observation that the down-regulation of PAL has much less effect on total lignin content and/or lignin composition in young internodes with small amounts of lignin than in mature internodes with high lignin production (Table S3 in Text S1; [6]). Expressed differently, PAL is not acting at capacity when the demand for lignin is relatively low, as is the case in young internodes. This conclusion is also supported by the observation that lignin production is not enhanced proportionately when PAL enzyme is over-expressed in transgenic plants [21].
In transgenic plants where C3H is down-regulated, the proportion of H lignin among total monomer yields is significantly increased over control plants, especially in mature internodes (Figure 4A). This finding is at first puzzling, because it is unlikely that the cell can detect changes in C3H activity and adapt accordingly by exerting appropriate flux control at an earlier branch point (i.e., p-coumaroyl-CoA) within the network. Arguably the simplest explanation is that HCT (possibly along with other plant acyltransferases) is reversible [22]. If so, the following scenario is possible: as p-coumaroyl-shikimate accumulates due to a reduced C3H activity, HCT converts it back to p-coumaroyl-CoA in the presence of free CoA, thereby allowing the cell to escalate the production of H lignin beyond the wild-type level. The catalytic efficiency of HCT acting on p-coumaroyl-shikimate as substrate remains to be experimentally determined, along with the possible competition for CoA between two shikimate esters (i.e., p-coumaroyl-shikimate and caffeoyl-shikimate).
The hypothesis of HCT being reversible prompts us to investigate whether C3H, which controls the material flow between two HCT-catalyzed steps, also permits catalysis in both directions. A slightly increased proportion of H lignin in CCoAOMT-deficient plants (Figure 4A) seems to suggest that C3H is mildly reversible and that part of the accumulated caffeoyl-CoA is therefore converted back to p-coumaroyl-CoA and subsequently channeled towards H lignin, a scenario which seems unlikely based on the known catalysis by cytochrome P450 enzymes. However, the amounts of H lignin determined by thioacidolysis appear to be unaffected by the low CCoAOMT activity despite a noticeable decrease in total lignin content (Table S3 in Text S1; [6]). One plausible explanation is that thioacidolysis yields are highly correlated with the in vivo abundance of S lignin [5], which might suggest that plants may in effect produce more H lignin than was measured against the down-regulation of CCoAOMT.
If both HCT and C3H are reversible, the two CCR-catalyzed reactions—v10 and v13—can be regarded as the “committed” steps (i.e., they are essentially irreversible), because manipulation of any downstream enzyme, such as COMT and F5H, has no substantial effect on H lignin (Figure 4B). Interestingly, the postulate seems to echo the conclusion from a previous enzyme assay [23]: CCR purified from poplar stems was able to catalyze the conversion of coniferaldehyde into feruloyl-CoA in the presence of other co-factors but preferentially reduced CoA-esters, as judged by the calculated equilibrium constants.
In addition to a modest increase in H lignin, down-regulation of CCoAOMT leads to a noticeable increase in the S/G ratio of all internodes except for internodes 1 and 2 (Figure 5A). This finding is puzzling because coniferyl aldehyde is a common precursor to both S and G lignin and one would therefore expect a similar effect on both. The analogous situation arises in COMT-deficient plants, where the S/G ratio is reduced (Figure 5A). This case, however, is not quite as clear-cut because COMT also shows activities towards downstream intermediates like 5-hydroxyconiferyl aldehyde and 5-hydroxyconiferyl alcohol. Thus, in this case of COMT deficiency, the S/G ratio might not be a good indicator of the flux partitioning at coniferyl aldehyde towards G and S lignin.
As an explanation for the altered S/G ratios in cases of CCoAOMT or COMT down-regulation, we postulate that the enzymes controlling v12 and v16 (and maybe even v10 and v17) are organized into a functional complex through which the intermediates are channeled without much leakage. Similarly, we postulate that v13 and v14 form a corresponding complex without much leakage. This dual postulate for crossing channels is supported indirectly by literature information and by findings from our flux analysis, as outlined below.
First, an analysis of mature stems (internodes 6–9) collected from CCoAOMT down-regulated transgenic lines indicated that the levels of G lignin were greatly reduced, whereas those of S lignin were nearly unaffected (cf. CCOMT antisense line ACC305 in Table 1 of [24]). Similarly, down-regulation of CCR1, which actively catalyzes the subsequent reduction of feruloyl-CoA to coniferyl aldehyde, also resulted in an increased S/G ratio in mature internodes of alfalfa stems [25], again with G lignin being more strongly reduced than S lignin. Although the existence of the CCR2-COMT pathway helps sustain the lignin content in either CCoAOMT or CCR1 down-regulated lines, the findings do not explain why S lignin is synthesized at the expense of G lignin upon genetic modifications of the CCoAOMT-CCR1 pathway. Nevertheless, the findings are entirely consistent with the postulate of crossing channels.
Second, one of the constituent enzymes, F5H, is localized to the external surface of the endoplasmic reticulum [26], so that the proposed channel may exist in the form of an enzyme complex anchored in the endomembrane. Indeed, a labeling experiment in microsomes extracted from lignifying alfalfa stems suggested such a co-localization of COMT and F5H [27]. It showed that caffeyl aldehyde, when incubated with [methyl-14C]-labeled S-adenosyl L-methionine (a co-substrate necessary for COMT-mediated O-methylation) and NADPH (the reducing agent for F5H), is converted to coniferyl aldehyde, 5-hydroxyconiferyl aldehyde, and a small amount of sinapyl aldehyde.
Finally, our flux distribution analysis reveals a strong correlation between the computed flux values of v13 and v14 for all but the CCoAOMT-deficient plants (Pearson correlation coefficient ρ = 0.9952; p-value<0.001) (Figure 6). This correlation suggests that there is normally almost no exchange of products between v12 and v13, and that most of the coniferyl aldehydes produced through the CCR2-COMT shunt are directly utilized by F5H without having the opportunity of diversion into G lignin biosynthesis. A notable exception seems to be the situation where CCoAOMT is significantly down-regulated. In this case, caffeoyl-CoA tends to accumulate at least in the short term, thus providing the CCR2-COMT pathway and the associated metabolic channel with an abundance of substrate. The predicted flux distribution (Figure 3) and the observed lignin composition (Table S3 in Text S1) indicate that CCoAOMT-deficient plants produce a considerable amount of G lignin, although the levels of S lignin are comparable to those in the controls, which implies that only some of the extra caffeoyl-CoA can be converted efficiently into S lignin through the proposed channel. Overall, the proposed functional channels seem to be consistent with results of the flux analysis as well as with earlier discussions in the literature [8], [9]. The correlation between v12 and v16 is less pronounced, which is presumably due to the fact that F5H and COMT catalyze parallel pathways, with the latter (v20 and v21) buffering changes in earlier precursors.
An alternative explanation for an increased S/G ratio upon modifications of the CCoAOMT-CCR1 pathway could be that the kinetic features of the enzymes that catalyze coniferyl aldehyde and coniferyl alcohol are fine-tuned such that they could permit the adjustment of fluxes leading to G and S lignin and thus change the S/G ratio. For instance, given that down-regulation of CCoAOMT or CCR1 may alter the intracellular level of coniferyl aldehyde, the relative values of v14 and v16 at steady state could depend on whether the respective enzyme works within the linear or saturation region of its kinetic profile.
To investigate this alternate hypothesis, we designed and analyzed a kinetic Michaelis-Menten model that contains the two alternative pathways from caffeoyl-CoA to coniferyl aldehyde as well as the two principal branch points where the fluxes leading to G and S lignin diverge (see Text S1). The model was simulated 10,000 times with randomly sampled kinetic parameter values, as described in Materials and Methods and Text S1, and we recorded the percentage of admissible parameter sets that yielded a significantly increased S/G ratio in response to a 80% reduced CCoAOMT or CCR1 activity.
We first examined the case where CCoAOMT is down-regulated. Only ∼5% of all admissible systems (see Text S1 for definition) yielded a significantly increased S/G ratio, whereas nearly half of all systems resulted in an S/G ratio that differed by less than 5%. The few cases of significant increases in the S/G ratio did not reveal particular patterns, which may not be too surprising because the system involves 16 kinetic parameters that affect each other in a nonlinear fashion. Intriguingly, for the scenario of CCR1 down-regulation, none of the admissible systems showed a significant increase in S/G ratio; in fact, all changes in S/G ratios were less than 0.5%. Replacing the Michaelis-Menten kinetics with cooperative Hill kinetics allowed more flexibility. Still, only ∼3% of all admissible systems exhibited an increase in S/G ratio upon CCR1 down-regulation. Taken together, it seems that, theoretically, some precisely tuned sets of kinetic parameters could lead to the observed effects on the S/G ratio. However, these sets are rare and do not seem to be robust enough to render the kinetics-based hypothesis viable.
One of the most paradoxical findings among the collective results from the transgenic plants is the opposite effect on lignin composition (and specifically the S/G ratio) when either PAL or C4H is down-regulated. It seems that these alterations should not differentially affect monolignol biosynthesis, because both occur before the first branch point, but they do. Closer inspection of the data from different internodes reveals that the S/G ratio is consistently increased in PAL-deficient plants but decreased in C4H-deficient plants (Figure 5B). While experiments with tobacco have suggested that the differential co-localization of PAL isoforms and C4H might be the underlying cause of such observations [28], there is as yet no direct evidence for this intracellular association in alfalfa or other related legume species.
In accordance with the proposition of separate metabolic channels for G and S lignin, we postulate that the different effects of PAL or C4H down-regulation on the S/G ratio are due to feedforward regulation. Specifically, we suggest that this regulation is mediated by a downstream product of the cinnamic acid degradation pathway, which is represented collectively as v22 in Figure 1. Notice that this feedforward regulation had not been recognized by the scientific community and was postulated by the model analysis purely with computational means.
Consistent with the observation of all transgenic experiments, an appropriate control strategy by this unknown compound X is summarized in Figure 7 and discussed below. In the case of PAL-deficiency, where the biosynthesis of cinnamic acid from phenylalanine declines, a diminished pool of X could directly or indirectly reduce the expression of CCoAOMT/CCR1/CAD and/or activate the expression of CCR2/COMT/F5H, thereby altering the channeling towards G and S lignin and increasing the S/G ratio. Intriguingly, this proposed inhibition of CCoAOMT expression following PAL down-regulation is supported by a strong correlation of the proportion of G and S lignin in total monomer yields in internodes 4–8 of the PAL- and CCoAOMT-deficient plants (Figure 8).
In the case of C4H deficiency, however, the production of X through v22 is likely to increase because the consumption of cinnamic acid through a competing branch v2 is not as effective as in wild-type plants. Thus, an accumulation of X could in turn activate the expression of CCoAOMT/CCR1/CAD and/or reduce the expression of CCR2/COMT/F5H, leading to a smaller S/G ratio.
Salicylic acid (SA) is a notable endogenous signaling molecule that is known to be derived from cinnamic acid [29]. Down-regulation of one pathway enzyme other than C4H (e.g. HCT [30]) had recently been shown to lead to elevated levels of SA. To investigate whether SA is the postulated signaling compound X, we measured its intracellular levels in many independent transgenic alfalfa lines in which different monolignol biosynthesis genes had been down-regulated. Indeed, the results show that the intracellular levels of SA are highly proportional to the extent of lignin reduction (Figure 9). Based on our postulated feedforward regulation, this effect can be explained through the participation of SA in the inhibition of the metabolic channel committed to S lignin biosynthesis, thus reducing the total lignin content.
Functional genomics is a premier tool for identifying metabolic pathways in sequenced model species and for pinpointing genes involved in them [31]. However, it is known that many enzymes coexist in multiple isoforms with unique expression patterns and substrate specificities. A pertinent example seems to be the recent discovery of two CCR isoforms with distinct catalytic properties towards major CoA-esters in Medicago [7]. Steady-state flux analysis of an extended pathway system that accounts for the isoforms reveals that the alternative path is dispensable in wild-type plants, but that it may rise to significant levels in specific transgenic lines. Indeed, CCoAOMT-deficient plants support a much higher lignin production than lines where HCT or C3H is down-regulated (Table S3 in Text S1; [6]). The intricate differences in pathway operation among otherwise very similar transgenic lines point to the need of investigating flux patterns not only in different plants, but also in different strains, lines and even different internodes and tissues. The results shown here furthermore demonstrate that subtle variances among tissues and lines are difficult to discern with intuition alone, but that computational analyses can serve as objective and rigorous tools for explaining such differences.
Specifically, the new integrative modeling approach proposed here combines static flux-based models and a Monte Carlo simulation of randomly parameterized kinetic models. This approach has the advantage that it allows the collective analysis of many experimental results and sheds light on pathway features that are particularly important for functionality under normal and altered conditions. The analysis here revealed a quantitative trend of flux patterns during development, which in turn allowed the identification of principal branch-point metabolites at which internode-specific flux partitioning patterns control the observed mode of lignification. While it is relatively easy to single out principal metabolites in linear or slightly branched pathways, the system studied here is confounded by the plant's employment of the same enzymes, such as CCR and CAD, in different key positions. Due to this multiple use, manipulating the flux partitioning pattern towards a desired mode of lignification may incur undesired “side effects.”
The computational analysis indicates that a single flux analysis just for wild-type plants is insufficient for understanding because even a seemingly simple pathway like monolignol biosynthesis requires relatively minor, yet important, extensions to account for the overflow of some intermediate metabolites that only occurs in transgenic plants. At the same time, the analysis also demonstrates that the simultaneous analysis of several independent datasets, in this case transgenic lines and sequential internodes, can lead to insights that otherwise would have been difficult to obtain. Here, it led to several postulates that are specific enough for experimental validation or refutation.
Some model-free postulates refer to the need for reversibility or committedness of key reactions, which might not be too surprising. Two further postulates are more intriguing. They refer to the functional channeling within the pathway and its mechanistic control. Based on the observation of an increased S/G ratio in CCoAOMT or CCR1 down-regulated lines, the computational results suggest an S lignin-specific channel capable of converting caffeyl aldehyde directly into 5-hydroxyconiferyl aldehyde or sinapyl aldehyde. Different experiments in the literature suggested the co-localization of COMT and F5H in lignifying alfalfa stems [27] and the localization of F5H to the external surface of the endoplasmic reticulum [26]. These and our findings would imply the likely location for a functional S-channel complex to be associated with the endomembrane.
While the proposed membrane-bound channel for synthesizing S lignin could constitute an important control mechanism, it may only have comparatively limited capacity because even in CCoAOMT down-regulated lines G lignin is generated in a higher proportion of total monomer yields than S lignin (Table S3 in Text S1; [6]). One likely cause is that different O-methyltransferases (OMTs) are involved in converting caffeyl aldehyde to coniferyl aldehyde. These OMTs may have distinct sub-cellular localization (to cytoplasm or endomembrane) and therefore a different affinity to F5H. Thus, it could be that the cytosolic OMT in the transgenic lines with reduced CCoAOMT expression is up-regulated and helps consume extra caffeyl aldehyde outside the proposed channel. A corresponding labeling experiment in alfalfa [27] confirmed that only a small proportion of total cellular COMT activity against caffeyl aldehyde is associated with the microsomal membrane, and that adding excess recombinant COMT has little effect on the metabolism of caffeyl aldehyde by microsomes.
To examine whether the observed increase in the S/G ratio upon modifications of the CCoAOMT-CCR1 pathway could be explained alternatively by a kinetically-controlled mechanism, we generated 10,000 ODE models for a reduced pathway system (Text S1) and simulated both down-regulation schemes. Among all sampled parameter sets, only a minute percentage of systems had the ability to increase their S/G ratio significantly in either case. Although the results neither reject the possibility of a kinetically-controlled S/G ratio nor directly corroborate our channeling postulate, they do suggest that purely kinetic control might be unlikely, because it would require rather precise implementations of specific parameters in different tissues, which seems to compromise the robustness of the system. As shown in a structural study of alfalfa COMT [32], mutations of some key residues lining the active site result in significantly different substrate binding and/or turnover rate. Moreover, it is likely that the kinetic properties of other enzymes may also exhibit a similar, if not more severe, susceptibility to genetic perturbations (e.g., [33], [34]). Since the variation in the S/G ratio is typically small (s.d.≈0.03 in two control lines; [6]), the proposed functional channeling mechanisms seem to offer a more robust option to help maintain a physiologically proper S/G ratio.
The observed decrease in the S/G ratio of COMT down-regulated lines alone is not sufficient to prove the existence of a G lignin-specific channel, because a reduced COMT activity affects all fluxes that are specific for the synthesis of S lignin, thus leading to a smaller S/G ratio. Nevertheless, the strong correlation between v13 and v14 that emerged from our computations for most transgenic experiments lends further credence to such an inference. This correlation not only supports the operation of a G lignin-specific channel, but also hints at the possibility of CCR1 and CAD (and maybe CCoAOMT) being complexed or co-localized on internal membranes.
One option for testing this postulate would be to down-regulate CCR2 and record if the strain exhibits a greater decrease in S lignin than in G lignin, giving a smaller S/G ratio. Surprisingly, knocking out CCR2 in M. truncatula, a species closely related to alfalfa, leads to an increased S/G ratio, whereas M. truncatula CCR1 knock-out mutants show a reduction in the S/G ratio [7]. However, in spite of their close taxonomic relatedness, the operation and control of monolignol biosynthesis might be quite different in tetraploid alfalfa (M. sativa L.) and diploid M. truncatula. For instance, the S/G ratio in wild-type alfalfa stems (0.62; internodes 1–8) is approximately twice as large as that in wild-type M. truncatula stems (0.29; internodes 1–7). Consequently, further experimental work is required to validate or reject the postulate that a G lignin-specific channel is operational in alfalfa.
If the postulates of specific channels towards the synthesis of G and S lignin are valid, one may further surmise that the opposite effects of PAL or C4H down-regulation on lignin composition are the results of differential gene or enzyme expression, which could be mediated by a cinnamic acid derivative. However, the model could not identify this molecule, leading us to call it Compound X. Supporting this hypothesis, the transgenic experiments used here have shown that down-regulation of CCoAOMT, which we postulate to be involved in the G lignin-specific channel, yields similar proportions of G and S lignin among total monomers as does the down-regulation of PAL, which is postulated to inhibit and/or activate the functioning of the G lignin- and S lignin-specific channels, respectively (Figure 8).
Salicylic acid (SA), a phenolic phytohormone derived from phenylalanine, was proposed as a potential candidate for this unknown Compound X. Intriguingly, post-hoc experiments showed that the intracellular levels of SA are indeed highly proportional to the extent of lignin reduction in transgenics where different pathway genes are down-regulated (Figure 9). This result fits directly into the context of our feedforward control postulate. At the same time, it makes us wonder why putting a block on monolignol biosynthesis could affect the homeostasis of SA, especially if the blockage is located away from the pathway entrance. Based on previous findings that SA can be derived both from cinnamic acid and from isochorismate via the shikimate pathway [29], and that HCT uses shikimate as a preferred cofactor (Figure 10), we propose the following scenario: when the flux going through the pathway is decreased due to some genetic manipulation, fewer shikimates will be trapped in those shikimate esters (p-coumaroyl-shikimate and caffeoyl-shikimate) and thus become available to make SA. In other words, the shikimate recycling facilitated by HCT enables the shikimate pool to work as a sensor of the flux into lignin. Future in-depth studies, whether they are experimental or computational, are required to justify this hypothesis. It is noteworthy, however, that the reason why plants shuttle monolignol pathway intermediates between Coenyme A and shikimate esters has yet to be explained.
In conclusion, our analysis shows that a combined modeling effort can uniquely and effectively complement experimental studies of the type used here. In contrast to analyzing one dataset at a time, it allowed us to integrate all results from a comprehensive experimental investigation of various transgenic lines and internodes. This integration, in turn, revealed dynamic, developmental patterns and their dependence on key enzymes. Together, the analyses uncovered elusive control of monolignol biosynthesis and led to testable hypotheses regarding various pathway aspects that should be clarified before one attempts to generate and optimize viable, productive “designer” crops with minimal recalcitrance.
In a previous study [6], lignin content and composition were analyzed in transgenic alfalfa plants in which seven enzymes were independently down-regulated (cf. Figure 1). These enzymes were: L-phenylalanine ammonia-lyase (PAL), cinnamate 4-hydroxylase (C4H), hydroxycinnamoyl CoA:quinate/shikimate hydroxycinnamoyl transferase (HCT), coumarate 3-hydroxylase (C3H), caffeoyl coenzyme A 3-O-methyltransferase (CCoAOMT), ferulate 5-hydroxylase (F5H), and caffeic acid 3-O-methyltransferase (COMT). Each transgenic plant was cultivated to early flowering stage, and the mature stem consisting of eight internodes was harvested and divided into individual segments; all internodes were numbered according to their maturity, with internodes 1–2 representing the pooling of the two uppermost stem segments. The lignin content and monomer composition for each internode were determined for each transgenic line via established protocols [6]; the results are summarized in Table S3 in Text S1. The activities of all targeted enzymes were also measured and summarized elsewhere, with the exception of F5H, which showed no activity towards any documented substrates when assayed in crude alfalfa extracts in vitro (Table 2c of [6]). Thus, the F5H-deficient line is excluded from the following analysis.
|
10.1371/journal.pcbi.1003136 | New Universal Rules of Eukaryotic Translation Initiation Fidelity | The accepted model of eukaryotic translation initiation begins with the scanning of the transcript by the pre-initiation complex from the 5′end until an ATG codon with a specific nucleotide (nt) context surrounding it is recognized (Kozak rule). According to this model, ATG codons upstream to the beginning of the ORF should affect translation. We perform for the first time, a genome-wide statistical analysis, uncovering a new, more comprehensive and quantitative, set of initiation rules for improving the cost of translation and its efficiency. Analyzing dozens of eukaryotic genomes, we find that in all frames there is a universal trend of selection for low numbers of ATG codons; specifically, 16–27 codons upstream, but also 5–11 codons downstream of the START ATG, include less ATG codons than expected. We further suggest that there is selection for anti optimal ATG contexts in the vicinity of the START ATG. Thus, the efficiency and fidelity of translation initiation is encoded in the 5′UTR as required by the scanning model, but also at the beginning of the ORF.
The observed nt patterns suggest that in all the analyzed organisms the pre-initiation complex often misses the START ATG of the ORF, and may start translation from an alternative initiation start-site. Thus, to prevent the translation of undesired proteins, there is selection for nucleotide sequences with low affinity to the pre-initiation complex near the beginning of the ORF. With the new suggested rules we were able to obtain a twice higher correlation with ribosomal density and protein levels in comparison to the Kozak rule alone (e.g. for protein levels r = 0.7 vs. r = 0.31; p<10−12).
| Gene translation is an important step of the intra-cellular protein synthesis, which is a central process in all living organisms. Thus, understanding how translation efficiency is encoded in transcripts has ramifications to every biomedical discipline. The aim of the current study is to decipher the way translation initiation fidelity is encoded in eukaryotic transcripts, and how evolution shapes the beginning of transcripts. Based on the genomes of dozens of organisms we were able to derive a new, more precise, set of rules related to this process, facilitating a high resolution view of the mechanisms aiding translation initiation fidelity. Among others, we show that there is a universal trend of selection for low numbers of ATG codons upstream, but also in the 5–11 codons downstream of the START ATG, presumably to prevent translation of alternative ORFs over the main one. With the new suggested rules we were able to obtain a twice higher correlation with ribosomal density and protein levels in comparison to the previous translation initiation efficiency rule.
| Gene translation is the central cellular process of sequence decoding to produce a protein. This process occurs in every organism and consumes most of the cellular energy [1]–[3], thus it has important ramifications to every biomedical field [3]–[9]. Translation consists of three stages: initiation (the binding of the ribosome to the transcript and the association of the small and large subunits), elongation (the iterative translation of triplets of nucleotides to amino acids by the ribosome) and termination (the disassociation of the large and small subunits of the ribosome and the completion of the process), which form a recurring cycle of events.
In eukaryotes the initiation step usually involves formation of a pre-initiation complex (consisting of the small subunit, 43S or the 40S subunit, and initiation tRNA). According to the accepted scanning model [10]–[14], this complex accompanied by additional initiation factors scan the mRNA sequence starting from its 5′ end towards its 3′ end, until a start codon is recognized (usually an AUG that is identified by the initiation tRNA), which represents the beginning of the open reading frame (ORF). The recognition of the start codon triggers the association of the large subunit and the beginning of the elongation step [10]–[13]. However, ATG codons are expected to be present in all possible reading frames upstream and downstream the START of the ORF; how thus does the scanning pre-initiation complex recognize the start ATG?
Over 30 years ago Kozak suggested that a specific context (i.e. the nucleotides before and after a codon) surrounding the initiating ATG codon is required for its recognition by the pre-initiation complex; asserting that this context should appear only in the vicinity of the initiating (START) ATG codon of the ORF [10]. Nevertheless, several deviations from the aforementioned scanning model have been reported based on small scale experiments. For example, there are reported cases of leaky scanning where AUG codons with sub-optimal contexts are skipped and translation initiates at a downstream AUG [14], and there are cases of translation via internal ribosome entry site (IRES) and additional non-canonical mechanisms [15]–[17], but these have been reported to be relatively rare. In addition, though there is a relatively low number of ATG codons in the 5′UTR as expected by the scanning model [18], it was shown that there are many cases with non-optimal ATG context scores for the main ATG codon, and cases of ATG codons at the 5′UTR with relatively optimal context scores [19], [20], suggesting that the scanning model is an over-simplification of the reality.
Previous genomic studies [10]–[13], that were usually based on a relatively small number of transcripts and\or organisms, were aimed at understanding the nucleotide pattern surrounding the START ATG, with few considering alternative ATGs, and if so usually focusing on those in the 5′UTR and not the ORF. Specifically, amongst others, it was shown that there are a number of preferred nucleotide sequences surrounding the main translation initiation codon in eukaryotic genomes, which may vary among organisms [21], [22]; there is a relation between sequence features and context scores at the beginning of the ORF and alternative initiation START codons [23], [24]. In addition various papers studied alternative ORFs in the 5′UTR (uORFs) and their effect on translation regulation and re-initiation [17], [25]–[28], the effect of mRNA folding on translation and its initiation [17], [29]–[33], the relation between alternative ATGs near the beginning of the ORF and protein localization [34], and the length of the 5′UTRs [27], [35].
As mentioned, the fact that there are less AUGs at the 5′UTR [18], and the effect of uORFs on the main ORF downstream, have been previously suggested and are related to the scanning model [17], [25]–[28]. However, we are the first to quantify the 5′UTR (and ORF) region under selection for less ATGs, and the fraction of the protein levels variance that can be explained by these facts.
The aim of the current study is to infer new universal rules related to the way in which initiation efficiency and fidelity are encoded in transcripts, to quantify potential alternative translation initiation events, and discern the selection processes maintaining the integrity of translation initiation and the resultant protein product. To this end, we analyse large scale genomic data of dozens of eukaryotes. Based on this analysis we reformulate and refine the translation initiation rules, and improve the understanding of the biophysics of initiation and the evolution of transcripts. Specifically, we show for the first time that there is selection for less ATG codons downstream to the beginning of the ORF; we estimate the length of the region under such selection upstream (5′UTR) and downstream the beginning of the ORF; we report some additional sequence signals related to initiation fidelity such as anti-‘Kozak’ sequences surrounding ATG codons near the beginning of the ORF, and the appearance of stop codons close to them, that are under selection; and we are the first to quantify the partage of the protein levels and ribosomal density variance that can be explained by the different signals near the beginning of the ORF.
First, we re-studied the distribution of nucleotide allocation near the START ATG, focusing on Saccharomyces cerevisiae. As was reported in previous studies [21], we find that in most of the positions near the START ATG, the distribution of the nucleotide composition is relatively close to uniform in S. cerevisiae genes (Figure 1A; for example in position −1 the probability to see A/T/C/G is 0.45/0.21/0.18/0.16 respectively, and in position +1 the probability to see A/T/C/G is 0.3/0.27/0.13/0.3 respectively).
The meaning of the above result is that if we consider the nucleotide with the highest probability (e.g. A in positions ±1), there is a very high probability to see a different nt in this position (probability 0.55 and 0.7 respectively); even if we consider the two nt with highest probability (AT in position −1 and AG in position +1), there is still relatively high probability to see a nt different to these two (probability 0.34 and 0.4 respectively).
However, as previously reported [10], [36], [37], some of the positions (e.g. −3; i.e. 3 nt before the beginning of the ORF) exhibit a relatively non-uniform nucleotide distribution (Figure 1A–B, see also supplementary Figure S37 related to Schizosaccharomyces pombe).
Throughout the paper we use the measure, (ribosomal density)·(mRNA levels) (Methods), that we named ribosomal-load as a measure of the intensity of gene translation. This measure considers total ribosomal flux over the gene transcripts. As can be seen in Figure 1B, also in the case of highly translated genes with high ribosomal load, the nucleotide distribution near the beginning of the ORF is relatively uniform (Figure 1B; Methods).
Is it possible that the nt distribution near the START ATG is related to the gene translation rate? When we computed the correlation between the optimality of the context score (a score that is based on the comparison to the context of the most highly translated genes; details in the Methods section) and the protein levels (Figure 1C), we obtained relatively low but significant correlations; similar results were obtained when we correlated the context score with ribosomal density (Figure 1D; as we demonstrate later these correlations can be slightly improved when binning the data).
Similar results were obtained in the 33 eukaryotes that were analyzed (Figure 2A). Specifically, we find that the relative entropy (Methods), a measure of conservation of a nucleotide among the organism's genes, of position −3 upstream of the START codon is the lowest; thus, nucleotide −3 is the most conserved site in all the analyzed organisms (Figure 2A), concordant with previous small scale studies [10], [36], [37]. However, as can be seen in Figure 2A, positions −1, −2, and 1 are also relatively conserved in many of the analyzed organisms, also in agreement with previous small scale studies [10], [36], [37]. Specifically, it seems that the patterns of conserved positions vary along the evolutionary tree. For example, position 1 is relatively conserved in all the vertebrates. This result may reflect co-evolution between the ribosome and the region around the start ATG. The results reported in this subsection are in agreement with previous studies and motivated us to search for additional signals that are under selection near beginning of the ORF.
In the previous section we demonstrated that there is a relatively weak signal corresponding to the START ATG in eukaryotes. This result raises the hypothesis that ATG codons near the beginning of the ORF contribute to alternative translation initiation events. If such alternative initiation events are deleterious we expect a selection for less ATGs near the beginning of the ORF.
In the rest of the paper, to analyze signals related to such a selection, we considered three reading frames: frame 0 is identical to the reading frame of the gene ORF; frames 1 and 2 represent a frame shift of 1 or 2 nucleotides relative to the main frame (Figure 2B).
Indeed, as can be seen in Figure 3 for four model organisms (S. cerevisiae, Caenorhabditis elegans, S. pombe, and Homo sapiens), the number of alternative ATGs is lower near the beginning of the ORF. Similar results were obtained for the 33 eukaryotes that were analyzed (Figure 4). Specifically, in almost all the organisms and frame shifts there is a decrease in the number of ATGs near the end of the 5′UTR, but also the beginning of the ORF. For example, we found that the genomic region of significant decreased number of ATGs at the beginning of the ORF (see details in the Methods section regarding the estimation of this region) is 14/14/11 codons (i.e. triplets of nt) for 0, 1 and 2 nt frame shifts respectively in S. cerevisiae (Figure 3). Similar analysis was performed for the 5′UTR. The average region under selection for all the analysed organisms is 21/13/15 codons (16 across all frames) for the end of the 5′UTR, and 4/19/10 codons (11 across all frames) for the beginning of the ORF, for the three frames respectively (Figure 4).
In the previous section we showed that there is a universal trend for a significantly lower number of ATGs near the beginning of the ORF. This result raises the question of whether these ATG profiles are selected for. For example, it is possible that the decrease in the number of ATGs near the beginning of the ORF is due to specific amino acid bias. Though the evolutionary selection for the ATG profile can be performed at the amino acid level and by non-synonymous mutations, we performed for the first time, an analysis to show that the signal remains, also when controlling for genomic features such as amino acid bias, codon bias and GC content.
Specifically, we compared the ATG profile obtained in each of the analyzed genomes to the one obtained in randomized genomes with identical proteins, total GC content and codon bias (see more details in the Methods section). As can be seen in Figure 5, the number of ATG codons near the beginning of the ORF indeed tends to be significantly lower than in the randomized genomes, supporting the conjecture that the observed ATG pattern is under selection. Expressly, in the analyzed organisms the 5′UTRs include a lower number of ATG codons than expected, but also fewer ATGs than expected can be found more than five codons downstream of the START codon. Interestingly, there are organisms with positions with more ATGs than expected (green dots in Figure 5) after the beginning of the ORF. This signal may be related (probably indirectly) to yet unknown codon bias signals after the beginning of the ORF. Analysis of the genomes of 33 eukaryotes by comparing them to randomized genomes (Methods) demonstrates that in most of them a region at the beginning of the ORF is under selection for less ATGs (Figure 6). Specifically the mean region under selection in the ORF for the analyzed genomes is 3.6 and 6.8 codons for the first and second frames respectively (frame 0 is the same for real and randomized genomes as we maintain the original proteins in the random genomes, see Methods and Figure 6), and 29.3, 25.3, and 26.8 in the 5′UTR for frame 0, 1 and 2 respectively.
It was previously demonstrated that there is selection for lower folding strength at the beginning of the ORF, presumably to improve the efficiency of translation initiation [30], [31]. Thus, it is possible that the observed selection for a lower number of ATGs at the beginning of the ORF is a result of the selection for weak mRNA folding in these regions. To examine the relation between the number of ATGs in a short mRNA sequence and the folding energy, we randomized the sub-windows of mRNA sequences of the S. cerevisiae genome maintaining their amino acid content; for each sub-window across all genes we compared the folding energy of variants with at least one ATG codon to variants with no ATG codons (Methods). We found that decreasing the number of ATGs tends to decrease the folding energy (i.e. increase the folding strength); thus, the decrease in the number of ATG codons at the beginning of the ORF is clearly not a result of selection for weak mRNA folding at the beginning of the ORF (Figure 7A).
It was also demonstrated that there is a pattern of slower codons at the beginning (first 30–50 codons) of the ORF probably to improve the cost and fidelity of translation [38]; thus, it is possible that the observed signal of decreased number of ATGs at the beginning of the ORF is somehow related to this reported pattern. To show that this is not the case, we sampled randomized genomes that maintain the distribution of codons at the beginning of the ORF (Methods), and demonstrated that the signal of less ATGs at the beginning of the ORF in these randomized genomes is significantly weaker than in the case of the real genome (see Figure 7B).
In addition, if indeed translation tends to occur also from alternative ATG codons, and assuming that such an alternative translation initiation is usually deleterious, we expect stronger selection for less alternative ATG codons in highly translated genes (e.g. genes with higher ribosomal load), which potentially have higher effect on the organism fitness than genes with lower translation rates. Indeed, this is exactly the pattern that was observed when we compared the group of 15% top/bottom genes in terms of the ribosomal load; see p-values in figure 7C). This result is in agreement with the scanning model, according to which upstream ATGs should exert negative translational control (e.g. see [14], [18]); however, we found that such a signal appears also downstream of the start ATG.
Finally, our analysis shows that for each of the non-ATG codons there is no region with significantly lower number of codons before/after the beginning of the ORF. Thus, indeed the ATG codon behaves differently from non-ATG codons, supporting the hypothesis that the ATG depletion is related to translation initiation from the alternative ATG codons (see supplementary Figure S35).
In this and the next subsections we report additional signals that are encoded near the beginning of the ORF to prevent alternative initiation of the ribosome, and thus to decrease the cost of translation.
At the first step, we considered the context of the alternative ATGs, which is related to the nucleotides in the vicinity of the main ATG codon. It has been demonstrated that the ribosome commences translation more efficiently from ATG codons with a specific context [10], [11]. Thus, we decided to verify if there is selection for ATG codons with less efficient contexts upstream and downstream the main START ATG. To this end, we developed a context score that is based on the distribution of nucleotides near the main START ATG in genes with a very high ribosomal-load. This score can be computed for every ATG, and higher scores reflect a context that is more similar to the main START ATG context of highly translated genes, and thus it is expected to contribute to a more efficient initiation (See details in the Methods section).
We computed this score for each alternative ATG codon, and plotted the mean observed context score profile as a function of the distance from the beginning of the ORF over the entire genomes of S. cerevisiae and S. pombe (Figure 8). S. cerevisiae and S. pombe were selected as model organisms in this study due to the large evolutionary distance between them (they are estimated to have diverged approximately 350–1,000 million years ago [39]). Indeed, as expected, the context score surrounding the ATGs in the vicinity of the main START ATG tends to be lower in the real genomes relatively to the randomized genomes (Figure 8A; S. cerevisiae: 5′UTR p-value<10−142, ORF p-value<10−142; S. pombe: 5′UTR p-value<10−142, ORF p-value<10−142).
In addition, the context scores surrounding the ATGs s in the vicinity of the main START ATG in the case of genes that are highly translated, tend to be lower than in the case of genes that consume less ribosomes; specifically the difference between the two groups is larger around the beginning of genes (S. cerevisiae: 5′UTR p-value = 1.6·10−15, ORF p-value<10−142 ; S. pombe: 5′UTR p-value = 4.2·10−142, ORF p-value<10−142 ; Figure 8B).
An additional way to decrease the cost of alternative out-of frame undesired initiation events is to introduce a stop codon close to them. In such cases the translated peptide will be shorter and thus its translation and degradation will consume fewer cell resources. To check if indeed there is such a selection, we compared the distances of alternative ATG codons to the closest STOP codon in the same frame obtained for two groups of ATGs: 1) alternative ATG codons that are near the START ATG (less than 6 codons) and thus have higher probability of being involved in alternative translation initiation based on the scanning model; 2) alternative ATGs that are further from the START ATG. Indeed the distances were significantly shorter for the first group when looking across all frames, both when considering ATGs in the 5′UTR (mean 30.8 nt vs. 41.2 nt; p-value = 0.0008) and at the beginning of the ORF (mean 42 nt vs. 46.7 nt; p-value = 0.05). Similar results were obtained when we considered the metabolic biosynthesis cost (instead of peptide length, Methods) of the peptide potentially translated from alternative ATGs near the START ATG vs. further from the START ATG (5′UTR: mean 294.8 vs. 404 with p-value = 0.0001; ORF: mean 402.6 vs. 452.3 with p-value = 0.04); or if we also added the translation cost (Methods; in the case of the 5′UTR: 366.6 vs. 500.1 with p-value = 0.0002, in the case of the beginning of the ORF: 500.6 vs. 561.1 with p-value = 0.046). In addition, as can be seen in Figure 8C–D - the metabolic cost of translating alternative ATGs is lower than in randomized genomes (Figure 8C; S. cerevisiae: 5′UTR p-value<10−245, ORF p-value<10−245; S. pombe: 5′UTR p-value<10−245, ORF p-value<10−245; the results when considering the translation cost in addition to the metabolic costs were very similar – see the Methods section), and in highly expressed genes relatively to lowly expressed genes (Figure 8D; S. cerevisiae: 5′UTR p-value = 1.8·10−16, ORF p-value = 3.6·10−120; S. pombe: 5′UTR p-value = 10−71, ORF p-value = 10−245). Thus, these results support the conjecture that there is selection for close STOP codons near alternative ATG codons to decrease the metabolic cost of their translation.
The current study suggests and surveys a list of transcript features that are related to translation initiation; specifically these ‘rules’ include: 1) Fewer ATG codons at the end of the 5′UTR; 2) Fewer ATG codons at the beginning of the ORF ;3) Optimal Kozak/context sequence at the START ATG; 4) Anti-optimal Kozak/context sequence of ATG codons near the START ATG; 5) Close stop codon for alternative ATGs.
Do the new rules related to translation initiation described in the current paper also have predictive power? For instance, can they explain the variance in measured protein levels and ribosomal densities?
To answer this question, we designed four computational predictors of protein levels and ribosomal density based on features and rules related to translation initiation described in this study; each of the following predictors was based on more features/rules in comparison to the previous ones (Methods). Four rules/features were considered: (1) The Kozak sequence in eukaryotes from [10]; (2) The main START ATG context score; (3) The number of alternative ATGs less than 30 codons downstream from the main START ATG; (4) The mean context scores of alternative ATGs less than 30 codons downstream from the main START ATG. We considered four predictors (A, B, C, and D); each predictor was based on an additional feature relatively to the previous one (i.e. A is based on feature (1); B is based on feature (2); C is based on features (2) and (3); D is based on features (2), (3), (4); see Figure 9B).
Indeed, as can be seen in Figure 9B, the correlation with binned protein levels and ribosomal density in S. cerevisiae (Methods) increases A) < B) <C) < D), i.e. this result supports the conjecture that we have designed a better model of initiation: the correlations with protein levels are 0.313/0.534/0.664/0.695 (p-values = 4.4·10−3/4.6·10−7/<10−12/<10−12) respectively, and the correlations with ribosomal density are 0.305/0.352/0.538/0.564 (p-values = 3.4·10−6/7.8·10−8/<10−12/<10−12) respectively. Similar results were obtained for S. pombe (see Figure 9B (, or when considering and controlling for the number of features in each predictor (Methods). Thus, the rules reported in this study can be utilized for designing highly expressed heterologous genes with efficient translation initiation.
In the current study we update and refine the set of rules related to the efficiency and the fidelity of translation initiation in eukaryotes. Specifically, we show that there are ancillary relevant signals upstream as required by the scanning model, but also downstream of the START ATG, in addition to a specific context surrounding the start ATG of the ORF reported in previous studies [10], [36]. While previous studies usually mentioned the possible contribution of the 5′UTR structure to translation initiation [4], [10], [14], [18], [37], [40], [41], we are the first to emphasize the selection against alternative ATG codons in the ORF. In addition, we are the first to study translation initiation signals in eukaryotes in a universal genome-wide manner, and the first to show that there is actually selection for the reported signals, while controlling for other possible explanations for these signals. Finally, we are the first to infer the region under selection for the reported signals; as discussed in the Methods section the inferred regions cannot be explained by the 5′UTR lengths.
The additional novel signals reported in this study include selection for less ATG codons before and after the main START ATG, but also more refined signals such as non-optimal nucleotide contexts near the alternative ATGs upstream and downstream of the main START ATG. We found that these signals are universal as they appear in many eukaryotes. In addition, these signals are selected for as they are stronger in highly expressed genes, and weaker in randomized genomes with the same proteins, GC-content, and codon bias as the original genome (even when considering the unusual distribution at the beginning of the ORF [38], see Methods). Thus, although there are cases where alternative ATG codons produce functional proteins [42]–[44], the results reported in this study demonstrate that specifically in the vicinity of the beginning of the ORF these ATG codons are usually deleterious and decrease the fitness of the organism. The mean estimated region under selection for less ATGs is the last 16.1--27.1 codons of the 5′UTR, and the first 5.2--11.2 codons of the beginning of the ORF; per-frame estimations appear in Figure 9B.
As we discuss in this section, the new initiation rules should improve the biophysical models of translation, the understanding of how translation efficiency is encoded in transcripts, and the constraints on the evolution of transcripts.
From a biophysical point of view, the results reported in this study can be used to develop more detailed models of translation initiation. For example, a possible initiation model that has been suggested [45] includes a diffusion/random-walk-type motion of the pre-initiation complex (note that other possible models may include energy-assisted directional scanning [45]). Under this model, the pre-initiation complex at each step has the probability to move forwards but also backwards along the transcript [45]. When the pre-initiation complex approaches an ATG codon it may start translating it with a probability that is related to the context of the ATG, but also to the distance of the ATG from the main START ATG [14]. Specifically, the probability of translating the main START ATG is lower than one, and the pre-initiation complex may continue scanning downstream of the START ATG. However, our analyses suggest that the probability to continue scanning after more than 5--11 codons downstream of the main START codon is negligible (illustration in Figure 9B) in terms of its effect on the organismal fitness – the selection signals after this region are not significant. The length of the region under selection is probably related to the biophysical properties of the pre-initiation complex such as its vibrations [46], its geometry, and the biophysical features of the mRNA molecules.
It was previously suggested that the translation elongation step is coupled in various ways with the initiation step via various signals encoded in the codons at the beginning of the coding sequences. For example, it was shown that the first codons of the coding sequences are under selection to induce weak local mRNA folding to improve the efficiency of translation initiation [30], [31], [47]. It was also shown that the codons at the beginning of the coding sequences have lower adaptation to the tRNA pool than the codons afterwards [38], and that the codons ∼10–25 downstream from the main ATG are under selection for strong mRNA folding [29], probably to improve ribosomal allocation by decreasing the initiation rate and increasing the distances between ribosomes.
The current study further demonstrates the coupling between the translation initiation and elongation steps. We suggest that additional signals for efficient and low-cost initiation appear not only at the 5′UTR but also at the beginning of the ORF (see also Figure 9B). These signals include amongst others lower numbers of ATGs and non-efficient ATG context scores in all reading frames. Thus, we suggest that the accepted paradigm that divides the translation process into three stages: initiation (occurs at the 5′UTR), elongation (at the coding sequence), and termination, is inaccurate. We propose an additional step that can be named late-initiation, which occurs at the region near the beginning of the ORF (Figure 9B). This step is part of the elongation stage and its efficiency is affected by signals encoded in the ORF, but is also related to the initiation step.
The results reported in this study are also related to the ‘cost’ of translation. It is known that translation is the metabolic process that consumes the largest amount of cellular energy [2]. Here we suggest that to accurately estimate the translation cost we should also take into account the alternative initiation events near the beginning of the ORF. The various signals related to selection against such alternative initiation events suggest, that in the case of genes that do have ATG codons near the beginning of the ORF, there may be a non-negligible number of alternative initiation events. These events presumably consume significant amounts of energy both at the synthesis stage of these short peptides and at their degradation stage.
A related result in this context is the fact that the decrease in the in-frame codon frequency at the beginning of the ORF is weaker than in the case of the out-of-frame codons (see, for example, Figure 4; the effect of in-frame codons is significant in 9 organisms). However, we actually do not expect that the signal in the in-frame codons will be as strong as in the case of out-of-frame or the 5′UTR signals due to the following reasons:
The analyses described in this study suggest that evolution tends to eliminate events of alternative translation initiation. However, many alternative ATG codons still appear near the beginning of the ORFs in all the analyzed eukaryotes. It is possible that these ATGs still exist since the fitness advantage of eliminating these ATG codons is not high enough. An additional possibility is that at least some of the resultant alternative proteins from these alternative short peptides are functional. Indeed, recently it was demonstrated that many ATGs in the 5′UTR probably initiate translation in two model organisms, Mus musculus [48] and S. cerevisiae [49]. Cases where alternative translation initiation events affect the localization of the produced protein have also been reported in recent years (see, for example [50]–[56]). Our analyses suggest that this phenomenon may be much more widespread and appears in many other eukaryotes.
Finally, in agreement with the above paragraph, the observed selection for less ATG codons near the beginning of the ORF reported here does not contradict the fact that the alternative ATG codons that do appear near the beginning of the ORF are relatively conserved [57]. Actually, these two results support each other – the fact that alternative ATG codons may trigger alternative initiation events means that the alternative ATG codons that do appear near the beginning of the ORF probably tend to initiate translation; if they appear there (and haven't been selected for) it may mean that they have a (relatively important) functional role related to such alternative initiation, and thus should be more conserved relatively to other codons in this region.
Analysis of the conservation of the nucleotides near the START ATG for 33 eukaryotes emphasized the importance and conservation of nucleotide position −3 [10], [36], [37] (Figure 2A). However, here we clearly show that the pattern of conservation of additional nucleotides varies along the evolutionary tree – different groups of eukaryotes have different conservation patterns. This result may be related to co-evolution between changes in the eukaryotic ribosomes along the evolutionary tree, and the nucleotide sequences that have the optimal interaction with them. Further future studies in this direction may teach us about the evolution of structural changes in the eukaryotic ribosome.
We would like to conclude the discussion with a comparison between the transcription and the translation processes. Specifically, do we expect to see similar/analogous signals related to the transcription initiation fidelity as the ones we reported regarding translation initiation? We believe that the answer to this question is negative – these signals will be weaker in the case of transcription. There are several major dissimilarities making the two processes significantly different in this context. First, only in the case of translation are there three different frames that usually correspond to very different proteins; in the case of transcription there are no out-of frame initiation events, and a small shift should not affect the resultant protein. Second, alternative transcription initiation is also expected to affect the length of the 5′UTR and not the ORF, again not altering the produced proteins. Thus, we believe that the effect of alternative transcription on the organism fitness should be significantly lower than the effect of alternative translation.
|
10.1371/journal.pntd.0003899 | Epidemiology of Leptospirosis in Africa: A Systematic Review of a Neglected Zoonosis and a Paradigm for ‘One Health’ in Africa | Leptospirosis is an important but neglected bacterial zoonosis that has been largely overlooked in Africa. In this systematic review, we aimed to summarise and compare current knowledge of: (1) the geographic distribution, prevalence, incidence and diversity of acute human leptospirosis in Africa; and (2) the geographic distribution, host range, prevalence and diversity of Leptospira spp. infection in animal hosts in Africa.
Following Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines, we searched for studies that described (1) acute human leptospirosis and (2) pathogenic Leptospira spp. infection in animals. We performed a literature search using eight international and regional databases for English and non-English articles published between January 1930 to October 2014 that met out pre-defined inclusion criteria and strict case definitions.
We identified 97 studies that described acute human leptospirosis (n = 46) or animal Leptospira infection (n = 51) in 26 African countries. The prevalence of acute human leptospirosis ranged from 2 3% to 19 8% (n = 11) in hospital patients with febrile illness. Incidence estimates were largely restricted to the Indian Ocean islands (3 to 101 cases per 100,000 per year (n = 6)). Data from Tanzania indicate that human disease incidence is also high in mainland Africa (75 to 102 cases per 100,000 per year). Three major species (Leptospira borgpetersenii, L. interrogans and L. kirschneri) are predominant in reports from Africa and isolates from a diverse range of serogroups have been reported in human and animal infections. Cattle appear to be important hosts of a large number of Leptospira serogroups in Africa, but few data are available to allow comparison of Leptospira infection in linked human and animal populations. We advocate a ‘One Health’ approach to promote multidisciplinary research efforts to improve understanding of the animal to human transmission of leptospirosis on the African continent.
| Leptospirosis is an important bacterial zoonosis that affects people and animals worldwide. It is common in tropical areas where people and animals live in close contact, but the disease has been widely neglected in Africa. In this study we aimed to demonstrate the extent of leptospirosis in Africa and describe the diversity of the causative agent Leptospira spp. in human and animal infections across the continent. Through a systematic literature review, we identified 97 studies from 26 African countries that described human disease or animal infection and met inclusion criteria. Leptospirosis was the cause of illness in 2 3% to 19 8% of hospital patients with a fever. Where population-level data were available, leptospirosis was estimated to affect 3 to 102 people per 100,000 every year. A variety of animal hosts of Leptospira spp. were identified. Cattle were reported as carriers of a variety of serological types of Leptospira spp. infection. The role of cattle and many other different animal hosts in human disease transmission remains unclear. Our review demonstrates that leptospirosis is a substantial cause of human illness in Africa, and we recommend integration of human and animal studies in the future to help us understand the epidemiology of leptospirosis on this continent.
| Endemic zoonotic diseases affect impoverished and developing communities worldwide but are frequently overshadowed in public and clinician awareness by high profile infections such as malaria and HIV/AIDS [1, 2]. In Africa, zoonotic infections are both directly responsible for human illness and death and indirectly impact human well-being as a result of reduced livestock productivity and food security [3–5]. However, bacterial zoonoses including leptospirosis remain under-diagnosed and under-reported in Africa, and as a result are overlooked as public health priorities [1, 2, 6].
Leptospirosis is one of the most common and widespread zoonotic infections in the world and is recognised as a neglected disease by the World Health Organisation (WHO) [7]. Human leptospirosis is caused by infection with pathogenic strains of Leptospira spp. bacteria [8, 9]. More than 250 pathogenic Leptospira serovars are known to exist worldwide, which are classified into 25 serogroups on the basis of their serological phenotype [10, 11]. Recent species determination by DNA homology has identified 13 pathogenic Leptospira spp., and seven of these (L. interrogans, L. borgpetersenii, L. santarosai, L. noguchii, L. weilli, L. kirschneri and L. alexanderi) are considered as the foremost agents of human and animal disease [10, 12]. Both serological and DNA-based classification systems are currently in use for clinical diagnosis and in understanding the pathogenesis and epidemiology of the disease [11, 13, 14].
A wide range of animals can carry pathogenic Leptospira bacteria and act as a source of infection [8, 11]. Leptospira serovars often demonstrate a degree of animal host preference and some common relationships between serovars and their hosts are reported [9, 15]. Following infection, the bacteria colonise the renal tubules and urogenital tract and are shed in the urine of infected animals. Animal species may be asymptomatic carriers of infection (maintenance hosts) or develop clinical disease (accidental hosts) depending on the infecting serovar [11, 16]. In food producing animals, cattle and pigs are relatively susceptible to clinical infection resulting in production losses including reduced milk yield, reproductive failure and abortions [16, 17].
In people, disease occurs through direct or indirect contact with infected urine from an animal host [8, 9, 15]. Good knowledge of Leptospira serovars circulating in local animal populations is important to determine sources and transmission routes for human infection [8]. In the early stages, human leptospirosis manifests most commonly as a non-specific febrile illness that is hard to distinguish from other aetiologies of febrile disease particularly in tropical areas [11, 18, 19]. Infection can result in severe secondary sequelae including renal failure and pulmonary haemorrhagic syndrome, and a case fatality ratio of up to 50% has been reported in complicated cases [15, 19].
Leptospirosis is particularly common in the tropical areas where people and animals live in close contact, and warm and humid conditions favour environmental survival and transmission of the pathogen [8, 9]. In South-East Asia and South America, leptospirosis is recognised as an important cause of renal failure and febrile disease [18–22]. However, despite its global importance, large gaps persist in our knowledge of the burden and epidemiology of leptospirosis in Africa. Reports from the WHO Leptospirosis Epidemiology Reference Group (LERG) indicate that leptospirosis incidence may be high in Africa, but also highlight the lack of available data [7, 23]. Although reported seroprevalence data demonstrates widespread exposure to Leptospira spp. in humans and animals in Africa, [24] little is known about the extent of human disease or the epidemiology of Leptospira infection in different animal species in Africa.
To tackle these gaps in current understanding and awareness of human and animal Leptospira infection in Africa, we performed a systematic review of peer-reviewed and grey literature following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines [25]. Our aims were summarise and compare: (1) current knowledge of the geographic distribution, prevalence, incidence and diversity of acute human leptospirosis in Africa; and (2) the geographic distribution, host range, prevalence and diversity of Leptospira spp. infection in animal hosts in Africa.
A detailed protocol for this study can be found in the supplementary material (S1 File). Following the PRISMA guidelines and checklist (S1 Checklist) references for this review were identified through searches of eight international and regional databases (Table 1) using the search string ‘Leptospirosis’ OR ‘Leptospira’ and ‘Africa*’ for articles published between January 1930 and October 2014 inclusively. Additional articles for inclusion were identified by bibliography hand searches of relevant articles [26].
Abstracts and titles were compiled in EndNote (Thomson Reuters, Philadelphia, PA, USA) and reviewed independently by two researchers (KJA, HMB) to determine whether each article met pre-determined abstract inclusion and exclusion criteria (S1 File). A third researcher (JEBH) served as a tiebreaker for any discordant decisions. Citations were included if they presented data on human or animal Leptospira spp. infection from any country within the United Nations (UN) definition of Africa [27]. We excluded abstracts that did not refer to original human or animal leptospirosis research data, or did not describe naturally occurring cases of leptospirosis in human or animal populations. We included case reports but excluded reports of returned travellers because of potential uncertainty around the specific location where infection was acquired.
Articles classified as eligible for inclusion were retrieved in full text format and assessed against pre-defined case definitions (Table 2) of human acute leptospirosis and carrier animal status agreed upon by three authors (KJA, HMB, JEBH). Rigorous diagnostic criteria were specified in accordance with WHO and international reference laboratory guidelines (Table 2) [7, 11, 16]. Serological diagnostics were not included in the case definition for carrier animals because of the inability to differentiate between previous exposure and current infection status. We also excluded articles describing studies that used laboratory animal inoculations as a diagnostic test for leptospirosis because of concerns over the risk of false positive results as a consequence of pre-existing infection in experimental animal colonies, diagnostic sensitivity and cross-contamination [16]. Full text articles were reviewed by two authors (KJA, HMB) and were excluded if they failed to meet case definitions, if results from the same cohort were presented more comprehensively in another eligible article, or if insufficient information was provided in the study methodology to determine whether the case definitions were met. Non-English language articles identified for full text review (n = 97) included French language articles translated by KJA with assistance from a native language speaker (n = 83); German language articles translated by a native language speaker (n = 7); Italian articles translated by a native language speaker (n = 4); Afrikaans (n = 2) and Dutch language articles (n = 1), which were translated using online translation software with support from a Dutch language speaker [28].
Two reviewers (KJA, HMB) independently extracted pre-determined qualitative and quantitative data from each included article. Data on infection prevalence and incidence for comparable studies (i.e. similar study inclusion criteria and diagnostic methodologies) were compiled, and ranges were presented by study type (human studies), location or host species (animal studies) if three or more citations reporting comparable data were identified. Data on serological and genetic typing of leptospiral isolates from people and animals were compiled and summarised by country and by animal species. Additional data on serogroup and genetic species of reported serovars was obtained from the Leptospirosis Library, maintained by the Leptospirosis Reference Centre, Royal Tropical Institute (KIT), Netherlands [29].
The risk of bias in included studies such as selection or reporting bias was assessed following the Cochrane guidelines for systematic reviews of medical interventions [30]. Full text study validity and methodological quality was assessed by comparison to pre-determined case definition criteria to control for heterogeneity in study design and diagnostic methodology (Table 2). Studies classified as high-risk for bias were not included in quantitative analysis of leptospirosis prevalence and incidence data.
Our searches yielded 681 unique articles from a total of 1201 abstracts identified by database searches. Data can be accessed through: http://dx.doi.org/10.5525/gla.researchdata.191. After abstract and full text review, 95 citations published between 1956 and 2014 were eligible for inclusion. Hand searches identified two additional articles that met inclusion criteria but were not identified in the original database search. Reasons for full-text exclusion are detailed in Fig 1. In total we included 97 articles that described human or animal studies conducted in 26 (44.8%) of 58 countries included in the UN macro-geographical definition of the African continent (Fig 2) [27]. Major risks of bias identified in eligible studies were selection bias, attrition bias in studies that relied on paired serology (MAT) for confirmatory diagnosis, and reporting bias, as descriptions of diagnostic methodology and results were often incomplete.
Acute human leptospirosis was reported in 46 eligible studies from 18 African countries (Fig 2) [31–76]. South Africa was the most frequently represented country with a total of six articles [43, 47, 54, 57, 65, 71], followed by Egypt [45, 55, 56, 58, 59] and Kenya [31, 37–39, 42] with five included articles. Twenty-one articles described acute human leptospirosis in hospital or health centre-based cohort studies (Table 3). Five articles described data from passive population-based surveillance [35, 41, 64, 70, 73], and two articles described active case-finding in the setting of an outbreak of acute febrile illness [31, 72]. Non-specific febrile illness was the most common clinical criteria described for cohort or surveillance study inclusion. Jaundice was stated as a primary inclusion criterion in three hospital-based cohort studies [44, 61, 66]. Haemoglobinuria was stated as the only inclusion criterion in one study conducted in the Democratic Republic of the Congo (DRC) [40].
The majority of studies (n = 41/46) used microscopic agglutination test (MAT) as a primary method to diagnose human cases of acute leptospirosis. IgM enzyme linked immunosorbent assay (ELISA) testing was the only diagnostic method used in three studies [31, 40, 63], but was more commonly used as part of a multi-faceted diagnostic approach (n = 6/46) [44, 45, 58, 64, 68, 73]. Fifteen (32 6%) of 46 eligible human studies demonstrated leptospirosis infection by blood culture in combination with serological diagnostics [34, 35, 37–39, 41, 44, 54–56, 58, 59, 64, 69], and nine (19 5%) studies also used PCR detection as well as culture and serology [34, 35, 41, 56, 58, 59, 64, 70, 73]. Genetic targets for diagnostic PCR assays included lbf1,[34, 35] lipL32 [34, 35], rrs [34, 35, 70], and ligA [58, 59]. No culture-independent genetic typing of Leptospira spp. was reported in any included human studies.
Leptospirosis prevalence varied by study design and inclusion criteria (Table 3). In hospital-based prospective cohort studies in mainland Africa that enrolled patients with non-specific febrile illness and used MAT serology for diagnosis of acute leptospirosis with or without adjunct diagnostics, prevalence ranged from 2 3% to 19 8% (n = 11; number of patients: median = 166; range = 39–2441) [33, 36–39, 42, 44, 45, 55, 58, 68]. A hospital-based prospective cohort study of febrile patients in Mayotte that diagnosed acute leptospirosis by PCR and culture without serology reported a prevalence of 13 7% (number of patients = 2523) [34]. In hospital-based cohort studies that used jaundice as the main study enrolment criterion, prevalence of acute leptospirosis ranged from 2 0% to 16 1% (n = 3; number of patients: median = 102; range = 99–392) [44–46]. Acute leptospirosis was also reported in one patient (2 3%) of 38 with haemoglobinuria [40], three patients (25 0%) of 12 involved in an outbreak of acute febrile disease in a pastoralist community in northern Kenya [31], and eight patients (9 8%) of 82 involved in an outbreak of acute pulmonary disease (pneumonia) in a mining camp in DRC [72].
Incidence estimates were calculated in five population-based surveillance studies [35, 41, 64, 70, 73] and two hospital-based prospective cohort studies [63, 74]. The only estimate of incidence from mainland Africa came from northern Tanzania, where regional incidence of 75 to 102 cases per 100,000 people per year was reported. This estimate was obtained by combining data on leptospirosis prevalence from hospital-based surveillance of febrile disease with multipliers derived from a population-based health-care utilisation survey [74]. For the Indian Ocean islands, incidence estimates were available for the Seychelles where the average annual incidence was estimated as 60 to101 cases per 100,000 [63, 70]; Réunion where the average annual incidence reported in three studies using a variety of data sources ranged from 3 1 to 12 0 cases per 100,000 [41, 64, 73] and Mayotte, where the average annual incidence calculated from cases identified through four years of active hospital-based surveillance between 2007 and 2010 was reported as 25 cases per 100,000 [35].
Sixteen case reports describing acute leptospirosis in a total of 34 individuals were considered eligible for study inclusion. A wide range of clinical manifestations were reported including febrile illness, jaundice, meningitis, and acute respiratory distress syndrome. Case reports described confirmed or probable acute leptospirosis in patients in South Africa (n = 6) [43, 47, 54, 57, 65, 71], Gabon (n = 3) [48, 62, 76], Morocco (n = 3) [50, 52, 53], Algeria (n = 1) [32], Mali (n = 1) [51], Réunion (n = 1) [75], and Senegal (n = 1) [60]. With the exception of Réunion and Senegal, case reports were the only eligible data on acute human leptospirosis from these countries.
Naturally occurring Leptospira spp. infection in animal hosts was reported by 51 eligible citations describing studies performed in 17 African countries (Fig 2) [77–127]. South Africa [84, 100, 101, 104, 117, 120–122] and Zimbabwe [83, 93–99] were the most frequently represented countries with a total of eight included articles per country, followed by Tanzania with seven articles [106, 110–112, 114–116]. Wild animal surveys were most commonly described (n = 21/51) followed by strain typing of Leptospira spp. previously isolated from naturally infected animal hosts (n = 13/51), livestock disease outbreaks (n = 7/51) and abattoir surveys (n = 7/51). Four citations (n = 4/51) reported human leptospirosis outbreaks as the inciting cause for investigations into animal carrier status [86, 109, 117, 123].
Leptospira spp. infection was demonstrated in a wide range of animal hosts (S1 Table), including cattle (Bos spp.) [85, 87, 89–91, 93–102, 111, 114, 119, 121, 127]; pigs (Sus scrofa domestica) [78, 79, 84, 85, 100, 104, 106, 122]; goats (Capra aegagrus hircus) [85]; Rusa deer (Rusa timorensis) [85]; dogs (Canis lupis familiaris) [85, 113, 116]; cats (Felis catus) [85, 113, 116]; rodents including the African grass rat (Arvicanthus niloticus) [87, 88], African giant pouched rat (Cricetomys gambianus) [110, 112], lesser tufted-tailed rat (Eliurus minor) [125], fringe-tailed Gerbil (Gerbilliscus robustus) [77, 88], rusty-bellied brush-furred rat (Lophuromus sikapusi) [109], multimammate mouse (Mastomys sp.) [83, 87, 103, 115], house mouse (Mus musculus) [80, 81, 83, 85, 118, 120, 124], brown rat (Rattus norvegicus) [82, 85, 103, 108, 117, 118, 120, 124], black rat (Rattus rattus) [83, 85, 86, 92, 103, 118, 120, 124], South African pouched mouse (Saccostomys campestris) [88]; and a range of other free-living mammal species including shrews (Crocidura spp. and Suncus murinus) [86, 103, 115, 118]; mongoose (Herpestes ichneumon, Mungo mungo and Paracynictic selousi) [80, 105]; Egyptian fox (Vulpes vulpes niloticus) [80]; shrew tenrecs (Microgale cowani, Microgale dobsoni, Microgale longicaudata, Microgale majori, Microgale principula) [125]; streaked tenrecs (Hemicentetes nigriceps, Hemicentetes semispinosus) [125]; and various bat species (Chaerephon pusillus, Miniopterus gleni, Miniopterus goudoti, Miniopterus griffithsi, Miniopterus griveaudi, Miniopterus mahafaliensis, Miniopterus majori, Miniopterus soroculus, Mormopterus francoismoutoui, Mormopterus jugularis, Mytotis goudoti, Otomops madagascariensis, Rousettus obliviosus, Triaenops furculus, Triaenops menamena) [107, 125]. Studies demonstrating infection in cattle were most common (n = 20/51) followed by pigs (n = 8/51), black rats (n = 8/51), brown rats (n = 7/51) and house mice (n = 7/51).
Culture and isolation was the most common detection method for Leptospira infection in animal studies (n = 43/51). PCR assays were used to demonstrate Leptospira spp. infection in 13 (25 5%) out of 51 studies [85, 86, 92, 103, 105, 107, 115, 118, 120, 123–126]. In three studies, culture and PCR were used in combination to determine infection status [92, 115, 118]. As with human studies, a variety of genetic targets were used in PCR assays to detect pathogenic leptospiral DNA, including lipL32/hap1, [85, 86, 118], secY, [103], rrl [105], and rrs [107, 115, 120]. PCR was predominantly used to demonstrate Leptospira spp. infection in rodents and wild animal species. Only one study in Réunion also used PCR assays to demonstrate infection in domestic animals [85].
Leptospira infection prevalence varied widely by target animal species and diagnostic methodology (S1 Table). Studies that used PCR diagnosis reported higher infection prevalence than studies that relied on Leptospira culture and isolation. Overall Leptospira infection prevalence reported in black rats tested by PCR ranged from 11 0% to 65 8% (n = 6; number of animals: median = 79, range = 33–141) [85, 86, 92, 103, 118, 124]. In two studies where black rats were tested by both PCR and culture, prevalence was higher by PCR (11 0%, n = 100; and 28.7%, n = 94) than by culture (4 0% and 3 2%) in Egypt [92], and Madagascar respectively [118]. A similar relationship was observed in brown rats, house mice and Asian house shrews tested in Madagascar [118]. Cattle and brown rats were the most common species tested by culture. Prevalence in brown rats ranged from 2 7% to 8 5% by culture (n = 3; number of animals: median = 256, range = 130–919) [82, 108, 117] but was considerably higher in three studies that used PCR to detect infection (10 0% to 4 7%; number of animals: median = 11, range = 10–96) [103, 118, 124]. In four abattoir-based surveillance studies of cattle from Egypt, Nigeria and Zimbabwe [87, 89, 93, 127], renal Leptospira spp. carrier status was detected by culture in 1 1% to 10 4% of sampled animals (number of animals: median = 480, range = 74–625), compared to 18 2% (number of animals = 77) in a single PCR-based study from Mayotte [85].
Serological typing of Leptospira spp. isolates from patients with acute leptospirosis was described in cohort studies conducted in the DRC [69], Egypt [55, 56], Ghana [44], Kenya [37–39] and Mayotte [34, 35], and in a case report from South Africa [54]. Isolates belonging to 15 serogroups were reported (Table 4). Mini and Icterohaemorrhagiae were the most commonly reported serogroups. Isolates that were equally cross-reactive with representative serovars from more than one serogroup (Mini/Hebdomadis and Pyrogenes/Ballum) were reported by two studies in Mayotte [34, 35]. In animal studies, isolates belonging to 12 serogroups were reported from 33 articles. At least one animal host was identified within Africa for 11 (73 3%) out of the 15 human-infecting serogroups identified in this review (Table 4). However, only six of these serogroups were detected in human and animal populations from the same country. These were serogroup Autumnalis in Kenya [39, 88]; and serogroups Canicola [56, 92, 113], Grippotyphosa [56, 80, 81, 92], Icterohaemorrhagiae [55, 80, 92, 127], Pomona [55, 56, 127] and Pyrogenes [56, 92] in Egypt. Serogroups associated with human febrile illness were frequently isolated from multiple animal hosts. One of the most commonly reported serogroups isolated from patients in Africa, serogroup Icterohaemorrhagiae, was isolated from cattle, brown rats, Egyptian mongoose and an Egyptian fox. Cattle were identified as carrier hosts for the widest range of Leptospira serogroups (n = 9) but several other animal species, such as African grass rats and black rats were also identified as carrier hosts for multiple serogroups.
Leptospira spp. isolated from human patients with acute leptospirosis belonged to five pathogenic Leptospira species (Table 5). L. interrogans was the most widespread and common species reported in either human or animal studies in Africa. Multiple animal hosts were identified for L. interrogans as well as the other common species, L. borgpetersenii and L. kirschneri, from a variety of countries. The widest diversity in Leptospira spp. was reported from two Kenyan studies of acute human leptospirosis, where isolates belonging to five species were identified (L. borgpetersenii, L. interrogans, L. kirschneri, L. noguchii and L. santarosai) [38, 39]. However, L. noguchii and L. santarosai were not detected in any other studies. Four Leptospira species: L. borgpetersenii, L. borgpetersenii-like, L. interrogans and L. kirschneri; were identified in two human studies on Mayotte, as well as by a concurrent study of black rats performed during the same period [34, 35, 86]. Divergent Leptospira spp. described as L. borgpetersenii-like and L. borgpetersenii Group B were detected in human and animal studies respectively in Mayotte, and in a study of indigenous small mammals in Madagascar [34, 35, 86, 125]. Sequencing and alignment of the atypical isolates from rat kidneys in Mayotte [86] showed perfect identity with isolates derived from people [35].
This systematic review is the first to synthesize and compile data on the epidemiology of acute human leptospirosis and pathogenic Leptospira spp. infection in animals in Africa. Leptospirosis remains amongst the neglected tropical diseases and is frequently overlooked in research priorities for African countries [1]. Yet, through this systematic review we have revealed a wealth of scientific evidence for acute human infection demonstrating that acute leptospirosis is an important cause of febrile illness in hospital patients across the African continent. Few studies providing population-level data on leptospirosis incidence in Africa were identified but available estimates indicate that the disease incidence is high in both island and mainland populations. In reports of human disease and animal infection, three predominant species, Leptospira borgpetersenii, L. interrogans and L. kirschneri, and a variety of Leptospira serogroups were diagnosed. Leptospira infection was reported in a wide range of domestic and wild animal species from across Africa but studies linking data on animal infections with studies of acute human disease were rare.
Acute leptospirosis was diagnosed in up to 19 8% of inpatients with non-specific febrile illness in hospital-based cohort studies conducted in several countries identified by this review. In sub-Saharan Africa, recent studies have highlighted that clinical over-diagnosis of malaria may conceal other aetiologies of febrile illness [20, 128]. Consistent with findings in other resource-limited tropical settings (e.g. South America [15, 129] and South-East Asia [130–132]), the evidence synthesised here demonstrates that acute leptospirosis infection is geographically widespread across the continent and should be considered as an important differential diagnosis for non-specific febrile illness in Africa.
Few estimates of leptospirosis incidence in Africa were identified by our review, revealing a key gap in research and surveillance outputs to date. The majority of incidence estimates identified came from the Indian Ocean islands where reports of annual incidence ranged from 3 1 to 101 cases per 100,000 people. In the African continent, the western Indian Ocean Islands appear to be the best-characterised region with regards to the human leptospirosis burden, possibly as a consequence of greater access to public health laboratories through French Territorial links [133]. We identified only one report of annual leptospirosis incidence from mainland Africa. This estimate of 75 to 102 cases per 100,000 people [74] was calculated from Tanzanian hospital-based surveillance data and is consistent with the WHO leptospirosis burden epidemiology reference group (LERG) predicted median African incidence of 95 5 cases per 100,000 [7]. At present, given the lack of population level data highlighted by LERG and by this review, estimates of the incidence of leptospirosis in Africa should be interpreted with care. However the data that are available from the continent indicate that the overall leptospirosis burden is likely to be high relative to other global regions. If the incidence figures identified by this review are close to the true burden of disease, up to 750,000 people in Africa will develop acute leptospirosis each year, representing a substantial disease burden that would far exceed current worldwide estimates (500,000 annual cases worldwide) [23].
Our review has revealed three predominant Leptospira species and a considerable diversity in reported pathogenic Leptospira serogroups in people and animals across the continent. Animal hosts, including livestock and invasive and indigenous rodent species, were reported for the majority of species and serogroups detected in human cases. However, there was poor geographical overlap in serogroup reporting between human and animal studies. Based on the findings of this review, we suggest that the major animal hosts of human-infecting serovars may vary across Africa and that both livestock and rodents may play important roles in human disease transmission. Few data were identified that described Leptospira spp. diversity in human cases and animal populations from the same country, and few studies attempted to link data on acute human leptospirosis with evidence of Leptospira infection in local animal populations. Studies on the Indian Ocean Islands of Mayotte and Madagascar were the exception to this. Isolates with unusual patterns of genetic and serological diversity, recently reclassified as a new pathogenic species Leptospira mayottensis [134], were detected from both human and black rat infections, implicating the black rat as the source of these human infections [35, 86]. These studies demonstrate the value of integrated human and animal research to identify sources and transmission routes of human leptospirosis, which can in turn help prioritise investment in disease prevention and control efforts.
The data included in this review most likely represents only the tip of the leptospirosis iceberg in Africa. Underreporting of leptospirosis is thought to be substantial and an overall lack of awareness about the disease and poor accessibility of diagnostic facilities are likely to contribute to this underreporting in Africa populations [135–137]. Patterns in reporting characteristics such as over-representation of study areas with greater research infrastructure, logistical connections or prior knowledge of a disease burden may also have resulted in reporting bias, particularly in assessing the geographic distribution of reports. We observed patient selection bias in some human studies, which limited the usefulness of reported prevalence data from these sources. Methodological limitations identified in this review include the use of the broad geographical search term ‘Africa’ rather than individual country names in our initial database searches. This approach may have missed eligible citations that are not indexed to the term ‘Africa’ in our selected databases. Our inclusion criteria may have created a bias towards more recent citations because of diagnostic technological advancements since the early era of our search period. Marked heterogeneity in methods and reporting criteria for serological diagnostic data prevented the meaningful synthesis and analysis of data on the reactive serogroups in human studies. We chose to include non-English language articles to allow inclusion of articles published in the colonial era, or in local language journals. Wherever possible, a proficient language speaker, in partnership with a study author, performed article translation. However, it is possible that some eligible studies may have been overlooked due to translation limitations.
Addressing the neglect of leptospirosis in Africa will be a major challenging for the future of leptospirosis research. Systematic review studies such as this can help to raise awareness of the human health threat of leptospirosis in Africa among researchers and policy makers. For medical clinicians, the non-specific presenting signs of acute leptospirosis in patients poses a substantial diagnostic challenge in developing countries where laboratory capacity rarely exists to diagnose the infection [18–20]. Hence, increasing clinician awareness and the development of treatment guidelines for the management of febrile patients should be a priority in resource-limited settings [138]. Integration of risk factor analysis in human cohort studies of febrile disease is also strongly advocated and would be a valuable next step in identifying groups at high risk of infection, and defining important animal to human transmission routes.
Knowledge of reservoir or carrier animal hosts is considered essential to understanding the epidemiology, transmission and control of leptospirosis in each setting [9, 11], yet our review has revealed that the linkages between Leptospira infections in people and animals are rarely addressed in the existing literature. Human and animal Leptospira infections are inextricably linked, and the multi-host epidemiology of leptospirosis means that there may be many potential sources of infection in a given setting. In the future, greater emphasis should be placed on performing multidisciplinary human and animal leptospirosis studies in the same geographical settings. Connecting investigations of animal reservoir populations with confirmed human cases would improve our understanding of the role that different animal species play in the transmission of pathogenic Leptospira serovars in a variety of geographic and environmental settings [8] [139]. Using an integrated ‘One Health’ approach to explore the relationship between human and animal Leptospira infection in areas where human disease is identified would also provide invaluable evidence to quantify the direct and indirect impacts of leptospirosis on human and animal populations in Africa [140, 141].
Control measures to prevent human leptospirosis often focus on rodent hosts of the disease. However, this review reveals that livestock are also important hosts of Leptospira infection in Africa, and may play a more substantial role in human disease transmission than is widely recognised. The clinical and sub-clinical productivity impacts of Leptospira infection in domestic animal populations in Africa are poorly understood. Around the world, several Leptospira serovars are considered to be of economic importance and cause production losses in a variety of livestock farming species including cattle, sheep, goats and pigs [17, 100, 142, 143]. More than 300 million of the world’s poorest people live in Africa, and at least 60% of these are in some part dependent on livestock for their livelihood [144]. Therefore, we consider that evaluating the impact of Leptospira infection on livestock health and productivity is also an important priority for prospective research in Africa. In the future, control of Leptospira infection in livestock species may have considerable scope to directly and indirectly improve human health and well-being in Africa, through reduced human leptospirosis transmission and increased productivity in livestock that subsistence farming communities [8, 142, 143, 145].
Finally, in 1967, the German leptospirosis researcher Kathe commented that ‘The world map of leptospirosis is, in fact, the world map of leptospirologists’ [67]. This is particularly true with regards to Africa. With this systematic review, we have started to outline the map of African leptospirosis; it is now time to fill in the gaps.
|
10.1371/journal.pntd.0007254 | Assessment of false negative rates of lactate dehydrogenase-based malaria rapid diagnostic tests for Plasmodium ovale detection | Currently, malaria rapid diagnostic tests (RDTs) are widely used for malaria diagnosis, but test performance and the factors that lead to failure of Plasmodium ovale detection are not well understood. In this study, three pLDH-based RDTs were evaluated using cases in China that originated in Africa. The sensitivity of Wondfo Pf/Pan, CareStart pLDH PAN and SD BIOLINE Pf/Pan in P. ovale detection was 70, 55 and 18%, respectively. CareStart was worse at detecting P. o. curtisi (36.5%) than at detecting P. o. wallikeri (75.0%), and SD could not detect P. o. curtisi. The overall detection ratio of all three RDTs decreased with parasite density and pLDH concentration. Wondfo, CareStart and SD detected only 75.0, 78.1 and 46.9% of the P. ovale cases, respectively, even when the parasitemia were higher than 5000 parasites/μL. Subspecies of P. ovale should be considered while to improve RDT quality for P. ovale diagnosis to achieve the goal of malaria elimination.
| Plasmodium ovale (P. ovale) are under-estimated and overshadowed by other malaria parasites in tropical countries, which can cause chronic infections that last from months to years. The chronic infection caused by P. ovale should be of concern in the context of the long-term goal of eliminating malaria. Rapid diagnostic tests (RDTs) is one of the WHO recommended tools to confirm the infection of plasmodium parasites, which can distinguish Plasmodium falciparum and non-falciparum species as well. However, little is known about their performance detecting P. ovale, and the factors that affect the efficiency of RDTs in the detection of P. ovale have not been systemically investigated. This study suggested that the performance of the three pLDH-based RDTs for P. ovale detection was not optimal, the low parasite density and pLDH concentration contributed to the failure of the RDT test for P. ovale. It provided information for the application of malaria RDTs in the field and for research and development to improve RDTs for malaria diagnosis.
| Plasmodium ovale has a wide geographic distribution across tropical countries, especially in Africa, Asia and some Western Pacific islands [1]. P. ovale has been overshadowed by other human malaria parasites in the field of medicine and medical research because of the relatively low morbidity and infections can be easily treated with conventional antimalarial drugs [2]. Although considered mild, P. ovale can cause acute respiratory distress syndrome and acute renal failure [3]. In addition, in the context of the long-term goal of eliminating malaria, it is becoming important to diagnose P. ovale in a timely manner. P. ovale, which shares with P. vivax the ability to form hypnozoites [4, 5], can cause chronic infections that may last from months to even years.
Rapid and reliable diagnosis is one of the key factors for malaria control and elimination. Accurate identification of Plasmodium infections is critical for administration of a targeted therapy, having a positive impact on patient health, disease management, and preventing transmission risk. Accurate diagnosis of malaria is needed to prevent the emergence and spread of drug resistant strains and to reduce the cost of medicine. The gold standard for malaria diagnosis remains the examination of Giemsa-stained smears by microscopy. This technique requires considerable training and experience. It is very challenging to maintain the capacity for microscopic examination in areas where malaria is being or has been eliminated [6]. In addition, P. ovale infection is difficult to diagnose microscopically owing to the generally low parasite density in patients and the morphology of P. ovale resembles that of P. vivax.
Malaria rapid diagnostic tests (RDTs) allow many countries to provide access to accurate malaria diagnosis even in the most remote areas by means of a simple-to-use, point-of-care test [7]. RDTs are easy to use and require no specific training or equipment. The results are visually readable as colored lines on a strip, and no special expertise is required. In recent decades, RDTs have replaced microscopy as the method of choice for diagnosing malaria in various settings. Reported sensitivities vary among different RDTs but are generally good for the detection of Plasmodium falciparum [8–11]. However, the sensitivity of RDTs for P. ovale detection is much lower, only 5.5% to 80%, with a sharp decrease observed at parasite densities lower than 500 parasites /μL [8–9, 11–13]. Based on a few assessments with a very limited number (n = 69–76) of tested samples, failure of P. ovale detection by RDTs has been reported [12, 14–15]. Malaria RDTs exhibited suboptimal performance in the detection of P. ovale infections [16], but the factors that affect the efficiency of RDTs in the detection of P. ovale have not been systemically investigated.
Targets of malaria RDTs are specific antigens of one or more Plasmodium species, such as histidine-rich protein (HRP2), lactate dehydrogenase (LDH), and aldolase. Among these antigens, Plasmodium-specific lactate dehydrogenase (pan-pLDH) is commonly used as a target of RDTs to detect all human Plasmodium species [14]. In addition, P. ovale is comprised of two genetic subspecies, namely, P. ovale curtisi and P. ovale wallikeri, and genetic variations based on P. ovale LDH gene polymorphism could also be involved in RDT failure [17]. We hypothesized that parasite density, level of pLDH, subspecies of P. ovale or polymorphism of pLDH gene might be involved in the failure of P. ovale detection by pLDH-based RDTs.
In this study, the performance of three pLDH-based RDTs (Wondfo diagnostic kit for malaria (Pf/pan) (colloidal gold), CareStart Malaria pLDH (PAN), and SD BIOLINE Malaria Ag Pf/Pan) for P. ovale detection were retrospectively evaluated with blood samples from returned international travelers and laborers from Africa to China. Moreover, the possible factors affecting RDT detection, i.e., parasite density, pLDH concentration, genetic subspecies and pLDH gene polymorphism, were investigated.
This study was reviewed and approved by the Institutional Ethics Committee of Jiangsu Institute of Parasitic Diseases (JIPD) (IRB00004221). All participants were adults and written informed consent was obtained from all participants before the interview or evaluation.
Blood samples were selected from venous EDTA-blood samples stored at -70°C, which were obtained from febrile patients at the clinics of local hospitals in Jiangsu province, China, and sent to the provincial reference laboratory. The patients were international travelers and laborers from African countries. The diagnosis of P. ovale was determined by both microscopy and confirmed by a nested PCR assay using the following commonly used protocols [18].
Thick and thin blood films were prepared from peripheral blood. Blood smears were stained with 3% Giemsa for 30 min at room temperature to identify parasites. Smears were analyzed by experienced microscopists at the JIPD. The standard method recommended by the World Health Organization (WHO) was used to estimate the number of circulating parasites per μL of blood. Parasite density was determined by counting the parasites and leucocytes, assuming 8,000 leucocytes /μL [19]. All the slides were subjected to double-blind verification by another independent microscopist, and the results were combined.
Three pLDH-based RDTs were used in this study: the Wondfo diagnostic kit for malaria (Pf/pan) (colloidal gold) (Guangzhou Wondfo Biotech Co., Ltd.; lot W05440903WC), CareStart Malaria pLDH (PAN) (Access Bio, Inc.; lot MN13G01) and SD BIOLINE Malaria Ag Pf/Pan (Standard Diagnostics Inc.; lots 05ED14111 and 05ED15003). The RDTs were selected based on the WHO/FIND malaria RDT performance evaluations and national guidelines of China. All of the RDTs were packed and sealed individually with desiccant and used immediately after opening and were performed based on the instructions of the manufacturers. The major target antigens of these three RDTs were pLDH, which are specific for all human-associated Plasmodium species. Five microliters of stored whole-blood samples were added to the pad, and three to four drops of specific lysis agent were added. The RDT result was read in 15–20 min according to the manufacturer’s instructions and immediately recorded by one person, a second person read the results 5 min later after the first person and was blinded to the initial reading. In the event of discordant results, a third person read the test blindly also and final results was the most common reading. The test was considered valid when the control line on the immunochromatographic test strip was visible. For the Pf/Pan test device (Wondfo diagnostic kit for malaria (Pf/pan) and SD BIOLINE Malaria Ag Pf/Pan), the result was recorded as non-falciparum only when the pan-pLDH line was positive. For the pLDH PAN test device (CareStart Malaria pLDH (PAN)), the presence of two colored bands (one band in the control and another band in the test) indicated a positive result for Plasmodium infection. All three RDTs were performed once.
Quantitative levels of pLDH antigens in P. ovale samples were determined using the Quantimal pLDH Malaria CELISA test (Cellabs), a sandwich ELISA for the detection of human Plasmodium pLDH. All P. ovale blood samples were tested in triplicate, and the manufacturer’s instructions were followed. Plates were read on a Zenyth 340 microplate spectrophotometer (Autobio) at 450 nm, with a reference wavelength of 620 nm. The reading was completed within 30 min after the stop solution was added. The mean optical density (OD) was calculated with the cut-off value as the means plus three standard deviations (SDs) of the wells containing healthy human blood alone.
Genomic DNA was extracted from 200 μL of whole-blood samples of P. ovale-infected patients using a QIAamp Blood Mini Kit (Qiagen) according to the manufacturer’s instructions. DNA extracted from healthy individuals living in nonendemic areas was used as a negative control in the amplification process. The real-time TaqMan PCR (qPCR) assay was used to detect P. ovale subspecies as described in a previous publication [20]. Amplification was performed with the following set of primers: POF (5′-ATAAACTATGCCGACTAGGTT-3′) and POR (5′-ACTTTGATTTCTCATAAGGTACT-3′). The probe pPOW HEX-AATTCCTTTTGGAAATTTCTTAGATTG-BHQ1 was used for detection of P. o. wallikeri, and pPOC FAM-TTCCTTTCGGGGAAATTTCTTAGA-BHQ1 was used for detection of P. o. curtisi. The qPCRs were carried out using LightCycler TaqMan Master (Roche, Germany) on a Roche LightCycler 480 (Roche, Germany) under the following conditions: one step at 95°C for 10 min; 45 cycles at 95°C for 15 sec and 60°C for 60 sec; and a final step at 4°C for 10 sec.
Nucleotide sequences corresponding to P. ovale LDH genes were amplified with the primers LDHovD21 (5′-GTTCTCGTTGGTCAGGAATGATA-3′) and LDHovC915 (5′-GGCATCATCAAACATCTTCTTTTCT-3′) by conventional PCR using Dream Taq Green PCR Master Mix (Thermo Scientific). Primer design and PCR conditions were based on a previous publication [20]. The PCR products were sequenced by Genescript Biological Technology Co., Ltd. (Nanjing, China). Nucleotide sequences of P. ovale LDHs were aligned using BioEdit software and compared with available P. ovale sequences in GenBank (accession number AY486058) and a paper by Bauffe et. al.[20]. Moreover, amino acid sequences were derived using GeneDoc software and also compared with available P. ovale sequences.
The RDT results for P. ovale detection were compared with each other, and sensitivity was calculated with 95% confidence intervals (CIs) by STATA (version 12.0). Categorical variables were determined by Chi-squared tests, with Fisher’s exact correction applied when the expected frequency in any cell was 5 or less. Correlation analyses between parasite density and pLDH OD level were performed as Pearson’s correlation analyses. A Pearson r value greater than 0.6 was considered a strong correlation. P values less than 0.05 were considered significant for all statistical analyses.
A total of 100 samples containing P. ovale parasites were studied. The samples had been collected from Feb 2012 to June 2015. The average age of the patients was 41.27 years (range, 22–58 years), and all the P. ovale infections were acquired in Africa. A majority (38/100) of the P. ovale samples were from Equatorial Guinea, followed by Nigeria (18/100), Angola (17/100) and the Republic of Congo (9/100) (Table 1).
The performances of the three pLDH-based RDTs (Wondfo Pf/pan, CareStart pLDH (PAN) and SD Pf/Pan) were compared for P. ovale detection. Of all the 100 confirmed P. ovale samples, 70 tested positive with the Wondfo Pf/Pan device, 55 tested positive with the CareStart pLDH device, and only 18 tested positive with the SD Pf/Pan device (Table 2). All three RDTs exhibited very high false negative rates for P. ovale detection.
To investigate whether the subspecies of P. ovale and the variations in P. ovale LDH polymorphism were associated with the sensitivity of the three RDTs for P. ovale detection, a real-time TaqMan PCR (qPCR) assay was carried out for subspecies determination, and the pLDH gene was sequenced for polymorphism analysis. Among the 100 P. ovale samples, 52 were P. o. curtisi and 48 were P. o. wallikeri, as determined by qPCR. The two subspecies exhibited no difference in parasitemia and pLDH levels (P>0.05).
There was no significant difference in the sensitivity of the detection of the two subspecies by the Wondfo Pf/Pan device (χ2 = 0.49, P = 0.485). A higher sensitivity was observed for P. o. wallikeri than that for P. o. curtisi (approximately 75% vs 36.5%) with the CareStart Pan device with statistically significance (χ2 = 14.92, P<0.0001). On the other hand, the SD Pf/Pan device could not detect P. o. curtisi at all, and a high false negative rate (62.5%) was observed for P. o. wallikeri detection with this device (χ2 = 20.88, P<0.0001, Fisher’s exact<0.0001 and one-sided Fisher’s exact<0.0001) (Table 3).
The amplified P. ovale LDH gene yielded approximately 890 base pairs, coding for 294 amino acids. A total of 100 of the amplified genes were sequenced to analyze the genetic variations in the P. ovale LDH gene using Clustal W2 software. No nucleotide substitution was detected within the sequences of each subspecies compared to the reference sequences (P. o. curtisi from GenBank (AY486058) and P. o. wallikeri from a paper by Bauffe et. al.). There were twenty-four single nucleotide polymorphisms (SNPs) between the two subspecies, and three of these SNPs were nonsynonymous mutations (S143P, N168K, I204V), which was consistent with the results of a previous study [20].
The performance of each RDT was evaluated according to parasite density levels. Based on P. ovale parasite densities, the 100 samples were divided into three groups. For the Wondfo Pf/Pan device, the sensitivity for cases with parasite densities greater than 500 parasites/μL (75.4% and 75%) was higher than that for cases with densities lower than 500 parasites/μL (27.3%), and this difference was statistically significant (χ2 = 10.75, P<0.05, Fisher’s exact = 0.007). However, for the SD Pf/Pan device, regardless of parasite density, the sensitivity was less than 50%, and the cases with parasite densities lower than 500 parasites/μL could not be detected with this device and a significant difference was observed (χ2 = 26.76, P<0.0001, Fisher’s exact<0.0001). The sensitivity of the CareStart Pan device reached 78.1% for only those cases that had parasite densities greater than 5000 parasites/μL, and a significant difference was observed (χ2 = 10.49, P = 0.005, Fisher’s exact = 0.005) (Table 4).
To determine the relationship between pLDH concentration and the sensitivity of the three RDTs for P. ovale detection, the P. ovale samples were divided into three groups according to the pLDH OD levels, and the sensitivity of the three RDTs was evaluated in each group. For the Wondfo Pf/Pan and CareStart Pan device, the sensitivity for P. ovale detection reached 100% and 89.7% when the pLDH OD levels in the samples were more than 0.5; the sensitivity increased with the pLDH levels; and a significant difference was observed (χ2 = 82.05, P<0.0001, Fisher’s exact<0.0001 for Wondfo; χ2 = 32.09, P<0.0001, Fisher’s exact<0.0001 for CareStart). On the other hand, the sensitivity of the SD Pf/Pan device increased with the pLDH levels, and a significant difference was also observed (χ2 = 38.52, P<0.0001, Fisher’s exact<0.0001), but the sensitivity reached only 55.2%, even when the pLDH level was high (>0.5 OD) (Table 5).
To determine whether the pLDH levels were associated with parasitemia in these P. ovale samples, the correlation between the pLDH levels and peripheral blood parasitemia was assessed. A moderate correlation was observed between pLDH levels and parasitemia (r = 0.5510, P value<0.0001). Some disagreement was observed between parasitemia and pLDH levels. Ten cases of P. ovale samples with parasitemia greater than 10,000 parasites/μL presented low levels of pLDH (OD<1), and five cases with low parasitemia (<10,000 parasites/μL) presented high pLDH levels (OD>1) (Fig 1).
Currently, hundreds of RDTs are available in the global market for malaria diagnosis, and these tests play a very important role in current malaria control/elimination programs. However, the quality of malaria RDTs varies among different companies, even among different lots from the same company. The FIND-WHO global RDT evaluation program was carried out on P. falciparum and P. vivax clinical samples (https://www.finddx.org/malaria/) [21], but there is limited information regarding the effectiveness of RDTs in P. ovale and Plasmodium malariae detection. This study evaluated three pLDH-based RDTs (Wondfo Pf/Pan, CareStart PAN, and SD Pf/Pan) for P. ovale malaria diagnosis using cases of malaria infections acquired in Africa and brought to China. The results showed that the three RDTs performed poorly in the detection of P. ovale. The most sensitive test, namely, Wondfo Pf/Pan, detected only 70% of the confirmed P. ovale samples; similar results were obtained in a previous study [9, 11]. In this study, the sensitivity of the SD Pf/Pan test for P. ovale was only 18%, which was much lower than that observed in previous reports (76.3% to 90.5% and 76.9%) [9, 11]. This discrepancy may be due to differences in product design or product load. The high rate of false negative results of the RDTs in P. ovale detection is a major challenge for malaria elimination in areas with a prevalence of both P. falciparum and non-falciparum strains.
Normally, the performances of malaria RDTs are analyzed by comparing the values with the parasite density levels observed by microscopic examination. The poor performance of RDTs in the detection of P. ovale may be explained by low parasite density, but several exceptional cases could not be explained by parasite density alone. All three RDTs had similar limits of detection, and all three exhibited poor performances for infections with parasite densities less than 500 parasites/μL. The sensitivity for the high parasitemia group (higher than 5000 parasites/μL) was also unsatisfactory, which was observed in other studies [8, 9, 11–13].
The pLDH antigen is one of the most promising antigens explored so far and is assumed to be a specific marker for the presence of Plasmodium in blood. pLDH does not persist in the blood [22, 23], and the amount of pLDH indicates the metabolic presence of Plasmodium parasites due to the low stability of pLDH in the body [24]. In this study, lower sensitivities of all three RDTs were observed in groups with lower pLDH antigen levels. The sensitivity reached 100% and 89.7% when the pLDH OD was greater than 0.5 for the Wondfo Pf/Pan and CareStart Pan test. Thus, the pLDH levels of P. ovale were one of contributing factors to the variations in the performance of pLDH-based RDTs, which was consistent with the results of a similar study on the detection of P. vivax with RDTs [24]. On the other hand, the relationship between pLDH levels and parasite density was assessed in this study, and a positive correlation was observed (r = 0.5510, P value<0.0001). A similar result was also observed for P. vivax that pLDH levels showed moderate correlation with parasite density (r = 0.4, P < 0.05) [24]. Some discrepancies were observed between parasite density and pLDH levels, while similar results were also observed in study of P. vivax [24] and in a rodent malaria model [25]. There are 36.4% (4/11) cases with low parasitaemia (<500 p/μL) and relatively high pLDH (OD>0.1), while 33.3% (5/15) cases have high parasitaemia (>10,000 p/μL) and relatively low pLDH (OD<0.5). A total of 9 cases (9% of the total population in the study) were not coordinate with the correlation between parasitaemia and pLDH level. Since the evaluation of RDTs is currently based on patients’ parasitaemia levels, while little is known about the targeted biomarkers in clinical patients, which maybe one of the factors that cause the discordant results in the present study.
Except for low parasite density and pLDH levels, the failure of the RDT test for P. ovale detection is also currently hypothesized to be contributed to the natural variability among tested species. The false negative rate observed for P. o. curtisi was higher than that for P. o. wallikeri (approximately 60% vs 43%, respectively) [20]. No substantial difference was observed with the Wondfo Pf/Pan test between both P. ovale species detected in this study. With the CareStart Pan device, the false negative rate observed for P. o. curtisi was higher than that for P. o. wallikeri (approximately 75% vs. 36.5%, respectively), which was consistent with the results of a previous study [20]. The SD Pf/Pan device could not detect P. o. curtisi at all, while a high false negative rate (62.5%) for P. o. wallikeri was observed. As genetic variations based on P. ovale LDH could be involved in RDT failure [17], P. ovale LDH gene polymorphism was evaluated for P. o. curtisi and P. o. wallikeri, and no nucleotide substitution was observed in either subspecies. Similar result was observed on P. falciparum ldh sequences, which also highly conserved with haplotype and nucleotide diversity value 0.203 and 0.0004 [26]. However, the isoforms of pLDH maybe different and the binding site of the antibody is unknown, these factors could be in involved in the failure of the P. ovale detection by RDTs.
The present study has its limitation. For the retrospective samples, it was impossible to explore the cause of discordant results, such as the sensitivity for the high parasitemia group (higher than 5000 parasites/μL) was unsatisfactory. Another limitation was that readers of RDTs were not blinded to the results of microscopy and PCR, which will cause the subjective bias. Furthermore, as frozen blood samples were applied in the study and an influence of sample storage, such as cell lysis and decreased level of pLDH caused by frozen-thraw, cannot be excluded.
The results of this study suggested that the performance of the three pLDH-based RDTs for P. ovale detection was not optimal. The low parasite density and pLDH concentration contributed to the failure of the RDT test for P. ovale. The subspecies of P. ovale can affect the sensitivity of the detection of P. ovale for the CareStart Pf/Pan and SD PAN RDTs but not the Wondfo Pf/Pan RDT, and the P. ovale LDH gene was relatively well conserved among the subspecies. Therefore, malaria diagnosis might be difficult using only RDTs, especially for P. ovale infections. The present results in the study could provide more aspects for producing better RDTs with significantly improved sensitivity for P. ovale.
|
10.1371/journal.pgen.1002603 | Balanced Codon Usage Optimizes Eukaryotic Translational Efficiency | Cellular efficiency in protein translation is an important fitness determinant in rapidly growing organisms. It is widely believed that synonymous codons are translated with unequal speeds and that translational efficiency is maximized by the exclusive use of rapidly translated codons. Here we estimate the in vivo translational speeds of all sense codons from the budding yeast Saccharomyces cerevisiae. Surprisingly, preferentially used codons are not translated faster than unpreferred ones. We hypothesize that this phenomenon is a result of codon usage in proportion to cognate tRNA concentrations, the optimal strategy in enhancing translational efficiency under tRNA shortage. Our predicted codon–tRNA balance is indeed observed from all model eukaryotes examined, and its impact on translational efficiency is further validated experimentally. Our study reveals a previously unsuspected mechanism by which unequal codon usage increases translational efficiency, demonstrates widespread natural selection for translational efficiency, and offers new strategies to improve synthetic biology.
| Although an amino acid can be encoded by multiple synonymous codons, these codons are not used equally frequently in a genome. Biased codon usage is believed to improve translational efficiency because it is thought that preferentially used codons are translated faster than unpreferred ones. Surprisingly, we find similar translational speeds among synonymous codons. We show that translational efficiency is optimized by a previously unknown mechanism that relies on proportional use of codons according to their cognate tRNA concentrations. Our results provide important molecular details of protein translation, answer why codon usage is unequal, demonstrate widespread natural selection for translational efficiency, and can guide designs of synthetic genomes and cells with efficient translation systems.
| Eighteen of the 20 amino acids are each encoded by two or more synonymous codons in the standard genetic code, yet the synonymous codons are often used unequally in a genome. Such codon usage bias (CUB) has been extensively documented in all three domains of life [1]–[3]. Within a genome, highly expressed genes tend to have stronger CUB than lowly expressed ones [4], and the codons preferentially used in highly expressed genes of a species are referred to as preferred codons.
Although codon usage is clearly determined by the joint actions of mutation, drift, and selection [5]–[6], the fitness benefit of CUB is less clear. There are two prevailing, non-mutually exclusive, hypotheses on the selective utility of CUB: accuracy and efficiency of protein translation [6]. The translational accuracy hypothesis asserts that different synonymous codons have different probabilities of mistranslation, and that the use of accurately translated codons is beneficial because mistranslation reduces the number of functional molecules, wastes energy, and/or induces cytotoxic protein misfolding. Unequivocal evidence for this hypothesis exists [7]–[10].
By contrast, the translational efficiency hypothesis lacks direct evidence. This hypothesis holds that different synonymous codons are translated at different speeds, and that faster translation is beneficial because it minimizes ribosome sequestering and so helps alleviate ribosome shortage [5], [11]–[12]. The relevance of ribosome shortage is evident from the findings that most ribosomes are actively engaged in translation during rapid cell growth [13]–[14] and that ribosome concentration increases with the rate of cell growth [15]. An important observation invoked to support the efficiency hypothesis is that cognate tRNAs of preferred codons tend to have higher cellular concentrations (or more gene copies) than those of unpreferred codons [4], [16], which may allow faster translation of preferred codons than unpreferred codons. While results from several earlier studies are consistent with this hypothesis [12], [17], these studies do not exclude the possibility that the observed differences in activity or fitness caused by synonymous mutations are entirely due to CUB's influence on translational accuracy (see Discussion). Here we directly test the efficiency hypothesis and its presumed underlying mechanism.
The translational efficiency hypothesis assumes that synonymous codons have different translational speeds, caused by disparities in codon selection time (CST), the time needed for ribosomal A site to find the cognate ternary complex of aminoacylated tRNA+eEF-1α+GTP. To test this proposition, we took advantage of a genome-wide ribosome profiling study of Saccharomyces cerevisiae that surveyed ribosome-protected mRNA fragments at a nucleotide resolution in a cell population at a given moment by Illumina deep sequencing [18]. Because the probability that a codon is docked at the A site is proportional to its CST, we estimated the relative CSTs of all 61 sense codons (Figure 1A) by the ratio of the observed codon frequencies at the A site in the ribosome profiling data and the expected codon frequencies estimated from mRNA-Seq data generated under the same condition in the same experiment (Figures S1, S2, S3; see Materials and Methods). The standard errors of the CST estimates, measured by bootstrapping genes from the original datasets, are on average 12% of the CST estimates (Figure 1A), indicating that our CST estimates are overall quite precise.
CUB is commonly measured by the relative synonymous codon usage (RSCU), defined by the frequency of a codon relative to the average frequency of all of its synonymous codons in a set of highly expressed genes [19]. To compare the usage of all 61 sense codons, we also use RSCU', which is the proportion of use of a given codon among synonymous choices in a set of highly expressed genes (see Materials and Methods). Another commonly used measure of CUB is the codon adaptation index (CAI) [20], which is calculated for a gene, and measures its usage of high-RSCU codons (see Materials and Methods). The greater the CAI, the more prevalent are preferred codons in the gene.
Contrary to the widely held presumption that preferred codons are translated faster than unpreferred codons, no significant negative correlation between RSCU' and CST was observed among the 61 sense codons (Figure 1B). It is also believed that codons with abundant cognate tRNAs tend to have low CSTs. Because tRNA gene copy number and tRNA concentration are highly positively correlated [21]–[22], the former is often used as a proxy of the latter. However, neither tRNA gene copy number (Figure 1C) nor tRNA concentration (Figure 1D) correlates negatively with CST. Because codons and tRNAs do not have one-to-one correspondence, in the foregoing analysis, we considered the best-matching tRNA species for each codon. This codon-tRNA relationship has been shown to be more accurate than the wobble rule, at least in yeast [22].
We also examined each amino acid separately. Among the 18 amino acids with at least two codons, 12 (Ala, Asn, Cys, Gln, Glu, Gly, Ile, Lys, Ser, Thr, Tyr, and Val) showed a negative correlation between RSCU' and CST, while 6 (Arg, Asp, His, Leu Phe, and Pro) showed a positive correlation, when statistical significance of the correlation was not required (Figure 1A). The number of negative correlations is not significantly more than the chance expectation of 9 (P = 0.12, one-tail sign test).
Using the standard errors of the CST estimates for the foregoing 18 amino acids (Figure 1A), we tested whether the CSTs are significantly different between the synonymous codon with the highest RSCU' and that with the lowest RSCU'. After the control for multiple testing by the Bonferroni correction, only two amino acids showed significant differences. The highest-RSCU' codon has a lower CST than the lowest-RSCU' codon for glycine (nominal P = 0.002), while the opposite is true for arginine (nominal P<0.001). Our results are robust to different multiple-testing corrections, as no other amino acids show a nominal P<0.01. Furthermore, when RSCU' is not considered, arginine is the only amino acid for which synonymous codons show significant heterogeneity in CST at the 5% significance level after the correction for multiple testing. Following an earlier study [1], we also tried defining preferred codons without using gene expression data, but the results are not different (Figure S4). The overall lack of a significant negative correlation between CST and synonymous codon usage is real rather than an artifact of imprecise CST estimation, because the standard errors of CSTs are quite small (Figure 1A) and CSTs of several nonsynonymous codons differ significantly from one another (see below).
To validate the above findings, we also directly compared RSCU' values of individual codon positions of Illumina reads from the ribosome profiling data, without estimating CSTs. If unpreferred codons are translated more slowly and therefore stay at the ribosomal A site longer than preferred codons, codons at the A site should have a lower RSCU' on average than its neighboring sites of the same read, after the correction of sequencing bias by mRNA-Seq data. However, we observed no dip in RSCU' at the A site (Figure 1E). We further calculated, within each gene, the ratio between the frequency of preferred codons and that of unpreferred codons at the ribosome A site of Illumina reads from the ribosome profiling data, after correction by mRNA-Seq. This ratio is expected to be 1 if preferred and unpreferred codons are translated equally fast. Indeed, after combining the ratio for all amino acids and all genes using the Mantel-Haenszel procedure [23], we found the overall ratio to be 0.984, not significantly different from 1 (P = 0.21, two-tail χ2 test).
The above findings are puzzling, because the first step in the interaction between tRNA and mRNA is non-specific [24] and the relative waiting time for the cognate tRNA to arrive at the ribosome A site is expected to be inversely proportional to the relative concentration of the cognate tRNA. It was also reported that CST is the rate-limiting step in translational elongation [25]. The only plausible explanation of similar CSTs among synonymous codons is that, in wild-type yeast cells for which the ribosome profiling was conducted, available cognate tRNAs for translating synonymous codons have effectively the same concentration.
In rapidly growing yeast, ∼80% of total RNA is rRNA and ∼15% is tRNA [15]. The mean length of yeast tRNAs is ∼72 nucleotides and the total length of rRNAs per ribosome is 5469 nucleotides [15]. Thus, the number of tRNA molecules per cell is approximately (15%/72)/(80%/5469) = 14.2 times the number of ribosomes per cell, substantially exceeding the expected ratio of two tRNAs per active ribosome (at A and P sites, respectively) if tRNA recharging and diffusion is instantaneous.
In reality, however, tRNA recycling takes time and thus cannot be ignored. Each tRNA, after completing its job of transferring an amino acid to the elongating peptide and then exiting the ribosomal E site, needs to be recharged with the cognate amino acid and then with eEF-1α+GTP to form a ternary complex before it can be reused in translation. It has been estimated that each ribosome translates ∼32.6 codons per second in yeast [26]. This implies that on average a tRNA molecule needs to be used 32.6/14.2 = 2.3 times per second, or once every 0.44 second. It is possible that the time for ternary complexes to form and diffuse to ribosomal A site is a substantial fraction of 0.44 second, so that the local concentration of ternary complexes is much lower than the total tRNA concentration. A recent study reported that consecutive synonymous codons in an mRNA tend to use the same tRNA and proposed that this codon choice is beneficial because a tRNA does not diffuse far from the ribosome after exiting its E site and is reused for translating the next synonymous codon when the ternary complex is formed again [27]. This observation and its explanation strongly implies that the local concentration of ternary complexes is low; otherwise, the addition of one cognate tRNA molecule among on average 20 tRNAs (because identical amino acids are expected to be on average 20 residues apart) cannot significantly increase the relative concentration of the cognate tRNA around the ribosome. Based on available information in E. coli, we calculated that the physiological concentration of ternary complexes is only ∼4.3% of the total concentration of tRNAs and ∼22% of the concentration of ribosomes (see Materials and Methods). These observations strongly support our hypothesis that available tRNA is in shortage during translation. Consistent with our hypothesis, total tRNA concentrations increase with the rate of cell growth in E. coli [28] and tRNA gene copy number increases with the shortening of the minimal generation time across species [29].
Under tRNA shortage, the optimal usage of synonymous codons in minimizing the total CST (i.e., maximizing translational efficiency) is to use isoaccepting tRNAs in proportion to their concentrations (see Materials and Methods). That is, pi = qi, where pi is the relative usage of the ith synonymous codon of an amino acid (Σpi = 1) and qi is the relative concentration of the corresponding tRNA (Σqi = 1). Under this codon usage, available cognate tRNAs of synonymous codons have equal concentrations and synonymous codon selection times become identical (see Materials and Methods). We will refer to this theoretical optimal codon usage under tRNA shortage as the proportional rule. The proportional rule is not predicted by other models. For example, without tRNA shortage, two optimal solutions in minimizing the total CST exist. When codon usage is fixed, isoaccepting tRNA concentrations should follow , which is referred to as the square rule [30]–[31]. When tRNA concentrations are fixed, only the codon corresponding to the most abundant tRNA species should be used [30], which is referred to as the truncation rule.
To test if the actual codon usage of yeast follows the proportional rule, we examined the 12 amino acids that are each translated by at least two tRNA species in yeast. For each amino acid, the relative transcriptomic usage of a codon among synonymous codons (i.e., pi) is quite close to the relative gene copy number of its cognate tRNA among isoaccepting tRNAs (i.e., qi), as predicted by the proportional rule (Figure 2A). We measured the Euclidian (Figure 2B) and Manhattan (Figure 2C) distances in synonymous codon usage from the observed values to those predicted by the proportional rule, and found these distances significantly shorter than expected by chance (Figure 2B–2D; Table S1; see Materials and Methods). Not surprisingly, genomic codon usage fits the proportional rule less well than the transcriptomic codon usage (Figure 2A), reflected by greater distances from the predicted values (Figure 2B, 2C).
The better fitting of the transcriptomic codon usage to the proportional rule than to the square rule and truncation rule can be seen from a comparison of the distances under these three models (Figure 2D). We also compared the likelihood of the three models, given the observed codon usage (Figure 2D). The proportional model has a much higher log10(likelihood) than the square model. Because the likelihood of the truncation model is 0, this model is much worse than the other two models. The same conclusions are reached for the transcriptomic codon usage of all other model eukaryotes we examined (Figure 2A, 2D).
In the above analysis, we combined synonymous codons that are recognized by the same tRNA species (referred to as iso-synonymous codons). Because the relative usage of such iso-synonymous codons does not affect the relative usage of isoaccepting tRNAs, it presumably does not affect translational efficiency. Nonetheless, iso-synonymous codons are not used equally, and factors other than translational efficiency (e.g., translational accuracy) may be at work (Table S2).
The observation of similar CSTs among synonymous codons and the empirical validation of the proportional rule strongly support the following model that includes three elements: (1) available tRNAs are in shortage during translation, (2) translational efficiency is optimized in nature by balanced codon usage according to tRNA concentrations, and (3) synonymous codons are translated with similar speeds under the codon-tRNA balance. Our model predicts reduced translational efficiency due to ribosome sequestering when the codon-tRNA balance is broken. It further predicts lower efficiency under exclusive use of preferred codons than balanced use of preferred and unpreferred codons.
We experimentally tested the above predictions by quantifying the cellular efficiency in translation, represented by the protein expression of a reporter gene, under different levels of codon-tRNA imbalance induced by the expression of another gene. Unlike previous studies [12], [17], our separation of the inducer and reporter allows the distinction among several potential mechanisms of CUB's impact on protein expression. We inserted our reporter gene, the Venus yellow fluorescent protein (vYFP) gene controlled by the GPD promoter, into Chromosome XII of a haploid strain of S. cerevisiae (Figure 3A). We then designed four synonymous sequences encoding another fluorescent protein, mCherry, as our inducer (Figure S5). The four mCherry sequences, named mCherry-1, 2, 3, and 4, cover the entire range of CAI of native yeast genes (Figure 3B). We developed an index, distance to native codon usage (Dncu), to measure the difference between the codon usage of a (heterologous) gene and the overall codon usage of the host cell, which is proportional to tRNA concentrations (see Materials and Methods). The four mCherry versions also span a large range of Dncu (Figure 3C) and show different degrees of codon-tRNA imbalance for individual amino acids (Figure S6). Other than synonymous codon usage, the four mCherry versions are nearly identical: they encode the same protein sequence, have similar G+C content (42–44%), and have identical sequences in the first 56 nucleotides of the coding region, because this region may affect the level of protein expression [12], [32]–[33]. Each mCherry gene is expressed from a constitutive and strong promoter on a high-copy-number plasmid (see Materials and Methods). The four plasmids were separately transformed to yeast cells carrying the vYFP reporter gene (Figure 3A). Our model predicts that the higher the Dncu of mCherry, the lower the vYFP expression.
The four yeast strains were grown in rich media to the log phase, and the expression levels of vYFP and mCherry proteins were inferred from their fluorescent signals, which were simultaneously measured for each cell by fluorescence-activated cell scanning of at least 300,000 cells. We found mCherry expression levels to be significantly different among the four strains (see Materials and Methods). Within each strain, expression levels of mCherry and vYFP are negatively correlated among cells (see Materials and Methods). Hence, the expressions of vYFP cannot be directly compared among strains. Instead, we separated the cells of each strain into three bins on the basis of mCherry expression and then compared vYFP expressions among the four strains for cells with similar mCherry expressions (Figure 3D). We found that, across the range of mCherry expressions shared by the four strains, the higher the Dncu of mCherry, the lower the expression of vYFP (Figure 3D). Furthermore, the vYFP expression-level difference among the strains increases with the mCherry expression level (Figure 3D). Of special interest is the comparison between mCherry-3 and mCherry-4, which clearly shows that it is a low Dncu rather than a high CAI that enhances translational efficiency (Figure 3D). A multivariate regression analysis of all cells from the four strains further demonstrated that Dncu is significantly more important than CAI in explaining the variation of the vYFP signal (P<0.001).
The above results were not due to different random mutations fixed in the genomes of the four strains during our experiments, because the vYFP signals were not significantly different among the strains upon removal of the plasmids (Figure 3E). We also sequenced the entire plasmid DNA from each strain and found no mutation. Using quantitative polymerase chain reaction, we further verified that the vYFP mRNA abundance is not different among the four strains (Figure 3F). Thus, the among-strain variation in vYFP signal must be due to a variation in translation. We also confirmed our results by a finer control of mCherry expression and ruled out the possibility that our observation is a byproduct of potential differences in translational accuracy among different mCherry versions (7; see Materials and Methods). Furthermore, because the accuracy hypothesis is based on CAI and thus predicts a higher vYFP expression in the strain carrying mCherry-4 than that carrying mCherry-3, our results (Figure 3D) are inexplicable by this hypothesis. Similarly, mechanisms resulting from translational errors, such as protein misfolding or aggregation, cannot explain our observation either.
In the experiment, we used vYFP to represent native genes in the yeast genome. However, because vYFP and mCherry have 71/220 = 32% of protein sequence identity, one might ask whether our observation can be generalized. Specifically, could the negative influence of mCherry expression on vYFP expression be caused entirely by the similarity in codon usage between mCherry and vYFP? We measured the codon usage dissimilarity between a pair of genes by a Euclidian distance and examined the distribution of this distance between each mCherry version and all yeast genes (Figure S8). The distribution is approximately bell shaped and the distance between mCherry and vYFP falls in the central part of the bell, suggesting that mCherry is no more similar to vYFP in overall codon usage than to average yeast genes. Furthermore, our results cannot be explained by amino acid similarity between mCherry and vYFP, because all mCherry versions have the same amino acid sequence and should not differentially affect vYFP expression through amino acid usage. Thus, our observation from vYFP can be extrapolated to native genes in the yeast genome.
If translational efficiency is maximized when the cellular codon usage follows the proportional rule, why do highly expressed genes necessarily prefer codons with highly abundant cognate tRNAs and have stronger CUB than lowly expressed genes? We hypothesize that these phenomena are due to differential selective coefficients associated with synonymous mutations occurring in highly expressed and lowly expressed genes in the regain of the codon-tRNA balance upon a genetic perturbation. Let us imagine an amino acid with two synonymous codons (codon1 and codon2) that each uses a distinct tRNA species (tRNA1 and tRNA2) and assume that the present codon usage follows the proportional rule. Now, if the proportion of tRNA1 rises due to a mutation, natural selection will promote the fixations of synonymous mutations from codon2 to codon1 to reestablish the codon-tRNA balance. Such advantageous mutations occurring in highly expressed genes affect tRNA usage more than those occurring in lowly expressed genes and hence have a greater selective advantage and are fixed faster. This difference becomes even bigger when clonal interference [34] is considered. As a result, highly expressed genes use more codon1 and fewer codon2 than before and show stronger CUB. The contrasting scenario, in which the tRNA usage is rebalanced by frequent use of codon1 in lowly expressed genes, requires many synonymous substitutions in many lowly expressed genes, which will not happen because it takes much longer than rebalancing the tRNA usage by increasing codon1 frequency in highly expressed genes. Indeed, in a computer simulation of codon usage evolution that starts from the equal usage of 4 synonymous codons whose cognate tRNAs have different concentrations, the final usage of the codons, after 500 generations of random mutation, genetic drift, and natural selection for translational efficiency, follows the proportional rule (Figure 4A). More importantly, the preferential use of high-concentration tRNA species and strong CUB in highly expressed genes are seen from both the average of 1000 simulation replications (Figure 4B) and any one replication (Figure 4C). The standard deviations presented in Figure 4B indicate an extremely low probability for CUB to be stronger or a preferred codon to be used more frequently in lowly expressed genes than highly expressed genes. As expected, the phenomena in Figure 4 disappear when the natural selection for translational efficiency is removed in the simulation (Figure S9). These observations support our model that the high CAI of highly expressed genes is a byproduct of natural selection for an overall cellular efficiency in translation, rather than the direct product of stronger selection for translation efficiency in more highly expressed genes [6].
Analogous to synonymous codon usage, we predict that the optimal amino acid (or nonsynonymous codon) usage in speeding up translation is in proportion to the corresponding tRNA concentrations. Indeed, amino acid frequencies inferred from transcriptome data were reported to correlate positively with the corresponding tRNA gene copy numbers in yeast [35] and C. elegans [36]. More importantly, actual amino acid usage is significantly closer than random usage to our predicted optimal (i.e., the diagonal line in Figure 5A; P<10−6, simulation test). This phenomenon is also true in all other model eukaryotes examined, although the level of match between the observation and prediction varies among species (Figure 5A). Transcriptomic amino acid usages instead of proteomic amino acid usages are plotted here because the latter are unavailable for most species. Nevertheless, S. cerevisiae data showed an almost perfect correlation between transcriptomic and proteomic amino acid usages (Figure S10), indicating that the former is a good proxy for the latter. We also predict a positive correlation between aminoacyl tRNA synthetase concentration and corresponding tRNA concentration to enhance the efficiency of amino acid charging. Such a correlation is indeed found in S. cerevisiae (r = 0.45, P = 0.03; Figure S11).
If amino acid frequencies are in perfect proportion to tRNA concentrations, the mean CST for an amino acid should not vary among amino acids. This uniformity, however, is not observed in yeast (Figure S12), suggesting that amino acid usage is only roughly proportional to tRNA concentrations (Figure 5A), which may be due to mutational bias [37] or antagonistic selective pressures from factors such as physiochemical properties [38] and synthetic costs [39] of various amino acids. Our model predicts that the average CST of an amino acid increases with the decrease of the relative availability of tRNAs for the amino acid. Indeed, a negative correlation exists between the tRNA availability and CST for the 20 amino acids (Pearson's r = −0.40, P = 0.03, permutation test; Figure 5B). This finding reconfirms tRNA shortage in translation, explains in part why CSTs of nonsynonymous codons vary, and indicates compromised translational efficiency due to other fitness effects of amino acid usage.
Results from several earlier experiments are consistent with the role of CUB in enhancing translational efficiency or reducing ribosome sequestering [12], [17]. For example, when expressing many synonymous versions of a green fluorescent protein (GFP) gene in E. coli, Kudla and colleagues reported that strains harboring high-CAI GFP genes tend to grow faster than those harboring low-CAI GFP genes, despite the lack of a correlation between the GFP protein expression level and its CAI [12]. Although these authors found no correlation between CAI and protein misfolding, their experiment was unlikely to be sensitive enough for quantifying GFP misfolding [12]. Thus, it could not rule out the possibility that the observed variation in fitness was entirely caused by CUB's influence on translational accuracy. By contrast, we were able to demonstrate CUB's impact on translational efficiency after excluding its impact on translational accuracy.
A recent study in E. coli showed that the ribosome shortage induced by over-expression of unneeded proteins can be alleviated by physiological adaptation in 30 to 40 generations, owing to the manufacture of additional ribosomes [40]. This finding suggests that the disadvantage of suboptimal codon usage may also be mitigated by physiological adaptation. Nevertheless, physiological adaptation takes time. If the growth rate fluctuates rapidly due to frequent environmental changes, the fitness of the individual with suboptimal codon usage is expected to be much lower than the individual with balanced codon usage.
We hypothesized and demonstrated that translational efficiency is optimized by codon-tRNA balance. This new model of translational efficiency by unequal codon usage differs substantially from the prevailing model (Table 1). One critical piece of evidence for our model is similar CSTs of synonymous codons in wild-type yeast. Our CST estimation is based on the assumption that the time a codon occupies the ribosomal A site equals the waiting time for the cognate tRNA. Our estimates of all CSTs would be biased upward to a similar level if downstream “traffic jams” happen during translational elongation. However, a recent study suggested that downstream traffic jams are unlikely, due to slow “ramps” at the beginning of an mRNA [21]. Furthermore, even if downstream traffic jams occur, it should affect synonymous codons as well as nonsynonymous codons and thus cannot explain why only synonymous codons but not nonsynonymous codons have similar CSTs.
Over two decades ago, Curran and Yarus indirectly estimated relative CSTs for 29 sense codons in E. coli, under the assumption that the probability of a frame shift in the translation of a codon is proportional to the CST of the codon [41]. They reported that only codons of very low CSTs tend to be preferentially used [41]. However, because their fundamental assumption about the frame-shift rate is incorrect [42], their CST estimates are unlikely to be correct. It is also possible that prokaryotes and eukaryotes have some differences in using CUB to regulate translational efficiency (e.g., translational attenuation in prokaryotes). In another E. coli study, Sorensen and colleagues reported faster translation of a multicopy-plasmid-borne lacZ gene when a segment of the gene comprises mainly preferred codons than when it comprises mainly unpreferred codons [43]. This result cannot be used to infer relative CSTs of synonymous codons in wild-type cells, because the extremely high expression of synonymous versions of the endogenous lacZ gene from plasmids potentially breaks the codon-tRNA balance and alters CSTs. Nevertheless, their observation is fully compatible with our finding of different levels of translational efficiency induced by the expressions of different synonymous versions of mCherry. Several other studies reported similar findings [25], [44]. Recently, some authors calculated CSTs by assuming that the CST of a codon is determined by the relative concentrations of its cognate, nearly cognate, and non-cognate tRNAs without considering tRNA shortage or using ribosome profiling data [45]. Because of the violation of the fundamental assumption they made, their estimates are likely to be incorrect. Indeed, their estimated CSTs would predict a slower translation of mCherry version 3 than 4, contradictory to our experimental result (Figure 3D). While the present work was under review, Ingolia and colleagues reported estimates of translational elongation speeds in mouse embryonic stem cells using a pulse-chase strategy that does not involve expressions of heterologous genes [46]. Although their method is different from ours, their finding of similar elongation speeds among synonymous codons is highly consistent with our results from yeast.
Our discoveries require reinterpretation of several earlier observations. For example, higher prevalence of codons with abundant cognate tRNAs in genes with higher expressions is often interpreted as a result of a stronger demand for fast translation of more abundant proteins [19]–[20]. This interpretation is not supported by our results. Rather, we suggested and demonstrated by simulation that, the selection coefficient for synonymous mutations that help achieve the codon-tRNA balance is greater in highly expressed genes than in lowly expressed genes, leading to quicker and more acquisitions of codons with abundant cognate tRNAs in the former than in the latter. In this regard, our results support that CUB serves as a global strategy to enhance the efficiency of the translation system [12], [47].
Within an organism, the transcriptome can vary among cell cycle stages, developmental stages, and tissues. How do such variations affect the codon-tRNA balance? We found pairwise Pearson's correlations in transcriptomic usage of all 61 sense codons to be nearly 1 among different time points in the S. cerevisiae mitotic cell cycle (Figure 6). We further analyzed the transcriptomic usage of all 61 codons across tissues and/or developmental stages in the worm, fruitfly, and human. If multiple replications of the same cell type exist in a dataset, we randomly chose one replication in our analysis. Similarly high correlations were observed among different cell types within species (Figure 6). By contrast, the correlation is generally below 0.5 between any pair of the four species examined here. The high correlation in codon usage across cell cycle stages, developmental stages, and tissues of the same species is likely due to house-keeping genes, which are always highly expressed. Thus, within-organism gene expression variations have little impact on the maintenance of the codon-tRNA balance. Further, tRNA concentrations may covary with the transcriptomic codon usage to maintain the codon-tRNA balance across tissues [48].
A byproduct of our CST estimation is the translational initiation rate of each gene. We found that the translational initiation rate is significantly positively correlated with the mRNA concentration (ρ = 0.34, P = 6×10−81), suggesting a coordinated regulation of gene expression at the transcriptional and translational levels. We also observed a strong positive correlation between the translational initiation rate and CAI (ρ = 0.51, P<10−196), suggesting that CAI provides a moderate amount of information about the translational initiation rate. This may explain why the protein concentration correlates with the product of mRNA concentration and CAI better than with the mRNA concentration alone [49]. Several studies revealed reduced mRNA stability near the translation initiation site, suggesting that the reduced stability may enhance the translational initiation rate [12], [32]–[33]. Indeed, we found a weak but significant positive correlation between the reduction in mRNA stability [32] and our estimated translational initiation rate (ρ = 0.08, P = 1×10−5).
Given that CUB improves both translational efficiency and accuracy, one wonders whether one of these effects is a side-effect of the other. For instance, it was previously suggested that the variation in translational accuracy among synonymous codons may be a byproduct of the variation in translational efficiency, because (i) most translational errors are believed to occur during codon selection, (ii) codon selection has been assumed to be faster for preferred codons than unpreferred codons, and (iii) faster codon selection is thought to result in fewer errors [50]. Because our result invalidates assumption (ii) for wild-type cells, the above argument no longer holds. Thus, even though translational accuracy may be affected by relative concentrations of tRNAs in engineered yeast cells with grossly imbalanced codon-tRNA usage [51], this impact is not expected in wild-type cells because our results strongly suggest that isoaccepting tRNA species have effectively the same concentrations in wild-type cells. In addition, the enrichment of preferred codons at evolutionarily conserved amino acid residues cannot be explained by the translational efficiency hypothesis [7]–[10]. Furthermore, experimental data showed that translational accuracies of iso-synonymous codons vary [52], suggesting that the variation in accuracy cannot be entirely caused by the variation in cognate tRNA concentration, because iso-synonymous codons use the same cognate tRNA. Rather, comparative genomic analyses strongly suggest that translational accuracy is likely to be intrinsically different among synonymous codons [1], [53]. Further, we were able to establish CUB's impact on translational efficiency even after we controlled its impact on translational accuracy (Figure 3, Figure S7). In addition, because translational accuracy is not entirely determined by translational efficiency [7]–[10], the proportional rule, which is predicted from selection for efficiency, is not predicted from selection for accuracy, especially because translational errors at different residues have different fitness effects. Thus, the impact on efficiency cannot be a byproduct of the impact on accuracy. Taken together, we conclude that translational accuracy and efficiency are two separable benefits of CUB.
Let us compare three evolutionary models of CUB that differ in the roles of translational accuracy and efficiency as the selecting agent. We also consider mutational bias and genetic drift, two known factors in the evolution of CUB, in these models. In model I, translational efficiency is the sole selecting force (Figure 7). This model predicts co-evolution of codon usage and cognate tRNA concentrations and a codon-tRNA balance at which the relative frequency of a synonymous codon (pi) equals the relative abundance of its cognate tRNA (qi). The expected values of pi = qi are determined by the mutational bias, which directly affects codon usage and indirectly affects tRNA concentrations. However, this model cannot explain the observation that, although preferred codons of an amino acid vary among species, this variation decreases substantially (but does not disappear) after the control of genomic GC content [1]. For example, GTT and GTA both code for valine and have the same GC content, but GTT is frequently used as the preferred codon when the genomic intergenic GC content is below 50% [1]. When the GC content exceeds 50%, GTG rather than GTC is often used as the preferred codon for valine [1]. This observation suggests that, in addition to translational efficiency, there is a separate selecting force with a relatively constant direction.
In model II, translational accuracy is the sole selecting agent on CUB (Figure 7). The demand for translational accuracy, coupled with the mutational bias, determines the expected CUB, whereas selection for translational efficiency determines tRNA concentrations based on codon frequencies. The phenomenon of stronger CUB in more highly expressed genes is explainable by the protein-misfolding-avoidance hypothesis which predicts that highly expressed genes are translated more accurately by using accurate codons more frequently [7], [54]. Model II predicts that, after the control for the mutational bias, accurate codons are always the preferred codons in a species. If the translational accuracy of a codon is an intrinsic property of the codon and does not vary among species [29], we should observe no variation in the choice of preferred codons, after the control of mutational bias. This prediction, however, is incorrect, because preferred codons are not always the same in different species with the same mutational bias [1], [29]. A more rigorous test of this model is to compare the accurate and preferred codons of each amino acid in a species, because model II predicts a complete match between them. For each codon, we calculated an odds ratio by the relative use of the codon over other synonymous codons at conserved amino acid positions divided by that at non-conserved amino acid positions; the synonymous codon with the highest odds ratio is regarded as the most accurate codon because it is most preferentially used at important amino acid positions [7]–[10]. By comparing S. cerevisiae with its relative S. bayanus, we identified conserved and non-conserved amino acid positions. We calculated the odds ratio for each codon in each gene and then combined the odds ratios from all genes using the Mantel-Haenszel procedure [23]. By definition, the preferred codon of an amino acid is the one with the highest RSCU'. We found that, in 6 (Ala, Asp, Gly, His, Thr, and Val) of the 18 amino acids that have at least two synonymous codons, the codon with the highest odds ratio is different from the codon with the highest RSCU' (Figure 8). Furthermore, for three amino acids (Asp, His, and Thr), the codon with the highest RSCU' has an odds ratio significantly lower than 1 (Figure 8). We also used the 10% most highly expressed genes to calculate odd ratios; 8 (Ala, Arg, Asp, Cys, Ile, Leu, Thr, and Val) of the 18 amino acids show mismatches between the codon with the highest RSCU' and the codon with the highest odds ratio (Figure 8). These results provide unambiguous evidence for the inadequacy of model II.
In model III, selections for translational accuracy and efficiency jointly determine CUB (Figure 7). Let us consider three types of synonymous mutations with regard to their impacts on translational accuracy and efficiency. First, a synonymous mutation is likely to be fixed when it enhances both translational accuracy and efficiency, but is likely to be lost when it decreases both. Second, a synonymous mutation may increase the accuracy but reduce the efficiency. One possible outcome is that selection for higher accuracy will gradually alter the codon usage, which is followed by tRNA concentration changes that recover the loss of efficiency. Eventually, accurate codons will be the preferred codons. Alternatively, selection for higher accuracy may not be able to alter the codon usage permanently if the loss of efficiency is either too large or cannot be recovered by a corresponding tRNA change as quickly as the switch back of the codon usage. Consequently, accurate codons cannot become the preferred codons and the system is trapped in a local fitness peak that is the maximum for efficiency but not accuracy. For example, while codon CCA is more accurate than CCT for proline (Figure 8), there are still about a quarter of bacterial species with GC%<40 that use CCT as their preferred proline codon [1], suggesting that it is not rare for codon usage to be trapped in a local fitness peak. Third, a synonymous mutation may increase the efficiency but reduce the accuracy when the system is at a codon-tRNA imbalance. Although the fate of this mutation is determined by the relative strengths of the two forces, changes of tRNA concentrations could resolve the conflict better because they can increase efficiency without reducing accuracy. So, the final codon usage pattern will also depend on the rate of mutations that alter tRNA concentrations. While the quantitative aspects of model III require further exploration, it is clear that the model is able to explain, at least qualitatively, both the matches and mismatches between the accurate and preferred codons (Figure 8). It is also able to explain the codon-tRNA balance and the phenomenon of stronger CUB in genes with higher expressions. Thus, model III is most compatible with and best supported by available data. In addition to translational accuracy and efficiency, synonymous codon usage of individual genes may also be shaped by other forces, for example, those related to RNA splicing and stability [55]. But these forces are gene-specific and do not create genomic patterns of CUB.
Synthetic biology designs and constructs novel biological functions not found in nature. It has long been known that, in many but not all cases, increasing the CAI of a transgene boosts its protein expression [12], [56]–[57]. Different protein expression levels of synonymous transgenes are likely caused by CST differences created by various degrees of codon-tRNA imbalance induced by transgene expressions. Consistent with this idea, overexpression of rare tRNAs of E. coli (the bio-reactor) can rescue the tRNA depletion when heterologous human genes are expressed in E. coli [56]. When an artificially designed gene is added to a host cell, the potential imbalance between the overall cellular codon usage and the tRNA pool also affects the expressions of native genes and hence the growth of the host cell. We showed that Dncu, a newly devised index measuring the distance in codon usage between the transgene and the host cell, is an accurate indicator of the impact of per transgene protein molecule production on the expressions of native genes. We demonstrated that it is the Dncu rather than CAI of the transgene that predicts its impact on the host protein expression. Therefore, Dncu should be considered in synthetic biology when the impact of transgene expression on host gene expressions is a concern. Further, when genes from multiple species are assembled into a synthetic genome, designing tRNA gene numbers in proportion to the usage of their cognate codons will likely make protein expressions in the entire cell most efficient.
The yeast ribosome profiling data [18] were downloaded from Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/) under accession number GSE13750. Gene expression and protein expression levels were from http://web.wi.mit.edu/young/expression/ [58], http://www.imb-jena.de/tsb/yeast_proteome/ [59], and the supplementary data of a previous study [60]. Transcriptomic data for the yeast mitotic cell cycle were from a previous study [61]. Gene sequences and reading frames were downloaded from Saccharomyces Genome Database (SGD, www.yeastgenome.org). Numbers of tRNA gene copies were retrieved from an earlier study [22].
Gene expression levels in A. thaliana, D. melanogaster, M. musculus, and H. sapiens were downloaded from Gene Expression Omnibus (GDS416, GDS2784, GDS592 and GDS596, respectively). Gene expression levels in S. pombe and C. elegans were retrieved from two earlier studies [62]–[63], respectively. Peptide and cDNA sequences of S. pombe, A. thaliana, C. elegans, D. melanogaster, M. musculus, and H. sapiens were from Ensembl (www.ensembl.org/). Numbers of tRNA gene copies in the above species were obtained from the genomic tRNA database (http://lowelab.ucsc.edu/GtRNAdb/).
Using the S. cerevisiae ribosome profiling data [18], we identified codons docked at the ribosomal A site, from the Illumina Genome Analyzer sequencing reads. By comparing the observed codon frequencies in the ribosome profiling data with the expected codon frequencies estimated from mRNA-Seq data generated under the same condition in the same experiment, we calculated the relative CSTs of all 61 sense codons. Although Illumina sequencing may be biased toward certain sequences or nucleotides [64], this bias affects the mRNA-Seq and ribosome profiling data equally and thus will not affect our estimation of CST. For a sequencing read from the ribosome profiling data, nucleotide positions 16–18 were considered to be at the ribosomal A site where codon selection occurs [18]. Only those reads with exactly 28 nucleotides and 0 ambiguous sites were used to ensure the accurate determination of positions 16–18. We calculated the fraction of in-frame codons by comparing the read sequences with annotated yeast coding sequences. Consistent with what was previously reported [18], the majority of codons at positions 16–18 were in-frame in the ribosome profiling data. In the mRNA-Seq data, the fraction of each phase was close to one third, as expected. All out-of-frame codons were excluded. The probability of incorrect codon assignment was low, because only codons misaligned by at least 3 nucleotides may be assigned incorrectly. Transposons and uncharacterized genes were removed. Our CST estimation procedure (Figure S1) is as follows.
We first calculated fi, the observed frequency of codon i, in the ribosome profiling data by(1)where cij is the count of codon i in mRNA j positioned at the ribosomal A site measured by ribosome profiling and N is the number of genes with ribosome profiling data (N>3000 for both rich and starvation conditions). The expected ribosome footprint frequencies of codon i (Fi) when all codons have equal CST can be calculated based on the frequency of the codon in the mRNA-Seq data using(2)where Rj is the translational initiation rate of mRNA j and Cij is the count of codon i in mRNA j measured by mRNA-Seq. Then, the relative codon selection time for codon i is calculated by(3)We used an iterative approach to estimate the translational initiation rates that appear in Eq. 2. We first used Rj = 1 for all j. After the CST is calculated for each codon, the elongation rate ej of mRNA j (i.e., the number of codons translated per unit time) is calculated by(4)where Lj is the number of codons in each molecule of mRNA j and Dij is the number of codon i in each molecule of mRNA j. The translational initiation rate Rj can be estimated from(5)where dj is the ribosome density on mRNA j (i.e., the number of ribosomes per codon) and can be estimated by(6)We then used the newly estimated translational initiation rates to calculate CSTs. After 10 iterations, CST estimates converge (Figure S2) and are considered as our final estimates. Because our estimates of CSTs are relative values, we rescaled them by setting the maximal observed value at 1.
CST estimates from different experimental replicates were highly correlated (r = 0.79, P = 6×10−14) and were thus pooled for the rest of the analysis. Three different sets of initial values of translational initiation rates (uniform, proportional to CAI of each gene, inversely proportional to CAI) were used in CST estimation and they resulted in identical estimates of CSTs (Figure S3A, S3B). Thus, CST estimation does not depend on the initial values of R. The standard errors of the CST estimates were estimated by bootstrapping genes present in the ribosomal profiling data 1000 times. The CST estimates from two different media (rich and starvation) are also very similar (Figure S3C). To ensure no mistake in the estimation of CST, the first two authors of this paper independently derived the formulas, wrote the computer programs, and estimated the CSTs, and their results were virtually identical.
There are two commonly used measures of synonymous codon usage bias. The first is the relative synonymous codon usage (RSCU), defined by the frequency of a codon relative to the average frequency of all of its synonymous codons in a set of highly expressed genes [19]. Codons with RSCU>1 are preferred and those with RSCU<1 are unpreferred. To compare the usage of all 61 sense codons, we also used RSCU' = RSCU/n, where n is the number of synonymous codons of an amino acid. RSCU' of a codon is the proportion of use of a given codon among synonymous choices in a set of highly expressed genes. The second commonly used measure of synonymous codon usage bias is the codon adaptation index (CAI), which is calculated for a gene, and measures its usage of high-RSCU codons [20]. Briefly, CAI of a gene is the geometric mean of RSCU divided by the highest possible geometric mean of RSCU given the same amino acid sequence. CAI is a positive number no greater than 1. The greater the CAI, the more prevalent are preferred codons in the gene.
We first selected 200 most highly expressed genes based on a previous study [59]. Sixteen of these genes did not have expression information in another study [58] and 4 had expression levels lower than 4 times the genomic average (2.7 mRNA/cell reported in an earlier study [58]). The remaining 180 highly expressed genes were used to calculate RSCU and RSCU' for each codon. Our RSCU estimates were highly correlated with those previously reported [20] (r = 0.995, P<0.001, permutation test). CAI was calculated for each yeast gene and for each version of mCherry based on the RSCU values obtained above, following a previous study [20].
We also estimated the effective number of codons (Ncp) for each gene, after controlling the GC content of the gene [65]–[66]. We separately estimated the frequency (f) of each of the 61 sense codons in each gene. We then estimated Spearman's rank correlation (ρ) between Ncp and f among all genes for each codon. Among synonymous codons, those with more negative ρ values are considered to be more preferred [1]. This dataset was used in Figure S4 only.
It has been reported that the physiological concentration of the ternary complex is ∼200 nM for Phe tRNA and Lys tRNAs in E. coli [67]. Because the number of Phe tRNA and Lys tRNA molecules per cell is 1830 and 4300, respectively [68], we calculated that the Phe tRNA concentration is 1830/(6.02×1023)/(1.1×10−15) = 2.8×10−6 M = 2800 nM, where 6.02×1023 is the number of molecules per mole and 1.1×10−15 liter is the average volume of an E. coli cell. Similarly, Lys tRNA concentration is estimated to be 6500 nM. Thus, about 200/[(2800+6500)/2] = 4.3% of tRNAs are in ternary complexes. Because there are ∼1.2×104 ribosomes per E. coli cell [68], ribosome concentration is ∼18,000 nM. Thus, the ratio in the concentration of ternary complexes to that of ribosomes is expected to be 200×20/18000 = 0.22, if Lys and Phe can represent all 20 amino acids in ternary complex concentration.
Without loss of generality, we assume that an amino acid is encoded by synonymous codons 1 and 2, which are respectively recognized by isoaccepting tRNAs 1 and 2. Let the relative usage of the two codons be p1 and p2 = 1−p1 and the relative concentrations of the two tRNAs be q1 and q2 = 1−q1, respectively. Let the codon selection time for the two synonymous codons be t1 and t2, respectively. Thus, the expected codon selection time for the amino acid concerned is t = p1t1+p2t2. When tRNAs are in shortage, the local concentrations of tRNA 1 and 2 are aq1/p1 and aq2/p2, where a is a constant. Because codon selection time is proportional to the inverse of the local tRNA concentration, we have , where b is another constant. The above formula can be simplified to . It is easy to find that t reaches its minimal value of b/a when and . In other words, the expected codon selection time is minimized and thus translational efficiency is maximized when relative synonymous codon frequencies equal relative tRNA concentrations. Under this condition, codon selection time equals b/a for both codons and local tRNA concentration equals a for both tRNAs. A full treatment considering tRNA cycle and kinetics gave the same result [31].
We measured the Euclidian distance and Manhattan distance in synonymous codon usage from the observed values to the values predicted from the observed tRNA fractions using the proportional rule. To evaluate whether the observed distances are shorter than expected by chance, we conducted a computer simulation with 106 replications under random codon usage. That is, the frequency of a synonymous codon is uniformly distributed between 0 and 1 with the constraint of the total frequency of all synonymous codons being 1. We then obtained the distribution of the distance between a random codon usage and the codon usage predicted from the observed tRNA fractions. We also conducted a second simulation with 106 replications, in which tRNA factions vary randomly according to the above uniform distribution. We then obtained the distribution of the distance between the observed codon usage and that predicted from random tRNA fractions. This way, the potential confounding effect of genomic GC content on the assumed null distribution of codon usage becomes irrelevant to the test. We similarly tested the square rule and the truncation rule. Results from the first simulation are presented in Figure 2D, while those from the second simulation are in Table S1.
We developed an index, distance to native codon usage (Dncu), to measure how different the codon usage of a (heterologous) gene is from the overall codon usage of the host cell, which is presumably balanced with tRNA concentrations. First, the Euclidean distance in synonymous codon usage between the heterologous gene and the host is calculated for each of the 18 amino acids with at least two synonymous codons by(7)where Yij is the fraction of codon j among the synonymous codons of amino acid i for the heterologous gene and Xij is the fraction of codon j among the synonymous codons in the host transcriptome, ni is the number of synonymous codons for amino acid i. Dncu of the gene is defined as the weighted geometric mean of Di, or(8)where k≤18 is the number of amino acid types encoded by the gene excluding Met and Trp, which have no synonymous codons, mi is the number of amino acid i found in the protein, and l is the protein length excluding Met and Trp residues. By definition, Dncu is between 0 and 1.
The mCherry gene sequence was obtained from a previous study [69]. We designed four synonymous DNA sequences encoding the same mCherry peptide (Figure S5). The first 56 nucleotides were the same for all four sequences to avoid potential effects on the mRNA secondary structure, which affects protein translation [12], [32]–[33]. The GC contents of the four sequences (42–44%) were also made similar to each other and to the average value in yeast coding sequences (40%). In all sequences, synonymous codons were randomized in order and thus were unlikely to cause differences in order-related effects [27]. The different versions of mCherry DNA sequences were synthesized by Blue Heron Biotechnology. They were cloned into p426GPD [70] at SpeI and XhoI (New England Biolabs; Promega) and are under the control of the GPD promoter. The plasmids were subsequently transformed individually into a haploid yeast cell (BY4742) with vYFP [71] inserted into Chr XII [72]. The genotype of the cell is MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0 hoΔ0::PGPD-Venus.
We measured the expressions of mCherry and vYFP in log growth phase in Yeast extract/Peptone/Dextrose (YPD) media by florescence-activated cell scanning (FACSCalibur, BD). Fluorescence of mCherry was measured from FL4 with a 670 nm pass filter and fluorescence of vYFP was measure from FL1 with a filter having a 30 nm bandpass centered on 530 nm. Yeast cells with mCherry fluorescence signals greater than the BY4742 negative control cells (i.e., mCherry fluorescence signals >10) were gated. We retrieved the forward scatter (FSC, which is proportional to cell size) and mCherry and vYFP fluorescence signals for all gated cells. The expression levels of fluorescent proteins were defined as their fluorescence signals divided by FSC. The mean mCherry expression level is 3.388±0.002, 6.468±0.007, 14.003±0.032, and 14.544±0.022 for the strains carrying mCherry-1, 2, 3, and 4, respectively. Expression levels of mCherry and vYFP were negatively correlated for each strain (mCherry-1: r = −0.22; mCherry-2: r = −0.57; mCherry-3: r = −0.60; mCherry-4: r = −0.62; P<2.2×10−16 in all cases). All gated cells were then grouped into 3 (Figure 3D) or 15 (Figure S7) bins with equal mCherry expression ranges. For each genotype, multiple independently transformed strains were examined on different days, but the results were highly similar. We thus combined all results obtained from different strains of the same genotype. The total numbers of cells measured were 456333, 648792, 352863, and 793832, respectively, for the strains carrying mCherry-1, 2, 3 and 4 (Figure 3B). To confirm that our results were not due to random secondary mutations, we removed the plasmids from each strain by using 5′-FOA media to select against the plasmids, and then measured the vYFP fluorescence intensities. We also sequenced the entire plasmid DNA from each of the four strains.
To compare the vYFP mRNA levels among strains, we extracted the total RNA (RiboPure-Yeast Kit, Ambion) from three independently transformed strains of each genotype. The total RNA was reversely transcribed into cDNA (Moloney Murine Leukemia Virus Reverse Transcriptase, Invitrogen) with random hexamer primers. The vYFP mRNA level was measured by quantitative polymerase chain reaction (7300 Real-Time PCR System, Applied Biosystems) with ACT1 as an internal control. The primers for vYFP are 5′ – CATGGCCAACACTTGTCACT– 3′ and 5′ –TACATAACCTTCGGGCATGG– 3, while the primers for ACT1 are 5′ - CTGCCGGTATTGACCAAACT - 3′ and 5′ – CGGTGATTTCCTTTTGCATT – 3′.
The software package RELAIMPO (http://cran.r-project.org/web/packages/relaimpo/) was used for a multivariate regression analysis of the yeast experimental data from all cells of the four strains. We compared the relative importance of Dncu and CAI in explaining the among-cell variation in vYFP signal by the LMG method and used 1000 bootstrap replications to determine the statistical significance. Use of other methods (LAST, FIRST, and PRATT) implemented in RELAIMPO gave similar results.
Proponents of the translational accuracy hypothesis might argue that, because different synonymous codons have different mistranslation rates [52], [73] and preferred codons are considered to be more accurately translated than unpreferred codons [7], the mCherry with a low CAI is expected to produce fewer functional protein molecules than the mCherry with a high CAI even when the same numbers of protein molecules are produced. In other words, using red florescent signals may have led to a more severe underestimation of protein expression for the mCherry with a low CAI than for that with a high CAI. The average mistranslation rate has been estimated to be ∼5×10−4 per codon, and unpreferred codons have been posited to undergo mistranslation five times as often as preferred codons [7]. Based on these numbers and the CAIs of the four mCherry versions (Figure 3B), we assume that the mistranslation rate is 10×10−4, 8×10−4, 5×10−4, and 2×10−4 per codon for mCherry-1 to mCherry-4, respectively. Let us further assume that no mistranslated protein fluoresces. Given the length of mCherry (236 amino acids), we expect that 11.8%, 9.44%, 5.9%, and 2.36% of mCherry-1 to mCherry-4 proteins respectively fail to fluoresce due to mistranslation. On this assumption, we corrected mCherry expression levels from the observed florescent signals. We also conducted a better control of mCherry expression among strains by dividing cells of each strain into 15 bins based on the above corrected mCherry expression (Figure S7). Again, we observed a lower vYFP expression when the Dncu of the mCherry gene is higher, across the range of mCherry expressions shared by the three strains (Figure S7). This result is conservative, because only a minority of mistranslations are expected to prevent fluorescence, and it is likely that we have overcorrected the effect of mistranslation.
We simulated the evolution of synonymous codon usage in an asexual haploid unicellular digital organism. In this organism, we focused on a single amino acid with four synonymous codons (codon1 to codon4) that are respectively recognized by four distinct tRNA species (tRNA1 to tRNA4). We assume that the relative concentrations of the four tRNA species are 20, 21, 22, and 23, respectively. The digital organism has ten genes with relative (mRNA and protein) expression levels from 20 to 29, respectively. These genes each have 12 codons that are sampled from the four synonymous codons. We started the simulation with exactly the same usage of the four synonymous codons in each gene. Synonymous mutations among codons all have the same rates and the total mutation rate per genome is assumed to be one synonymous change per generation. The relative CST for a codon is assumed to equal the number of times the codon is used in translation divided by the number of corresponding tRNA molecules. The total time (T) required for translating all the proteins can be considered as the generation time. T can be calculated by summing up the CSTs of all codons in all transcripts if there is only one ribosome in the cell. If there are m ribosomes in the cell, the time required would simply be m times shorter. Thus, without loss of generality, we assume m = 1. A strain with a shorter generation has a higher fitness and will spread in the population. Genetic drift is simulated by random sampling of cells for the next generation. The population size is 104 individuals and the simulation lasts for 500 generations. We repeated the simulation 1000 times. Our results did not change when we simulated the evolution for more generations. By contrast, when we removed the natural selection for translational efficiency in simulation, the phenomena observed in Figure 4 disappeared (Figure S9).
Note that, in the simulation, we allow codon usage to evolve while fixing tRNA concentrations. If tRNA concentrations evolve while the codon usage is fixed, we also expect to observe the rebalance of codon-tRNA usage, but the correlation (or the lack of) between CUB and gene expression level will not change during this evolutionary process. In reality, tRNA concentrations and synonymous codon usage likely co-evolve to regain the balance. As long as codon usage is allowed to evolve, we expect stronger CUB to appear in more highly expressed genes, as demonstrated in Figure 4.
|
10.1371/journal.pgen.1005209 | The Nutrient-Responsive Hormone CCHamide-2 Controls Growth by Regulating Insulin-like Peptides in the Brain of Drosophila melanogaster | The coordination of growth with nutritional status is essential for proper development and physiology. Nutritional information is mostly perceived by peripheral organs before being relayed to the brain, which modulates physiological responses. Hormonal signaling ensures this organ-to-organ communication, and the failure of endocrine regulation in humans can cause diseases including obesity and diabetes. In Drosophila melanogaster, the fat body (adipose tissue) has been suggested to play an important role in coupling growth with nutritional status. Here, we show that the peripheral tissue-derived peptide hormone CCHamide-2 (CCHa2) acts as a nutrient-dependent regulator of Drosophila insulin-like peptides (Dilps). A BAC-based transgenic reporter revealed strong expression of CCHa2 receptor (CCHa2-R) in insulin-producing cells (IPCs) in the brain. Calcium imaging of brain explants and IPC-specific CCHa2-R knockdown demonstrated that peripheral-tissue derived CCHa2 directly activates IPCs. Interestingly, genetic disruption of either CCHa2 or CCHa2-R caused almost identical defects in larval growth and developmental timing. Consistent with these phenotypes, the expression of dilp5, and the release of both Dilp2 and Dilp5, were severely reduced. Furthermore, transcription of CCHa2 is altered in response to nutritional levels, particularly of glucose. These findings demonstrate that CCHa2 and CCHa2-R form a direct link between peripheral tissues and the brain, and that this pathway is essential for the coordination of systemic growth with nutritional availability. A mammalian homologue of CCHa2-R, Bombesin receptor subtype-3 (Brs3), is an orphan receptor that is expressed in the islet β-cells; however, the role of Brs3 in insulin regulation remains elusive. Our genetic approach in Drosophila melanogaster provides the first evidence, to our knowledge, that bombesin receptor signaling with its endogenous ligand promotes insulin production.
| Animals need to couple growth with nutritional availability for proper development and physiology, which leads to better survival. Nutritional information is mostly perceived by peripheral organs, particularly metabolic organs such as adipose tissue and gut, before being relayed to the brain, which modulates physiological responses. Hormonal signaling ensures this organ-to-organ communication, and defects in this endocrine regulation in humans often cause diseases including obesity and diabetes. In the fruit fly Drosophila melanogaster, adipose tissue (the “fat body”) has been suggested to play an important role in coordinating growth with metabolism. Here, we show that the Drosophila CCHamide-2 (CCHa2) gene, expressed in the fat body and gut, encodes a nutrient-sensitive peptide hormone. The CCHa2 peptide signals to neuroendocrine cells in the brain that produce Drosophila insulin-like peptides (Dilps) through its receptor (CCHa2-R) and promotes the production of Dilps. Mutants of both CCHa2 and CCHa2-R display severe growth retardation during larval stages. These results suggest that CCHa2 and CCHa2-R functionally connect peripheral tissues with the brain, and that CCHa2/CCHa2-R signaling coordinates the animal’s growth with its nutritional conditions by regulating its production of insulin-like peptides.
| Organisms need to coordinate growth and metabolism with their nutritional status to ensure proper development and the maintenance of homeostasis. In multicellular animals, nutritional information is mostly perceived by peripheral organs. It is subsequently relayed to other peripheral organs or to the central nervous system (CNS), which generates appropriate physiological and behavioral responses. Endocrine systems ensure this type of organ-to-organ communication via hormonal signals secreted from specialized glandular cells. For example, mammalian insulin is secreted from pancreatic β-cells in response to high blood glucose levels; insulin is then received by its receptor in the liver as well as in many other tissues to promote glucose uptake and anabolism, thereby reducing blood sugar levels [1]. In a similar manner, leptin secreted from adipose tissues is received by the hypothalamus, where it acts to alter energy expenditure and food intake [2] [3] [4] [5]. Caloric restriction reduces the secretion of leptin, leading to both an increase in appetite and a decrease in energy expenditure, which is known to be an adaptive response to starvation [6]. These findings demonstrate the significance of peripheral tissues in the maintenance of homoeostasis [7]. However, only a few peripheral hormones have been identified, and the mechanisms by which they regulate an organism's development or physiology in response to external stimuli remain elusive.
It has been reported that the endocrine system of Drosophila melanogaster allows adipose tissue, known as the fat body, to communicate with the CNS in a manner similar to that observed in mammals. This signaling depends on nutritional conditions and ultimately couples growth and metabolism with nutritional status. To date, two pathways have been described. In one pathway described from larvae, the fat body-specific down-regulation of either the Slimfast (Slif) amino acid transporter or the Target of Rapamycin (TOR) nutrient-sensing pathway affects systemic growth, suggesting that a hitherto unidentified amino acid-dependent signal(s) is secreted by the fat body for proper growth control [8]. In a second pathway that was identified in adults, Unpaired-2 (Upd2), which is a functional analogue of leptin, was identified as another fat body-derived growth regulator [9]. The expression of upd2 is both sugar- and lipid-sensitive and is apparently independent of the amino acid-activated TOR pathway [9]. Although no signaling molecules that act downstream of the Slif/TOR pathway have been identified yet, these fat body-derived signals ultimately regulate the production of insulin-like peptides (Drosophila insulin-like peptides; Dilps) secreted from the brain [10] [9].
Dilps are evolutionarily conserved peptide hormones with functions similar to those of mammalian insulin/insulin-like growth factor (IGF), including the control of tissue growth and blood sugar levels in response to nutritional conditions [11,12] [13] [14] [15] [16]. Eight dilp genes exist in the Drosophila melanogaster genome [11] [15] [16]. Unlike mammalian insulin, which is secreted from the pancreas, the major Dilps (Dilp2, -3, and -5) are specifically expressed in bilateral clusters of neurosecretory cells [insulin-producing cells (IPCs)] located in the anteromedial region of the brain hemispheres [11] [12]. With regard to the regulation of insulin-like peptides, the knockdown of the Slif/TOR pathway or upd2 in the larval fat body results in the down-regulation of Dilp2 secretion [9,10]. Upd2, a type-I cytokine, activates the JAK/STAT pathway through its receptor Domeless (Dome) [17] [18]. Dome is expressed in the GABAergic neurons juxtaposed to the IPCs in the adult brain. Activation of Dome by Upd2 blocks GABAergic inhibition of the IPCs and thereby facilitates Dilp secretion [9]. Therefore, signaling from peripheral tissues to the brain appears to be essential for the regulation of organismal growth and metabolism in response to nutrition availability in Drosophila melanogaster.
In this study, we investigated the roles of CCHa2 and its receptor in growth control in Drosophila melanogaster. CCHa2 was identified as a bioactive peptide that activates a G protein-coupled receptor (GPCR) encoded by CG14593 (now named CCHa2-R) [19]. Strong expression of CCHa2 in the larval fat body and gut motivated us to examine the roles of CCHa2 and its receptor in nutrient sensing and growth control. By generating mutants of CCHa2 and CCHa2-R, we show that CCHa2/CCHa2-R signaling from the periphery to the CNS can control the synthesis and secretion of Dilps. Our results demonstrate that CCHa2 is a novel hormone derived from peripheral tissues and that CCHa2/CCHa2-R form an additional afferent hormonal signaling pathway that coordinates systemic growth with nutrition availability.
We first measured the expression of CCHa2 mRNA in larval tissues using quantitative RT-PCR (RT-qPCR). As shown in Fig 1A, CCHa2 was predominantly detected in the fat body and the gut, with only very low expression detected in the CNS. We also examined CCHa2 expression using CCHa2 antisera, which detected punctate staining in the cytoplasm of the fat cells (Fig 1B and 1C). CCHa2 immunoreactivity was also detected in endocrine cells in the gut as previously reported [20] (Fig 1F). These signals were specific for CCHa2, because they were absent in CCHa2 mutants (Fig 1D and 1E and 1G).
We next examined the nutritional dependence of CCHa2 expression. In this assay, third-instar larvae [72 hours after egg laying (AEL)] were starved for 18 hours on water agar plates, and the relative amount of CCHa2 mRNA in the whole animal was quantified. After starvation, the expression of CCHa2 was significantly decreased, but levels recovered when the larvae were re-fed with yeast paste (Fig 1H).
To determine which nutrient signal controls CCHa2 expression, larvae were re-fed with different substances after starvation. Interestingly, both yeast and glucose induced CCHa2 expression (Fig 1I). It has been reported that the TOR nutrition-sensing pathway is activated by amino acids but not by glucose [10]. Nonetheless, we tested the involvement of this pathway in CCHa2 regulation in the fat body. When the TOR pathway was blocked in the fat body by the overexpression of signaling components TSC1/2, CCHa2 expression was significantly reduced (Fig 1J). (It should be noted that the knockdown of the TOR pathway in the fat body severely affects larval growth [8]; therefore, Lsp2-GAL4, which is expressed in the fat body at the wandering stage – after the growth period [13] – was used to overexpress TSC1/2 in order to avoid secondary effects from a systemic growth defect.) These observations suggest that CCHa2 expression is responsive to glucose and the TOR pathway. Given that glucose alone is sufficient to promote CCHa2 expression, CCHa2 appears to be distinct from the currently unidentified fat body-derived factor previously proposed to act downstream of the Slif/TOR pathway [10].
We performed RT-qPCR assays to examine the larval expression pattern of CCHa2-R (CG14593), which encodes the receptor for CCHa2. As shown in Fig 2A, CCHa2-R mRNA was detected specifically in the larval CNS. Fluorescent in situ hybridization (FISH) of larval brains with a CCHa2-R antisense RNA probe yielded signal in a subset of cells located at the anteromedial region of the brain (Fig 2B and 2C), which contains several types of neuroendocrine cells, including the IPCs [21]. To increase sensitivity for the mapping of CCHa2-R expression, its pattern was indirectly visualized using a GAL4 construct that recapitulates endogenous CCHa2-R expression. For this purpose, we established a transgenic fly line that carries an ~80-kb genomic region containing the entire CCHa2-R coding region, as well as substantial flanking sequence. The coding portion of the first CCHa2-R coding exon was replaced with sequence encoding the strong GAL4::p65 transcriptional activator (Fig 2E; CCHa2-R-GAL4::p65). Consistent with the endogenous expression pattern (Fig 2B and 2C), strong GFP signals were detected in the anteromedial region of the brain in transgenic flies in which CCHa2-R-GAL4::p65 drove the expression of UAS-nls::GFP (CCHa2-R>nlsGFP) (Fig 2F and 2G). Low levels of GFP expression were also observed in a number of cells in the brain and the ventral nerve cord (VNC) (Fig 2F and 2G), in which no CCHa2-R expression was detected by FISH, probably due to its lower sensitivity. When the transgenic larval brains were co-stained with anti-Dilp2 antibody, CCHa2-R>nlsGFP colocalized with the Dilp2 immunostaining (Fig 2H). We also found that neighboring peptidergic neurons that contain neuropeptide F (NPF) and SIFamide (SIFa), both also showed CCHa2-R>nlsGFP-expression (S1 Fig). Thus, CCHa2-R is expressed in a few types of neuroendocrine cells including the IPCs.
Since CCHa2-R is expressed in the IPCs, we examined whether CCHa2/CCHa2-R signaling is involved in insulin regulation. We generated two mutant CCHa2-R alleles, both expected to be nulls. Using gene targeting by homologous recombination [22] [23], most of the CCHa2-R coding region – from the translation-initiating methionine through the middle of the 7th transmembrane domain of the encoded GPCR – was deleted, generating the CCHa2-RKO51-2 allele (Figs 3A and S2A). In a second scheme, transcription activator-like effector nucleases (TALENs) were targeted to sequences near the translation-initiation site of CCHa2-R to create a frame-shift mutation in the CCHa2-R coding region [24]. By injecting pairs of TALEN-encoding mRNAs into wild-type embryos, we generated the CCHa2-RTAL-34 frameshift allele (Figs 3B and S2B and S2C). Although these two mutant alleles are both homozygous-viable, we used them in transheterozygous combination (i.e., CCHa2-RKO51-2/CCHa2-RTAL-34) to avoid the effects of any unexpected secondary mutations.
These mutant alleles were first used to examine whether CCHa2-R is required for the transcription of two Drosophila insulin-like peptide (dilp) genes. It has been reported that dilp2, dilp3, and dilp5 are expressed in IPCs [11] [12]. As previously reported, the expression level of dilp3 is very low during the growth period (S3A Fig) [13] [14], and neither growth nor developmental timing is altered in dilp3-null mutants [25]. Nonetheless, we tested whether dilp3 is under the control of CCHa2-R signaling. As shown in S3B Fig, dilp3 mRNA levels were not altered by CCHa2-R mutations. In contrast, dilp2 and dilp5 are highly expressed in feeding larvae and have substantial influence over larval growth [13] [14] [25]. Therefore, we focused on these two Dilps. For this analysis, total RNA extracted from whole larvae was used, as dilp2 and dilp5 are predominantly expressed in the CNS during the larval stages examined [26,27]. In mid-L3 (96 hours AEL) larvae, dilp2 expression was not significantly altered by the loss of CCHa2-R (Fig 4A). In contrast, the expression of dilp5 mRNA was remarkably reduced in CCHa2-R mutant larvae, compared to control larvae heterozygous for either the CCHa2-RKO51-2 or CCHa2-RTAL-34 allele, regardless of gender (Fig 4B). It was reported that starvation or Slif/TOR inhibition down-regulates the secretion of Dilp2, but not its transcription [28] [10]. We therefore examined the protein levels of Dilp2 in IPCs by staining larval brains with anti-Dilp2. It has been shown previously that, when dilp2 transcription is constant, increased Dilp2 within the cytoplasm of IPCs reflects decreased Dilp2 release into the hemolymph, and intracellular Dilp2 has been used as a sign of Dilp2 retention in the IPCs [10]. Consistent with this, stronger Dilp2 signals were observed in the IPCs of starved wild-type larvae than in those under fed conditions (Fig 4C). When CCHa2-R mutant larvae were fed, their IPCs showed increased Dilp2 immunoreactivity, which was much stronger than that observed in starved or fed control larvae (Fig 4C). To accurately compare the signals, the Dilp2 signal intensity in the IPCs for each condition was normalized against signals of membrane-bound GFP expressed under the control of the dilp2-GAL4 driver as an internal control (because, as noted above, dilp2 transcription is unaltered in the mutants). The quantification results clearly show that levels of Dilp2 protein in the mutant IPCs were significantly higher than those in the wild type (Fig 4D). Dilp5 protein levels were also quantified by the same methods in wild-type and CCHa2-R mutant IPCs. Despite a ~40% reduction in dilp5 mRNA in the mutant IPCs (Fig 4B), Dilp5 protein levels only dropped by about 20%, suggesting decreased release (Fig 4E and 4F). These results suggest that the secretion of both Dilp2 and Dilp5 is severely affected by the loss of CCHa2-R. We also noticed that Dilp2 levels in mutant IPCs under starved conditions were slightly higher than those under fed conditions (Fig 4C and 4D), suggesting that CCHa2-R is not the sole nutrient-sensitive Dilp regulator. Analysis of feeding activity using dyed yeast demonstrated no significant differences in dye ingestion between wild-type and CCHa2-R mutants (S4 Fig), suggesting that down-regulation of CCHa2/CCHa2-R signaling does not affect larval feeding behaviors. Taken together, these results show that CCHa2-R plays a crucial role in regulating the synthesis and secretion of insulin-like peptides in IPCs.
To examine whether CCHa2 acts together with CCHa2-R in the control of Dilps in vivo, we generated animals carrying a defective CCHa2 gene using the CRISPR/Cas system [29]. Short guide RNA was targeted to the sequence corresponding to the CCHa2 peptide-coding region, resulting in the isolation of putative null alleles for CCHa2 (CCHa2CR-1, CCHa2CR-2, and CCHa2CR-3) (Figs 5A and S5). All of these alleles displayed virtually the same phenotypes, so the results for CCHa2CR-1 are shown as representative. First, RT-qPCR analyses showed that the expression of dilp5 mRNA was significantly reduced in CCHa2CR-1/Df hemizygotes compared to heterozygous control larvae regardless of sex (Fig 5B). These results are similar to those seen in CCHa2-R mutants (compare Figs 4B and 5B). Consistent with these reduced dilp5 levels, the body weight of mid-third-instar larvae (96 hours AEL) was markedly lower in the CCHa2 hemizygotes than in the heterozygous control regardless of sex (Fig 5C). These results suggest that CCHa2 and CCHa2-R act together to regulate dilp5 expression in the brain.
To clarify whether peripheral tissues are responsible for the CCHa2-dependent regulation of dilps in the brain, CCHa2 was specifically knocked down in the fat body and gut using targeted RNAi driven by the ppl-GAL4 driver [8] [30]. As shown in Fig 5D, dilp5 mRNA levels were significantly reduced in CCHa2-knockdown larvae, indicating that peripheral tissue-derived CCHa2 activates dilp5 expression in the brain. To further confirm the importance of CCHa2 signaling from the periphery to the CNS, we expressed CCHa2 in the fat body in the CCHa2 mutant background using the cg-GAL4 driver [30]. To ensure the detection of brain-specific changes in dilp5 expression, RNA extracted from the brain was used for RT-qPCR. As shown in Fig 5E, CCHa2 expression in the fat body completely restored dilp5 expression in the brain of CCHa2 mutants. Consistent with the rescued dilp5 expression, the body weight of these larvae was mostly recovered (Fig 5F). These results demonstrate that CCHa2 released from peripheral tissues controls dilp5 expression in the brain.
To clarify whether CCHa2 directly signals to the brain, calcium imaging was performed using ex vivo culture of larval brains expressing the fluorescent calcium sensor GCaMP6s [31] in the IPCs. Dissected wild-type or CCHa2-R mutant brains were immersed in phosphate buffered saline (PBS). Peptides were added to the culture medium, and fluorescence changes from GCaMP6s were measured by live imaging. As shown in Fig 6A and 6B, signal intensities in wild-type IPCs were dramatically increased upon CCHa2 administration (S1 Movie), indicating that no other tissues are required for relaying the CCHa2 signal to the CNS. In contrast, no such increase in signal was observed in CCHa2-R mutant brains upon CCHa2 administration (S2 Movie). The difference in fluorescence intensities between wild-type and CCHa2-R mutant IPCs became significant as early as 2 minutes after CCHa2 administration and lasted at least for 7 minutes (S6 Fig). No increase in signal intensity was observed in wild-type brains treated with ghrelin or nociceptin (S3 and S4 Movies, respectively), mammalian peptide hormones which have no homologues in Drosophila. These results indicate that the observed activation of the IPCs is a specific response to CCHa2 mediated by CCHa2-R.
To examine whether the CCHa2 signal directly activates the IPCs, rather than being relayed by other neurons, CCHa2-R was knocked down specifically in the IPCs. We generated three UAS-CCHa2-R-shRNA lines (#3, 24, 30; S7 Fig). When any one of these shRNAs was specifically expressed in the IPCs, dilp5 mRNA levels were significantly decreased (Fig 6C). These results demonstrate that CCHa2-R is specifically required in IPCs for those cells’ response to the CCHa2 ligand, and suggest that CCHa2 secreted from the peripheral tissues directly signals to IPCs without being relayed by other tissues.
Since insulin-like peptides are major growth hormones in flies, we anticipated that abnormal expression and secretion of Dilps in the CCHa2-R mutants would lead to growth defects. To test this hypothesis, wild-type and mutant animals were weighed at time points from larval through adult stages. As predicted, CCHa2-R transheterozygous mutants weighed markedly less than heterozygous CCHa2-RTAL-34/+ control larvae from 72 to 108 hours AEL (Fig 7A). However, after 108 hours AEL, the mutant larvae showed rapid weight gain, surpassing wild-type weight at 120 hours AEL (Fig 7A). The body weight of the mutant larvae decreased from 144 hours AEL onward, resulting in pupae and adults of normal weight. In concert with these growth defects, CCHa2-R mutants displayed a developmental delay during the larval stages, with the feeding period extended for about 24 hours before entering the wandering stages. This prolonged feeding period resulted in an increase in body weight and delay in pupation and eclosion (Fig 7B). Since broad dilp2 overexpression caused lethality [11], we were unable to examine whether these developmental delays are a consequence of growth defects or whether developmental timing is regulated independently of growth. These results nevertheless suggest that CCHa2-R is a major growth regulator until around 108 hours AEL, after which time other mechanisms may operate to accelerate the growth of the mutant larvae.
Both the down-regulation of dilp5 expression (Fig 8A) and the retention of Dilp2 within the IPCs (Fig 8B and 8C) persisted in the CCHa2-R mutants through the last day of the extended larval stage (up to 144 hours AEL). These results support our conclusion that the reduction in Dilp2 and Dilp5 is not the consequence of feeding defects, because these larvae show abnormal Dilp regulation regardless of active feeding and growth. It has been reported that brain-derived Dilps (Dilp2, -3, and -5) are received by an insulin receptor (InR) in the fat body, leading to transcriptional repression of dilp6 in that tissue [25]. Hence, the reduction in Dilp expression in the brain in CCHa2-R mutants could lead to up-regulation of dilp6 transcription in the fat body. Consistent with this idea, we found that dilp6 mRNA levels were elevated in CCHa2-R mutant larvae (Fig 8D). Furthermore, we found that the removal of dilp6 from the CCHa2-R mutants abolished growth recovery between 96 to 120 hours AEL (Fig 8E). Taken together, these findings indicate that CCHa2/CCHa2-R signaling is a major regulator of Dilps until the mid- to late third-instar larval stages and that the growth recovery observed in later-stage mutants is due to the up-regulation of dilp6 resulting from the impairment of Dilp expression in the brain.
Some peripheral tissues act as monitors of the nutritional environment and metabolic status. Thus, communication between peripheral organs, particularly from metabolic organs such as adipose tissues or the gut, and the brain is imperative for proper development and the maintenance of homeostasis. Here, we have demonstrated that signaling from peripheral tissues to the CNS mediated by the CCHa2 hormone and its receptor is required for the proper regulation of growth in response to nutritional conditions.
A previous study suggested the existence of an amino acid-sensitive Dilp regulator(s) in larvae [10]. This as-yet-unidentified Dilp regulator(s) is regulated by the Slif/TOR pathway, and leucine and isoleucine, positive regulators of TOR signaling, are sufficient to promote the secretion of Dilp2 in both in vivo and ex vivo co-cultures of brain and fat bodies [10]. Our results demonstrated that the TOR pathway is required for CCHa2 expression during the larval stages (Fig 1J). However, feeding with amino acids, including leucine and isoleucine, was insufficient to promote CCHa2 expression (Fig 1I). CCHa2 expression was, however, induced by feeding with glucose (Fig 1I). Therefore, unlike the amino acid-dependent Dilp regulator(s) predicted by Géminard et al. [10], CCHa2 was found to be primarily sensitive to glucose. Some biological substances are produced by the metabolism of specific nutrients. For example, pyrimidine or purine bases are synthesized from amino acids. Therefore, it is possible that CCHa2 is down-regulated when glucose is abundant but other nutrients are not available, to limit growth in inhospitable environments. The reduction of CCHa2 mRNA in TOR-pathway knockdown larvae may recapitulate this scenario (Fig 1J).
In addition to CCHa2, Upd2 was reported to be a glucose-sensitive Dilp regulator expressed in the fat body [9]. The expression of upd2 in adult flies is up-regulated by feeding with a high-glucose or high-lipid diet. CCHa2 and Upd2, however, responded differently when the TOR pathway was disturbed: whereas CCHa2 expression was down-regulated in TOR-pathway-knockdown larvae, upd2 was up-regulated by the inhibition of the TOR pathway in adults [9]. Furthermore, the time course of CCHa2/CCHa2-R signaling is distinct from that of Upd2/Dome signaling. Disruption of upd2 down-regulated animals’ growth from larval to adult stages [9], whereas CCHa2-R mutations reduced growth until late-L3 stages, after which growth was recovered, leading to adults of normal size (Fig 7A). This growth recovery resulted from up-regulation of dilp6 expression (Fig 8D and 8E), which appears to be a consequence of dysregulated brain Dilps. The lack of growth recovery in upd2-knockdown animals in spite of abnormal Dilp production remains unexplained. Nevertheless, these results indicate that Drosophila melanogaster possesses multiple insulin regulators that have different nutrient sensitivities. Multi-input Dilp regulation might be advantageous under the imbalanced nutritional conditions that arise in the wild, and this could represent a general strategy for animal growth regulation.
In mammals, different hormones are secreted in response to long-term or short-term metabolic changes. For instance, gut-derived cholecystokinin, glucagon-like peptide-1, and PYY3-36, as well as stomach-derived ghrelin, all of which control feeding behavior, are secreted in response to food ingestion [32]. These hormones respond to acute metabolic changes and immediately signal to the feeding center in the brain. On the other hand, the synthesis or secretion of leptin and adiponectin is affected by the amount of lipid stored in adipocytes [33] [34], suggesting that leptin and adiponectin respond to long-term changes in metabolic status. The expression of CCHa2 responds to yeast and glucose within 6 hours (Fig 1H), indicating that CCHa2 mediates relatively rapid changes in metabolic status. Thus, it appears that CCHa2 functions as a short-acting metabolic regulator analogous to the mammalian gut- or stomach-derived hormones described above, and that Drosophila melanogaster CCHa2 might have an important role in the maintenance of energy homeostasis under volatile nutritional conditions.
The results from the calcium imaging experiments using brain explants and IPC-specific CCHa2-R knockdown strongly suggest that CCHa2 crosses the blood-brain barrier (BBB) to regulate the IPCs, although the underlying mechanism remains elusive. The Drosophila BBB consists of two different glial cell layers composed of either the perineurial glia (PG) or the subperineurial glia (SPG) [35] [36]. The SPG cell layer, which is adjacent to the neurons of the brain, forms septate junctions, which function as a barrier to separate the humoral space and the brain, analogously to the mammalian tight junctions formed between endothelial cells. Although several studies have identified important molecules involved in the formation of these septate junctions[36,37] [38] [39], little is known about functional aspects of the BBB [40]. CCHa2 could provide an ideal model for the study of BBB function as well as drug delivery across the BBB.
These experiments also show that peripheral tissue-derived CCHa2 directly activates IPCs in the brain. In mammals, direct sensing of blood glucose levels by pancreatic β-cells is a major trigger for insulin secretion. In these cells, glucose metabolism inhibits the ATP-dependent potassium channel (KATP channel) and opens voltage-dependent calcium channels (VDCCs), resulting in the exocytosis of insulin-containing granules [41]. The KATP channel also seems to be involved in insulin secretion in Drosophila IPCs [42]. Interestingly, a group of Gαs- and Gαq/11-coupled GPCRs can also activate the insulin secretion pathway in mammals [43]. The closest mammalian homologues of CCHa2-R—the Bombesin-related receptor subtypes 3, 1, and 2 (also known as gastrin-releasing-peptide receptor)—signal through Gαq/11 [44] [45] [46] [47] [48]. The slow rise in [Ca2+] in the IPCs in response to CCHa2 application is consistent with CCHa2-R’s mediation of Dilp release through the same pathway.
In contrast to Dilp2, dilp5 is also regulated by CCHa2/CCHa2-R signaling at the transcriptional level. Although the expression of dilp5 in the IPCs is activated by the conserved transcription factors Dachshund and Eyeless [27], whether CCHa2-R regulates these factors in IPCs remains unknown.
Overexpression of CCHa2-R in IPCs using the GAL4/UAS system displayed inhibitory effects on dilp5 expression, which prevented us from investigating whether direct CCHa2-R activation in IPCs is sufficient for Dilp regulation. CCHa2-R expression in the brain is not specific to IPCs but occurs in other central neurons (Fig 2B, 2C, 2F and 2G). Therefore, although we have shown that CCHa2-R expression in the IPCs is required for full dilp5 expression, it is possible that there may also be additional indirect pathways by which CCHa2 may up-regulate the Dilps. Although BBB glial cells are proposed to receive as-yet-unidentified signal(s) from the fat body and re-activate neural stem cells in the brain by secreting Dilp6 [49] [50], CCHa2-R>nlsGFP was undetectable in the BBB glial cells (S1G–S1I Fig). Thus BBB cells are unlikely to receive CCHa2 signals or to relay the signals to the IPCs.
The closest mammalian homologue of CCHa2-R is Brs3, an orphan GPCR, which is a member of the bombesin-like peptide receptor family [51]. Brs3-deficient mice develop obesity in association with a reduced metabolic rate and elevated feeding activity [52]. Interestingly, Brs3 is expressed in pancreatic β-cells both in mice and humans [53]. However, its involvement in insulin regulation has been controversial. Only if Brs3 knockout adult mice become obese (especially after 23 weeks old) do their plasma insulin levels increase [52]. Since hyper-insulinemia is generally observed in genetically obese mice, the elevation of insulin is most likely the consequence of the obesity rather than the loss of Brs3 function [52]. On the other hand, a Brs3 agonist promoted insulin secretion in both rodent insulinoma cell lines and in islets isolated from wild-type but not Brs3 mutants [53]. Our vigorous genetic approach combined with direct observations of Dilp production in IPCs provides the first evidence, to our knowledge, that Bombesin-related receptor signaling activated by its endogenous ligand promotes insulin production.
The following fly stocks were used: Oregon-R, y w, dilp2-GAL4 [12], Lsp2-GAL4 [13] [54], ppl-GAL4 [8] [30], cg-GAL4 [30] [55], UAS-TSC1/2 [56], UAS-CCHa2, UAS-dilp2 [11], UAS-CCHa2 RNAi (VDRC-102257), UAS-GCaMP6s [31], UAS-CCHa2-R-shRNA #3, #24, #30 (see below), UAS-dicer1 (see below), and dilp63932 [13]. CCHa2-RKO51-2 and CCHa2-RTAL-34 are putative null alleles, which were generated by gene targeting using homologous recombination and TALEN, respectively (see below). CCHa2CR-1, CCHa2CR-2, and CCHa2CR-3 were generated using the CRISPR/Cas9 system (see below). Df(3R)Exel7320 (Exelixis) was used as a deficiency that uncovers CCHa2.
Flies were raised at 25°C on regular fly food containing (per liter) 46 g yeast extract, 70 g cornmeal, 100 g glucose, and 6 g agar. For starvation experiments, third-instar larvae (72 hours AEL) were cultured on water agar plates for 18 hours. The starved larvae were later re-fed with different nutrients (yeast paste, 46 g/L peptone, 10% glucose, 0.3% L-isoleucine, or 0.2% L-leucine).
CCHa2-RKO51-2 was generated by gene targeting using homologous recombination [22] [23]. A 2.4-kb fragment upstream of the second exon and a 2.5-kb fragment downstream of the 6th exon were amplified by PCR and then cloned into the NheI and SpeI sites of the pGX-attP-WN vector [23]. The targeting vector was integrated into the fly genome to generate a donor line. We performed gene targeting as described in Zhou et al. (2012) [23] and obtained the CCHa2-RKO51-2 mutation in which the region from the first methionine through the middle of the 7th transmembrane domain was removed (Figs 3 and S2). CCHa2-RTAL-34 was generated by inducing double-strand breaks at the CCHa2 locus using TALEN [24]. A Golden Gate TALEN kit (Addgene) was used to generate two sets of RVD plasmids corresponding to the sequences found in the first coding exon of the CCHa2-R gene (1L, 1R, 2L, and 2R; S2 Fig).
1L: TAGTACCGTATGTGCCC
1R: GGAGACGTACATTGTCA
2L: TGCTGTACACGCTCATCTTC
2R: GGCAACGGCACGCTGGTCATCA
For in vitro transcription, the RVD fragments were cloned into the pCS2TAL3DDD or pCS2TAL3RRR vector (a gift from K. Hoshijima, The University of Utah). Capped and polyadenylated RNAs were synthesized by in vitro transcription from linearized 1L, 1R, 2L, and 2R plasmids, and they were mixed for injection into y w embryos. To detect mutated DNA, the target region was amplified by PCR from genomic DNA of mutant candidates, and the amplified fragments were re-annealed. The resulting fragments were digested by T7 endonuclease I (NEB). Sequencing of the mutated fragment revealed that the CCHa2-RTAL-34 mutation is a 74-bp deletion causing a frame-shift mutation at amino acid position 62 (Figs 3 and S2).
CCHa2 mutants were generated using the germline-specific CRISPR/Cas9 system as described in Kondo and Ueda (2013) [29] (S5 Fig). The following sgRNA target was used for the mutagenesis of the CCHa2 gene. Break points of the mutants were determined as described above.
CCHa2 (88): GCCTACGGTCATGTGTGCTACGG
The CCHa2-R-GAL4::p65 construct (Fig 2E) was generated with bacterial artificial chromosome (BAC) recombineering techniques [57] in P[acman] BAC clone CH321-87C13 [58] (Children’s Hospital Oakland Research Institute, Oakland, CA). A landing-site cassette was created by flanking the selectable marker in pSK+-rpsL-kana [59] with 5' GAL4 and 3' HSP70 UTR arms from pBPGUw [60], a gift of G. Rubin. Fifty-base CCHa2-R-specific genomic homology arms were added by PCR using the primers below (genomic homology in lower case, cassette homology in upper case):
CCHa2-R-GAL4-F: 5'-tagaaacaccattgagacatcttgcccaggagcagctccctcctccccacATGAAGCTACTGTCTTCTATCGAACAAGC
CCHa2-R-HSP70-R: 5'-acttccccaccttctgcgggacccccacagtgcgtgatatatccacttacGATCTAAACGAGTTTTTAAGCAAACTCACTCCC
The cassette was recombined into the BAC and then replaced with full-length GAL4::p65-HSP70 amplified from pBPGAL4.2::p65Uw [61] (a gift of G. Rubin) in a second recombination. The final BAC was verified by sequencing the recombined regions and was integrated into the attP40 site [62] (Genetic Services, Inc., Cambridge, MA).
UAS-CCHa2 was generated by cloning the coding region of the CCHa2 gene into the EcoRI site of the pUAST vector. Flies were transformed with the UAS-CCHa2 vector by P-element-mediated transformation to generate the UAS-CCHa2 stock.
UAS-CCHa2-R-shRNA lines were generated as described previously [63]. The following oligonucleotide pairs were annealed and cloned into the NheI and EcoRI sites of the pWALIUM20 vector. The UAS-shRNA constructs were integrated into the attP40 site [62].
#3 Top strand: 5'-CTAGCAGTGGCTGATCTGTTGGTTATATTTAGTTATATTCAAGCATAAATATAACCAACAGATCAGCCGCG
#3 Bottom strand: 5'-AATTCGCGGCTGATCTGTTGGTTATATTTATGCTTGAATATAACTAAATATAACCAACAGATCAGCCACTG
#24 Top strand: 5'-CTAGCAGTCGATTGTCTACACGCAGGAAATAGTTATATTCAAGCATATTTCCTGCGTGTAGACAATCGGCG
#24 Bottom strand: 5'-AATTCGCCGATTGTCTACACGCAGGAAATATGCTTGAATATAACTATTTCCTGCGTGTAGACAATCGACTG
#30 Top strand: 5'-CTAGCAGTCGAACTGACTTGGAGTTATGTTAGTTATATTCAAGCATAACATAACTCCAAGTCAGTTCGGCG
#30 Bottom strand: 5'-AATTCGCCGAACTGACTTGGAGTTATGTTATGCTTGAATATAACTAACATAACTCCAAGTCAGTTCGACTG
In order to amplify the efficiency of shRNA-mediated gene knockdown, UAS-dicer1 was constructed. The dicer1 fragment was amplified by PCR using the dicer1 cDNA as a template (a gift from Q. Liu, University of Texas Southwestern Medical Center, and Y. Tomari, University of Tokyo). The following primers were used with Q5 DNA polymerase (NEB):
Dcr-1 5': 5'-GGGGTACCAAAATGGCGTTCCACTGGTGCG-3'
Dcr-1 3': 5'-GGAGATCTTAGTCTTTTTTGGCTATCAAGC-3'
The dicer1 fragment was cloned into the KpnI and BglII sites of the UASp-K10attB vector (a gift from B. Suter, University of Bern), and the UAS-dicer1 construct was integrated into the attP2 site.
Total RNA from whole larvae or larval tissues was extracted using a PureLink RNA Mini Kit (Life Technologies). cDNAs were prepared by reverse-transcribing 1 μg of total RNA using ReverTra Ace qPCR RT Master Mix with gDNA Remover (Toyobo), and quantitative RT-PCR was performed using Thunderbird qPCR Mix (Toyobo). Expression levels were normalized against those of rp49. The following primers were used:
CCHa2 forward: 5′-AGTGCAGTTGGACTTTGGTAGTGT
CCHa2 reverse: 5′-AGGGATGCTGTTTAGCATCTATGAC
CCHa2R forward: 5′-CTCACTGTCTTTACTGCGGTGAT
CCHa2R reverse: 5′-CCACCATGAACTTTGCATACTC
dilp2 forward: 5′-GTATGGTGTGCGAGGAGTAT
dilp2 reverse: 5′-TGAGTACACCCCCAAGATAG
dilp3 forward: 5′-GTCCAGGCCACCATGAAGTTGTGC
dilp3 reverse: 5′-CTTTCCAGCAGGGAACGGTCTTCG
dilp5 forward: 5′-TGTTCGCCAAACGAGGCACCTTGG
dilp5 reverse: 5′-CACGATTTGCGGCAACAGGAGTCG
dilp6 forward: 5′-TGCTAGTCCTGGCCACCTTGTTCG
dilp6 reverse: 5′-GGAAATACATCGCCAAGGGCCACC
rp49 forward: 5′-AGTATCTGATGCCCAACATCG
rp49 reverse: 5′-CAATCTCCTTGCGCTTCTTG
Antibody staining was conducted as described previously [64], except that the gut samples were fixed in 4% formaldehyde in PBS for 4 hours. Rabbit anti-CCHa2 (1:1,000) [19], rabbit anti-Dilp2 (1:2,500; T. Ida), rabbit anti-Dilp2 (1:2,000, a gift from T. Nishimura, RIKEN Center for Developmental Biology) [27], rabbit anti-Dilp5 (1:2000, a gift from T. Nishimura) [27], mouse anti-Repo c8D12 (1:100; DSHB), mouse anti-GFP (1:500; Invitrogen), and rabbit anti-NPF (1:500; Ray Biotech, Inc) were used as primary antibodies, with anti-mouse-Alexa 488 (1:500; Invitrogen), anti-mouse-Cy3 (1:500; Jackson ImmunoResearch), anti-rabbit-Alexa 488 (1:500; Invitrogen), and anti-rabbit-Cy3 (1:500; Jackson ImmunoResearch) used as secondary antibodies. Alexa 488-conjugated phalloidin (Invitrogen) and TO-PRO-3 (Invitrogen) were used to label the cell membrane and nuclei, respectively. FISH was performed as described in Lehmann and Tautz [65] with a modification designed to amplify signals. Digoxigenin-labeled RNA probes were detected using the TSA Plus Fluorescein Kit (PerkinElmer). For the detection of CCHa2-R mRNA, the signals were further amplified using mouse anti-FITC (1:1,000; Jackson ImmunoResearch) and anti-mouse-Alexa 488 antibodies. Images were analyzed by LSM700 confocal microscopy (Zeiss) or DM5000B fluorescence microscopy with a DFC500 CCD camera (Leica).
To quantify Dilp2 and Dilp5 levels in IPCs, CNS samples were doubly stained with anti-Dilp2 or anti-Dilp5 and anti-GFP antibodies. Fluorescence images were acquired using a LSM700 confocal microscope (Zeiss). Constant laser power and scan settings were used to image the control and mutant samples. To quantify Dilp levels, Dilp and GFP signal intensities were measured on the same section using the ImageJ software (NIH). Dilp signals were normalized against the GFP signal intensities.
dilp2-GAL4/UAS-GCaMP6s and dilp2-GAL4, CCHa2-RTAL-34/UAS-GCaMP6s, CCHa2-RTAL-34 animals were used for imaging. Late-third-instar larvae were dissected in PBS. The ring gland and imaginal disks were removed from the brain. Dissected brains were immersed in 200 μl of PBS and tethered with tungsten wire (0.125 mm diameter, A-M Systems, Inc.) to the bottom of a culture dish (35x10 mm, Falcon). Custom-synthesized CCHa2 (SCRUM Inc.) and synthetic ghrelin and nociception (Peptide institute Inc.) were used. The ligand was pipetted directly into the bath in a volume of 100 μl to yield a final concentration of 10–9 M. Imaging was performed with a microscope (Axio Imager.A2, Carl Zeiss) equipped with a spinning disc confocal head (CSU-W1, Yokogawa). We used a 20x water-immersion objective lens (NA = 0.5; Carl Zeiss) mounted with a piezoelectric-activated lens mover (P-725K085 PIFOC, Physik Instrumente GmbH & Co. KG). GCaMP6s signals were excited with a 488-nm laser at 512 x 512 pixel resolution and collected at 250 ms/frame using an EM-CCD camera (ImagEM512, Hamamatsu Photonics) in water-cooled mode. For each IPC within an optical section, regions of interest (ROIs) were selected over multiple IPC somata using the ImageJ software (National Institutes of Health). Raw intensity values for GCaMP6s emission were recorded as mean pixel intensities (value range: 0–65,535) for each ROI at each time point and exported from ImageJ. GCaMP6s signal averaged over 30 frames before stimulation was taken as F0, and ΔF/F0 was calculated for each time point.
Larvae were synchronized at hatching (24 hours AEL) and cultured with regular fly food. Ten to thirty-five larvae were collected and weighed at each time point. For the analysis of developmental timing, pupation of synchronized larvae was examined every three hours. Results are shown as the averages of triplicated experiments.
The food-ingestion assay was modified from Edgecomb et al. [66]. In brief, 10 larvae were fed with yeast paste containing 1% Brilliant Blue FCF (Wako) for 1 hour or 2 hours. After rinsing with water, larvae were quickly frozen and homogenized in 200 μL of PBS (pH 7.0). The homogenates were centrifuged for 16,000 x g for 10 min, and the supernatants were analyzed spectrophotometrically for absorbance at 625 nm. Results are shown as the averages of triplicated experiments.
The data in each graph were presented as means ± SEM. Two-tailed t-test was used to evaluate the significance of the results between two samples. For multiple comparisons, one-way ANOVA was applied, then pair-wise comparison was performed by Tukey-Kramer (Figs 1H, 4D and 5E and 5F) or Dunnett test (Fig 1I). For the calcium imaging data, Kruskal-Wallis rank sum test was performed for comparison followed by Holm-Bonferroni post-hoc test between four groups of samples (S6 Fig). A significance level of p<0.05 was used for all tests, which was marked by an asterisk in the figure.
|
10.1371/journal.ppat.1003220 | Increased CD8+ T Cell Response to Epstein-Barr Virus Lytic Antigens in the Active Phase of Multiple Sclerosis | It has long been known that multiple sclerosis (MS) is associated with an increased Epstein-Barr virus (EBV) seroprevalence and high immune reactivity to EBV and that infectious mononucleosis increases MS risk. This evidence led to postulate that EBV infection plays a role in MS etiopathogenesis, although the mechanisms are debated. This study was designed to assess the prevalence and magnitude of CD8+ T-cell responses to EBV latent (EBNA-3A, LMP-2A) and lytic (BZLF-1, BMLF-1) antigens in relapsing-remitting MS patients (n = 113) and healthy donors (HD) (n = 43) and to investigate whether the EBV-specific CD8+ T cell response correlates with disease activity, as defined by clinical evaluation and gadolinium-enhanced magnetic resonance imaging. Using HLA class I pentamers, lytic antigen-specific CD8+ T cell responses were detected in fewer untreated inactive MS patients than in active MS patients and HD while the frequency of CD8+ T cells specific for EBV lytic and latent antigens was higher in active and inactive MS patients, respectively. In contrast, the CD8+ T cell response to cytomegalovirus did not differ between HD and MS patients, irrespective of the disease phase. Marked differences in the prevalence of EBV-specific CD8+ T cell responses were observed in patients treated with interferon-β and natalizumab, two licensed drugs for relapsing-remitting MS. Longitudinal studies revealed expansion of CD8+ T cells specific for EBV lytic antigens during active disease in untreated MS patients but not in relapse-free, natalizumab-treated patients. Analysis of post-mortem MS brain samples showed expression of the EBV lytic protein BZLF-1 and interactions between cytotoxic CD8+ T cells and EBV lytically infected plasma cells in inflammatory white matter lesions and meninges. We therefore propose that inability to control EBV infection during inactive MS could set the stage for intracerebral viral reactivation and disease relapse.
| There is general consensus that multiple sclerosis (MS) is associated with Epstein-Barr virus (EBV) infection but the mechanistic links are still debated. EBV is a B-lymphotropic herpesvirus widespread in the human population and normally contained as a persistent, asymptomatic infection by immune surveillance. However, EBV can cause infectious mononucleosis, is associated with numerous human malignancies, and is implicated in some common autoimmune diseases. While EBV infection alone cannot explain MS development, it has been postulated that in susceptible individuals alterations in the mechanisms regulating the immune response to the virus may contribute to MS pathogenesis. Here, we show that MS patients with inactive disease exhibit a lower CD8+ T-cell response to EBV when compared to healthy donors and active MS patients while the latter have a higher frequency of CD8+ T cells specific for EBV lytic antigens. Therapy with interferon-β and natalizumab, two treatments for relapsing-remitting MS, was associated with marked changes in the EBV specific CD8+ T cell response. We also demonstrate that one of the EBV lytic antigens recognized by CD8+ T cells expanding in the blood during active MS is expressed in the inflamed MS brain. Our results support a model of MS pathogenesis in which EBV infection and reactivation in the CNS stimulates an immunopathological response and suggest that antiviral or immunomodulatory therapies aimed at restoring the host-EBV balance could be beneficial to MS patients.
| Multiple sclerosis (MS) is the most common chronic inflammatory disease of the central nervous system (CNS) causing demyelination, neurodegeneration and disability. In most cases, MS is characterized by a relapsing-remitting course at onset which eventually develops into a progressive form; more rarely MS manifests as a primary progressive disease [1]. Immunomodulating and immunosuppressive drugs can reduce but not halt the disease process. Both the etiology and pathogenic mechanisms of MS are poorly understood. Genetic and environmental factors have been implicated in MS development, but the identity of the antigens (self or non-self) promoting chronic CNS inflammation remains elusive [2]. Several viruses have been linked to MS; however, Esptein-Barr virus (EBV) shows the strongest association with the disease [3]–[5]. EBV is a B-lymphotropic DNA herpesvirus that infects 95–98% of individuals worldwide, establishes a life-long, generally asymptomatic infection in B cells, and is the cause of infectious mononucleosis and of several lymphatic and non-lymphatic malignancies [6]. EBV has also been implicated in common autoimmune diseases, like systemic lupus erythematosus and rheumatoid arthritis [7], [8]. Numerous studies have consistently demonstrated a higher prevalence of EBV infection and higher titers of antibodies to EBV antigens, in particular to EBV nuclear antigen-1 (EBNA-1), in young and adult MS patients compared to age-matched, healthy individuals [9]–[14]. It has also been shown that high titers of anti-EBNA-1 antibodies prior to MS onset [15] or at the time of a clinically isolated syndrome [16] and a previous history of infectious mononucleosis [17] increase the risk of developing MS. Furthermore, MS patients have higher frequencies of CD4+ T cells specific for EBNA-1 relatively to healthy, seropositive individuals [18], while EBV-specific CD8+ T-cell responses in MS have been reported to be increased or decreased in different studies [19]–[23].
Although enhanced immune reactivity to EBV in MS suggests perturbed EBV infection, it is debated whether and how this can induce or influence the disease. EBV infection could contribute to MS through multiple mechanisms, including molecular mimicry, immortalization of autoantibody-producing B cell clones, and immunopathology [3], [24]. It has been shown that CD4+ T cells from some MS patients cross-react with EBV and myelin antigens but the relevance of this finding to disease pathogenesis is still unclear [25], [26]. EBV DNA load in the peripheral blood does not differ significantly between MS patients and healthy donors (HD) [16], [18], [23] and the possibility that a persistent EBV infection in the CNS drives an immunopathological response that damages myelin and neural cells is reasonable but remains controversial [27]. While several studies report absence of EBV in MS brain lesions [28]–[31], we [32]–[34] and others [35] have shown that an abnormally high proportion of B cells infiltrating the MS brain are latently infected with EBV. We have also shown that ectopic B-cell follicles present in the inflamed meninges of patients with secondary progressive MS harbour EBV infected B cells and that EBV can reactivate in plasma cells in immunologically active white matter lesions and meningeal B-cell follicles [32]–[34].
If MS were the result of an immunopathological response aimed at eliminating a persistent EBV infection in the CNS, a positive correlation should be found between disease activity assessed by magnetic resonance imaging (MRI) or clinical progression and immune response to EBV. In support of this hypothesis, it has been shown that serum levels of EBNA-1 IgG positively correlate with gadolinium-enhancing MRI lesions (characteristic of acute inflammation), lesion size and Expanded Disability Status Scale (EDSS) in patients with MS [36] and in patients with a clinically isolated syndrome who develop definite MS [16]. Another study reported higher disease activity on MRI in a subgroup of relapsing-remitting MS patients with stable levels of IgG specific for EBV early antigens expressed during the lytic cycle [37]. It has also been shown that CD8+ T-cell responses toward pooled EBV latent and lytic antigens in the blood of MS patients are high early in MS course and decrease during disease progression suggesting a possible association with more frequent episodes of CNS inflammation in early disease phases [21]. However, it is not known whether changes in the CD8+ T cell response to individual EBV latent and/or lytic antigens are associated with active and inactive MS phases. To address this issue, we have used pentamer staining to characterize the CD8+ T-cell response to EBV in the peripheral blood of patients with relapsing-remitting MS, both untreated and treated, and HLA-matched controls. Positivity to pentamer staining was then correlated with disease activity and inactivity, as assessed by clinical criteria and MRI of the brain. Our study reveals a lower prevalence of the CD8+ T cell response to EBV in inactive MS patients, a higher frequency of CD8+ T cells specific for EBV lytic and latent antigens during active and inactive disease, respectively, and marked changes in the EBV-specific CD8+ T cell response during treatment with approved disease-modifying drugs, such as interferon-β (IFN-β) and natalizumab. By analyzing post-mortem MS brain tissue, we demonstrate that the same EBV lytic antigen eliciting a higher CD8+ T cell response in the peripheral blood during active MS is expressed in inflammatory white matter lesions and meninges. We also show interactions between CNS-infiltrating cytotoxic CD8+ T cells and EBV lytically infected plasma cells, further supporting the link between EBV reactivation, higher cytotoxic immune responses to EBV lytic antigens and MS exacerbations.
The EBV-specific CD8+ T-cell response was studied with the pentamer technology in 43 HD and 113 patients (79 untreated and 34 treated) with relapsing-remitting MS who were selected according to their HLA genotype (HLA-A*0201 and/or HLA-B*0801) (Table 1). The choice of pentamers was guided by the high frequency of HLA-A2 family alleles in Caucasians, by previous characterization of the immunodominant peptide epitopes from EBV latent and lytic proteins that are most frequently recognized by CD8+ T cells [38]-[40], and by the possibility to evaluate the CD8+ T cell response to at least one viral latent and one viral lytic protein in the same subject. We therefore studied CD8+ T cell reactivities to the immunodominant peptides of LMP-2A (EBV latent antigen) and BMLF-1 (EBV early lytic antigen) restricted through the HLA-A*0201 allele, and of EBNA-3A (EBV latent antigen) and BZLF-1 (EBV immediate early lytic antigen) restricted through the HLA-B*0801 allele. As control for anti-viral MHC class I restricted CD8+ T cell responses we selected a HLA-A*0201 pentamer coupled with a peptide from cytomegalovirus (CMV) pp65 protein.
Freshly isolated PBMC from HD (n = 43) and untreated MS patients (n = 79) were stained with the above mentioned fluorochrome-labeled HLA-A*0201 and/or HLA-B*0801/viral peptide epitope pentamers. We first examined the prevalence of EBV-specific CD8+ T cell responses in our cohort, namely the proportion of individuals with detectable pentamer+ CD8+ T cells (Fig. S1). Positive pentamer staining specific for at least one EBV epitope was found in a similar proportion of HD (39%) and untreated MS patients (33%). The prevalence of CD8+ T cell responses to EBV latent and lytic antigens was similar in HD and untreated MS patients (Figure 1 A). In HLA-A2+ subjects where both EBV- and CMV-specific CD8+ T cell responses could be evaluated, no differences in the response to either virus were found between HD (n = 34) and untreated MS patients (n = 45) (Figure S2A).
We then explored whether the prevalence of EBV-specific CD8+ T-cell responses could be related to disease activity. Untreated MS patients were subdivided in two groups consisting of 31 active and 48 inactive patients, as defined by presence and absence of clinical relapses and acute brain inflammation assessed with gadolinium-enhanced MRI, respectively (Table 2). There was a similar prevalence of latent and lytic antigen-specific CD8+ T cell responses in HD and active MS patients but a significantly lower prevalence of lytic antigen-specific CD8+ T cell responses in inactive MS patients than HD (19% versus 39%, p = 0.05). Inactive MS patients also tended to have a lower prevalence of latent and lytic antigen-specific CD8+ T cell responses when compared with active MS patients (17% versus 35%, p = 0.1 and 19% versus 39%, p = 0.09, respectively) (Figure 1A). In contrast, no differences were found among HD, inactive MS and active MS patients in the CD8+ T cell response to CMV (Figure S2B). These findings therefore indicated a weaker immune response against EBV, particularly against lytic cycle antigens, in inactive MS patients compared to HD and active MS patients. Of note, 39% of active MS patients, but only 15% of inactive MS patients with a detectable EBV-specific CD8+ T cell response had a disease duration longer than 8 years, suggesting a decay with time of the EBV-specific immune response associated with inactive disease.
We then evaluated the percentage of pentamer+ CD8+ T cells within the circulating CD3+ CD8+ T cell population in the study subjects with a detectable EBV-specific CD8+ T cell response. Similarly to prevalence, the frequency of latent and lytic antigen-specific CD8+ T cells did not differ significantly between HD (n = 17) and total untreated MS patients (n = 26) (Figure 2 A). Also the frequency of CD8+ T cells specific for EBV and CMV antigens was similar in HLA-A2+ HD and total MS patients (Figure S3A). Again, differences in the immune response to EBV emerged only when patients were stratified according to disease activity (Figure 2 B–D). The frequency of latent antigen-specific CD8+ T cells in inactive MS patients (1.8±2.6%, mean ± SD) tended to be higher than in HD (0.3±0.2%, mean ± SD; p = 0.07) and was significantly higher than in active MS patients (0.22±0.19%, mean ± SD; p = 0.05) (cumulative data and representative plots are shown in Figure 2 B and D, respectively). This difference was mainly due to the fact that 6 out of 8 inactive MS patients recognized EBNA-3A and 3 of these displayed a very strong immune response against this EBV latent antigen (2.7, 2.8 and 7.6% of total circulating CD8+ T cells were EBNA-3A-specific). In contrast, the frequency of lytic antigen-specific CD8+ T cells in active MS patients (1.8±2.8%, mean ± SD) was significantly higher than in HD (0.34±0.28%, mean ± SD; p = 0.03) and tended to be higher than in inactive MS patients (0.3±0.2%, mean ± SD, p = 0.1). The latter difference did not reach statistical significance probably due to the low number of inactive MS patients analyzed (n = 9). These findings therefore indicated more frequent recognition of EBV latent and lytic antigens during inactive and active MS phases, respectively. In contrast, the frequencies of CMV-specific CD8+ T cells did not differ significantly among HLA-A2+ HD, inactive MS and active MS patients (Figure S3B, C). No correlation was found between the frequency of EBV-specific CD8+ T cells and disease duration in inactive and active MS patients (Figure S4).
We then analyzed the EBV-specific CD8+ T cell response in MS patients who were treated with IFN-β (n = 20) and natalizumab (n = 14) (Table 1). IFN-β, the most frequently used first-line treatment for relapsing-remitting MS, has antiviral and immunoregulatory activities and reduces relapse frequency and brain MRI activity in relapsing-remitting MS patients [41]. The monoclonal antibody natalizumab inhibits lymphocyte extravasation into the CNS and is highly effective in suppressing clinical relapses and disease activity in relapsing-remitting MS patients who fail to respond to first-line therapies [42].
Among 20 MS patients treated with IFN-β for 1 to 11 years (median = 4 years), 7 and 13 were in the active and inactive phase of disease, respectively (Table 2). The prevalence of the CD8+ T cell response to EBV latent and lytic antigens was similar in total untreated and IFN-β-treated MS patients (Figure 1 A, B). However, none of the 13 IFN-β-treated inactive patients had a detectable CD8+ T cell response to EBV (Figure 1 B) and CD8+ T cells for CMV were found only in 1 of 9 HLA-A02+ IFN-β-treated inactive patients (data not shown), indicating that effective IFN-β therapy is associated with a general inhibition of the antiviral response. In contrast, 57% (4/7) and 66% (4/6) of the IFN-β-treated MS patients with active disease had a detectable CD8+ T cell response to EBV (Figure 1 B) and CMV (data not shown), respectively. Moreover, the frequency of CD8+ T cells specific for EBV latent, but not lytic, antigens was significantly higher in IFN-β-treated (0.8±0.6%, mean ± SD) than in untreated (0.2±0.2%, mean ± SD ; p = 0.01) active MS patients (Figure 2 B, C).
After 8 to 16 months (median = 12 months) of treatment with natalizumab, all 14 MS patients analyzed were in the inactive phase of disease, both clinically and by MRI (Table 2). Unexpectedly, nearly all patients in this group had a detectable CD8+ T cell response to EBV latent and lytic antigens (87% and 93%, respectively) (Figure 1 B). This prevalence was significantly higher than that found in HD and any other group of MS patients (p<0.001). Nevertheless, the frequencies of latent and lytic antigen-specific CD8+ T cells in natalizumab-treated MS patients were similar to those found in untreated inactive MS patients (Figure 2 B, C). As expected for patients in the inactive phase of disease, the frequency of lytic antigen-specific CD8+ T cells tended to be lower in natalizumab-treated patients than in untreated active MS patients (p = 0.09) (Figure 2 C). The frequency of CMV-specific CD8+ T cells did not differ among HD, untreated and natalizumab-treated MS patients (data not shown).
We then asked whether changes in the EBV-specific CD8+ T cell response during active and inactive MS phases could be detected in longitudinal studies. Despite experiencing clinical relapses, two patients (HLA-B08+ B2/B2-2 and HLA-A02+ A14) in our cohort refused immunomodulatory therapy and agreed to be monitored periodically for 27 and 7 months, respectively. Patient B2/B2-2, who was clinically silent and was diagnosed MS one year before our analysis started, displayed a very highy frequency of EBNA-3A-specific CD8+ T cells (6% of circulating CD8+ T cells) at the beginning of the observation period (month 0; Figure 3 A). EBNA-3A-specific CD8+ T cells progressively decreased during the subsequent months and became undetectable between month 17 and 27. In parallel, the frequency of BZLF-1 specific CD8+ T cells, which were undetectable at previous time points, abruptly increased and reached a peak (11% of total circulating CD8+ T cells) at month 21 in concomitance with the presence of active MRI lesions. The percentage of BZLF-1-specific CD8+ T cells in the CD8+ T cell population then declined to 2.5% in the subsequent 6 months, denoting marked expansion and subsequent contraction of the immune response toward this EBV lytic antigen (Figure 3 A).
In patient A14, who experienced frequent clinical relapses, the frequency of CD8+ T cells specific for the EBV lytic antigen BMLF-1 ranged between 0.56 and 0.71% during a clinical relapse (months 0 and 3 of the observation period) and in the presence of an active MRI scan, and dropped to 0.11% in the subsequent 4 months (Figure 3 A). During the same period, the frequency of CD8+ T cells specific for the EBV latent antigen LMP-2A and for CMV pp65 antigen remained low and stable (Figure 3 A).
Longitudinal analysis of EBV-specific CD8+ T-cell responses was also performed in 2 HLA-B08+ MS patients (BTY5 and BTY8) treated with natalizumab and monitored periodically for 15 and 12 months, respectively, starting at 12–14 months after therapy initiation. Both natalizumab-treated patients were in the inactive phase of disease (according to clinical and MRI evaluation) and had a detectable CD8+ T cell response to EBNA-3A and BZLF-1 (Figure 3 B). However, while the frequency of BZLF-1-specific CD8+ T cells was stable during the whole observation period a steady rise in the CD8+ T cell response to EBNA-3A was observed in both patients after 15–18 months of therapy (Figure 3B). Three HD followed for 8 to 19 months did not show any significant variation in the frequency of CD8+ T cells specific for EBV latent and lytic antigens (Figure 3 C).
The increased frequency of CD8+ T cells specific for two EBV lytic antigens (BZLF-1, BMLF-1) in the blood of MS patients with an active MRI scan indirectly suggests a response to a previous or concomitant viral reactivation in the brain. In search for a link between immunological findings and brain inflammation, we examined the expression of BZLF-1 mRNA and protein in post-mortem brain tissue from 7 patients who died in the secondary progressive phase of MS and were characterized by a severe clinical course and substantial brain inflammation. The MS brain samples selected for this study contained immunologically active (both active and chronic active) lesions in the white matter and highly inflamed meninges with B-cell follicle-like structures that we previously showed to be major EBV reservoirs [32]–[34].
In preliminary experiments aimed at evaluating the specificity and binding of anti-BZLF-1 monoclonal antibody, we observed BZLF-1 immunoreactivity in the nucleus of EBV transformed B95-8 cells and in an EBV+ tonsil from a patient with infectious mononucleosis (Figure S5). Conversely, no BZLF-1+ cells were detected in sections of a non pathological human brain, of a brain from a patient with tuberculous meningoencephalitis and of an EBV-negative lymphoma (Figure S5). BZLF-1 immunoreactivity was detected in brain samples of all MS cases analyzed using immunohistochemical techniques (n = 5). Both isolated and small groups of BZLF-1+ cells were present in the inflamed meninges, at the periphery of B-cell follicle-like structures (Figure 4 A–H) and in diffuse inflammatory cell infiltrates (Figure 4 I–K). BZLF-1+ cells were also present in the perivascular cuffs of inflamed blood vessels in active white matter lesions characterized by a high density of intraparenchymal foamy macrophages (Figure 5), but not in demyelinated, chronic active and inactive white matter lesions (data not shown), thus linking EBV reactivation to acute inflammation. Nearly all BZLF-1+ cells infiltrating the meninges (Figure 4 C, D, K) and active WM lesions (Figure 5 D, E, G, H) were identified as Ig-producing plasmablasts/plasma cells, which is consistent with the knowledge that EBV reactivates upon B-cell differentiation into plasma cells [43]. At these sites the proportion of Ig+ cells expressing BZLF-1 ranged between 1 and 10%. BZLF-1+ plasma cells were detected in the same infiltrated brain areas where plasma cells expressing BFRF1, an EBV early lytic protein induced by BZLF-1 [44], were also found (Figure 4 E, Figure 5 F).
Expression of BZLF-1 was also investigated using quantitative real-time RT-PCR in 4 inflamed MS brain samples, 2 of which had been analyzed by immunohistochemistry. No BZLF-1 RNA was detected in whole MS brain sections (data not shown). This negative result was expected given the paucity of EBV lytically infected plasma cells relatively to the large and heterogeneous cell population of the MS brain. To enrich for EBV infected cells and increase the sensitivity of the technique, perivascular and meningeal inflammatory cell infiltrates and the surrounding, non infiltrated brain parenchymal regions were harvested from MS brain sections using laser capture microdissection and analyzed using pre-amplification, quantitative real time RT-PCR for BZLF-1. CD19 transcripts were also analyzed to optimally discriminate between infiltrated and non infiltrated brain areas. BZLF-1 transcripts were detected in the perivascular cuffs isolated from one active MS lesion and in 3 out of 4 meningeal B-cell follicles but not in 3 chronic active lesions, 4 sparse meningeal infiltrates and 7 non infiltrated parenchymal regions (Figure 6). A control lymph node was negative for BZLF-1. Thus, both immunohistochemical and RT-PCR findings corroborated BZLF-1 expression and therefore shift to EBV lytic infection in immunologically active white matter lesions and ectopic B-cell follicles.
Then, we searched for interactions between cytotoxic CD8+ T cells and EBV infected cells in the same MS brain samples in which BZLF-1 protein and/or RNA were detected. We first analyzed the presence and frequency of granzyme B-expressing CD8+ T cells and their relationship to EBV litically infected cells. We observed that most granzyme B+ cells in the MS brain co-expressed CD8 and that the fraction of CD8+ T cells expressing granzyme B ranged between 5 and 60% in the different MS cases and brain areas analyzed, the highest values being detected in the perivascular cuffs of active white matter lesions (Figure 7 A, B). In the meninges, granzyme B+/CD8+ T cells were present in diffuse meningeal infiltrates and at the periphery of B-cell follicle-like structures, but were rarely seen inside these structures (Figure 7 C, D).
Given the nuclear localization of BZLF-1 and the relatively small number of BZLF-1+ cells in the MS brain it was extremely difficult to see contacts between granzyme B+ cells and lytically infected cells using double immunofluorescence for BZLF-1 and CD8 or granzyme B. We therefore stained MS brain sections for the EBV lytic protein BFRF1 which has a perinuclear localization and has been detected in a higher fraction of plasma cells (up to 50%) compared to BZLF-1 [32]. We observed lymphoblastoid CD8+ T cells adhering to or secreting granzyme B toward BFRF1+ cells as well as contacts between CD8+ T cells and BFRF1+ cells displaying granzyme B immunoreactivity on their surface (Figure 7 E–H). Such cytotoxic contacts were frequently observed in sparse meningeal infiltrates and active white matter lesions, but not inside ectopic B-cell follicle-like structures. Finally, staining of MS brain sections for perforin and Ig allowed to visualize perforin granules polarized toward Ig+ cells inside the perivascular cuffs of active white matter lesions (Figure 7 I, J), supporting further the idea that EBV harbouring cells might be the target of a cytotoxic attack.
Altered control of EBV infection in individuals susceptible to MS is suspected to play a role in the development of immune dysfunction causing CNS pathology [3]–[5]. Higher serum titers of EBNA-1 IgG are associated with an increased risk of MS [15], increased conversion from a clinically isolated syndrome to definite MS [16] and more severe disease activity and clinical progression [16], [36]. Virus-specific CD8+ T cell responses play an essential role in the control of EBV infection [45] and have been investigated in previous studies in MS using mainly IFN-γ ELISPOT analysis in PBMCs stimulated in vitro with EBV+ lymphoblastoid cells [20], [22], viral lysates [21], individual [19] or pooled [21] EBV lytic and latent peptides, and more recently using MHC-peptide tetramer staining [23]. Several studies have shown that EBV-specific CD8+ T cell responses are significantly higher in MS than in HD or in patients with other inflammatory neurological diseases [19]–[21]. However, lack of significant differences between MS patients and controls [16], [23] and even reduced frequency of EBV-specific CD8+ T cells in MS patients [22] have also been reported. Use of cryopreserved versus freshly isolated PBMCs, analysis of patients with relapsing-remitting and progressive MS courses, and lack of stratification of patients according to disease activity may have hampered a clear understanding of the possible link between EBV-specific CD8+ T cell responses and MS pathogenesis.
To obtain an accurate pattern of CD8+ T cell in vivo specificities, we have used highly standardized flow cytometric analysis with EBV-specific pentamers, that unequivocally identify antigen-specific CD8+ T cells, on freshly isolated PBMCs obtained from HD and relapsing-remitting MS patients. Importantly, both untreated and treated MS patients were studied and disease activity was evaluated in most patients with gadolinium-enhanced MRI shortly before or at the time of blood collection. Such a rigorous selection justifies the relatively small number of MS patients analyzed.
The first main finding of this study is that differences in the prevalence and magnitude of the CD8+ T cell response to certain EBV latent and lytic proteins between MS patients and HD and within the MS cohort emerge only when patients are stratified according to disease activity and inactivity. By showing that fewer inactive MS patients have a detectable CD8+ T cell response against EBV lytic antigens compared with HD and active MS patients and that the frequency of lytic antigen-specific CD8+ T cells is higher in active MS patients than in HD and inactive MS patients, we demonstrate for the first time that changes in the immune control of EBV replication are associated with the active and inactive phases of MS. This is corroborated by the longitudinal study performed in two untreated MS patients showing a peak in the frequency of CD8+ T cells to EBV lytic antigens during active disease. Of the two EBV lytic antigens analyzed, BZLF-1 is a transactivator expressed at the very initiation of the lytic cycle and is involved in the induction of early lytic proteins, including BMLF-1 [6]. Thus, an increase in BZLF-1- and BMLF-1-specific CD8+ T cells in concomitance with acute brain inflammation on MRI strongly suggests an attempt of the immune system to control intracerebral foci of EBV replication. On the other hand, a logical explanation for the decrease in lytic antigen-specific T cells associated with inactive MS could be elimination of lytically infected cells brought about by the strong cytotoxic response occurring in the active disease phase. In this context, it is important to recall that EBV DNA load in the blood of MS patients does not differ significantly from that in healthy EBV carriers [16], [18], [23] indicating that fluctuations in EBV-specific CD8+ T cell responses in MS patients do not result in a generalized impairment of the immune control of EBV infection. In contrast with the present findings, Jilek et al. [23] did not observe differences in the prevalence and frequency of CD8+ T cell responses to BZLF-1 and BMLF-1 between MS patients and control subjects. However, this study was not restricted to patients with relapsing-remitting MS, did not distinguish between patients with active and inactive disease and used cryopreserved PBMCs. Importantly, in our study both the prevalence and magnitude of the CD8+ T cell response to CMV were similar in HD and untreated MS patients, irrespective of disease activity, indicating that the differences observed in EBV-specific immunity are not the consequence of a general activation of antiviral immune responses driven by a still unknown MS-associated immune dysfunction.
Despite the fact that the prevalence of the CD8+ T cell response to EBV latent antigens in inactive MS patients was similar to that in HD and tended to be lower than in active MS patients, we found that the magnitude of the CD8+ T cell response to EBNA-3A was higher in inactive MS patients than in HD and active MS patients. Very high numbers of EBNA-3A-specific CD8+ T cells were detected in half of the inactive MS patients harbouring this immune reactivity (2.7 to 7.6% of the circulating CD8+ T cells versus <1% in HD and active MS patients). Furthermore, longitudinal monitoring of a patient with a recent diagnosis of MS showed substantial reduction of EBNA-3A-specific CD8+ T cells just before the active disease phase and the rise of lytic antigen-specific CD8+ T cells. It is known that EBNA-3A is expressed shortly after EBV infection of B cells together with the whole set of EBV latent proteins (EBNA-LP, −1, −2, 3A, 3B, and −3C, LMP-1, LMP-2A, LMP-2B) that are essential to drive infected B cells into proliferation (latency III or growth program) [6] and elicit strong T-cell responses [45]. Most EBV-encoded latent antigens, including EBNA-3A, are then sequentially downregulated as EBV establishes a persistent infection in memory B cells (latency II and I programmes) [6]. Thus, the study of EBNA-3A-specific CD8+ T cells suggests that at least in some inactive MS patients there is an attempt by CD8+ T cells to control abnormal expansion of a latently infected B-cell pool. A decrease in the immune response to EBNA-3A could reflect a change in EBV latency programme and, possibly, set the stage for switching to the lytic cycle. Of interest, more abundant EBNA-3A-specific CD8+ T cells were detected only in MS patients with a short disease duration (<5 years). A decrease in immune reactivity toward EBNA-3A with disease progression could be due to reduced antigenic stimulation or to T-cell exhaustion which is known to occur during uncontrolled, chronic viral infections [46]. Relevant to this, we have shown that most EBV latently infected B cells accumulating in the inflamed MS brain during late-stage disease are memory B cells expressing the latency II programme [32], [33].
Based on the present findings, we propose that failure to fully control EBV latent infection in an immune privileged site like the CNS could lead to recrudescence of EBV reactivation. Exposure to newly synthesized viral antigens would promote expansion of lytic antigen-specific CD8+ T-cells targeting intracerebral viral deposits and hence inducing the active phase of MS. Of relevance for the present findings, it has been shown that after primary EBV infection and during establishment of EBV persistence CD8+ T cells specific for some EBV epitopes disappear from the circulation after having upregulated Programmed Death-1 (PD-1) inhibitory receptor, probably as a consequence of inadequate antigenic stimulation [47]. We are currently evaluating whether fluctuations in EBV-specific CD8+ T cells in relapsing-remitting MS might be associated with changes in PD-1 expression levels and T-cell function (i.e., cytokine profile and cytotoxic activity). It would be also interesting to compare the quality of the CD8+ T cell response to EBV in MS with that in systemic lupus erythematosus, an autoimmune disease associated with marked systemic EBV dysregulation [48] and impaired cytotoxic immune response to the virus [49].
The second main finding of this study is that treatment of relapsing-remitting MS patients with IFN-β and natalizumab is associated with marked changes in the CD8+ T cell response to viral antigens. We have shown that CD8+ T cells specific for EBV latent and lytic antigens and for CMV antigen were detectable in a substantial fraction of the patients entering active disease despite IFN-β treatment, but in none, except one, of the IFN-β-treated patients with inactive disease. It is likely that such a strong reduction in the CD8+ T cell response to both viruses might due to the direct antiviral activity of the drug [41]. Recently, Comabella et al. [50] reported that clinically effective IFN-β therapy is associated with downregulation of proliferative T cell responses to EBNA-1 without significant changes in the CD8+ T cell response against other (pooled) EBV antigens of the latent and lytic phase. Discrepancies with the present study could be due to technical issues, as discussed above. We have also shown that most natalizumab-treated MS patients, all of which were in the inactive phase of disease, had a detectable CD8+ T cell response to EBV. Such high prevalences could be related to the fact that natalizumab treatment causes a marked increase in lymphocyte numbers in the blood due to interference with lymphocyte extravasation and trafficking in lymphoid and non-lymphoid tissues [51]. Importantly, we found that in natalizumab-treated MS patients the frequency of CD8+ T cells specific for EBV lytic antigens was similar to that in HD and untreated inactive MS patients and stable over time (up to 22–27 months of therapy). It therefore seems significant that the present analysis, though limited to a relatively small number of donors, consistently showed no expansion of EBV lytic antigen-specific CD8+ T cells during the inactive phase of MS regardless of presence or absence of therapy. In contrast, the frequency of EBNA-3A-specific CD8+ T cells, which was comparable in untreated and natalizumab-treated inactive MS patients within 8–16 months of therapy, progressively increased during the second year of therapy in 2 longitudinally monitored patients. These observations suggest dysregulation of EBV latent infection upon prolonged treatment with natalizumab. Although further studies are needed to clarify these aspects, analysis of EBV-specific CD8+ T cell responses in MS patients may help identify biomarkers useful for therapy monitoring and shed light into the mechanisms underlying drug efficacy.
The third main finding of this study is that BZLF-1, one of the two EBV lytic proteins recognized by CD8+ T cells expanding in the blood of active MS patients, is expressed in post-mortem MS brains with prominent immune infiltrates. The demonstration of BZLF-1 protein and RNA in active white matter lesions, which likely correspond to gadolinium-enhanced MRI lesions, and in the inflamed meninges, where changes in water content cannot be detected on MRI, lends support to the idea that acute brain inflammation in MS is associated with switch to the viral lytic cycle. In line with our previous results [32], we have also shown that in the MS brain EBV reactivates in plasma cells and that the latter can be found in close contact with lymphoblastoid CD8+ T cells producing cytolotic enzymes. However, absence of CD8+ granzyme B+ T cells inside meningeal B-cell follicles, which contain a high frequency of EBV latently infected cells [32]–[34], suggests that a local suppressive environment created by the virus itself to elude immune control [52] could hamper virus clearance from these structures. A cytotoxic attack toward EBV infected cells in the MS brain is consistent with enrichment in EBV-specific CD8+ T cells in the cerebrospinal fluid (CSF) of patients with early MS [53], with increased CSF levels of granzymes during relapse in relapsing-remitting MS patients [54], and with preferential expansion of CD8+ T cells in MS brain lesions and CSF [55], [56].
Defects in the control of viral infections are suspected to promote the development of autoimmune diseases [57]. Nearly all of the genes whose variants have been associated with the risk of developing MS are implicated in immune system function [2], making it plausible that in susceptible individuals subtle differences in the regulation of the immune response might allow an EBV infection to be established in the CNS and become the target of an immunopathological response. Experimental studies suggest that upon infection with persistent viruses establishment of extralymphatic viral sanctuaries depends both on organ anatomy and defective synergies between CD8+ T-cell- and antibody-mediated immune responses [58]. The present results do not answer the question of whether EBV dysregulation is consequence or cause of MS but disclose a link between EBV reactivation, antiviral immune response and disease activity during the relapsing-remitting stage of MS. Such a scenario is consistent with the results of randomized, double-blind, placebo-controlled clinical and MRI studies of anti-herpesvirus therapy in relapsing-remitting MS showing that anti-herpesvirus drugs inhibiting viral replication have beneficial effects in subgroups of patients with higher exacerbation rates and more severe disease activity [59], [60]. Further work is needed to better understand whether and how an altered balance between EBV and the host immune system contributes to MS onset and verify the potential benefits of new antiviral drugs in controlling MS [61].
All blood samples were obtained following acquisition of the study participants' written informed consent. The study protocol was reviewed and approved by the local ethics committes of S. Camillo Forlanini Hospital, Tor Vergata University, S. Andrea Hospital, and Fondazione S. Lucia. Use of post-mortem human brain material for the study purposes has been approved by the ethics committee of the Istituto Superiore di Sanità.
MS patients and HD were recruited between 2008 and 2012 from Tor Vergata University, S. Camillo Forlanini and S. Andrea Hospitals in Rome. We enrolled 250 patients who were diagnosed the relapsing-remitting form of MS according to the 2005 revised McDonald's criteria [62]. A neurologist (SR, CG, FB, DC, MS) examined the patients, including assessment of the EDSS and confirmation of clinical relapse or remission. Of the 250 enrolled subjects, 113 MS patients were selected for this study according to their HLA genotype (HLA-A*02101, B*0801) for which well characterized EBV and CMV peptide antigens have been described [38]–[40]. Seventy-nine patients were free of therapy for at least 3 months, 20 patients were treated with IFN-β subcutaneously (12 with IFN-β 1a and 8 with IFN-β 1b) for 1–11 years (median = 4 years) and 14 MS patients were treated with natalizumab for 8–16 months (median = 12 months). The control subject group included 43 HD matched for sex and age and selected for their HLA genotype (HLA-A*02101, B*0801). The demographic and clinical characteristics of HD and MS patients are summarized in Table 1. At the time of peripheral blood collection 38 and 75 MS patients were in the active and inactive phase of the disease, respectively, based on clinical assessment and brain MRI (Table 2). Four MS patients (2 untreated and 2 treated with natalizumab) and 3 HD were monitored longitudinally for 7–27 months and blood was drawn every 3 to 6 months.
Seventy-six % (60/79) of untreated patients and all IFN-β-treated patients were examined by brain MRI with gadolinium enhancement on the same day or within 1 week before blood collection; only in one case MRI was performed 3 weeks before blood collection. All natalizumab-treated patients were monitored with MRI every 6 months. Acquisition of brain MRI scans was obtained in a single session. Conventional MRI scans were acquired including the following sequences: Fast Fluid Attenuated Inversion Recovery (FLAIR), T1 weighted images (T1-WI) before and after gadolinium administration covering the whole brain. The gadolinium enhanced T1-WI scans were obtained for all patients 15 minutes after admnistration of gadolinium (0,1 mmol/kg). MRI scans were classified as active if there was at least one gadolinium enhancing lesion. As shown in Table 2, the majority (86%) of active MS patients included in this study had both clinical manifestations and an active MRI scan, while a minority showed either clinical (6%) or MRI (8%) evidence of disease activity. Conversely, all inactive patients exhibited neither clinical manifestations nor disease activity on MRI.
PBMCs were isolated on a Ficoll gradient (Ficoll-Paque PLUS, GE Healthcare) and stained with pre-titrated antibodies. To evaluate the CD8+ T cell response to EBV latent and lytic antigens, PBMC from MS patients and HD were stained with fluorochrome-labeled pentamers (ProImmune, Oxford, UK). The analysis was conducted on freshly isolated PBMC with the exclusion of dead cells, providing an accurate pattern of CD8+ T cell in vivo specificities. We analyzed CD8+ T cells specific for two EBV lytic protein epitopes, the HLA-A*0201 restricted epitope (GLCTLVAML) from BMLF-1 and the HLA-B*0801 restricted epitope (RAKFKQLL) from BZLF-1, and for two EBV latent protein epitopes, the HLA-A*0201 restricted epitope (CLGGLLTMV) from LMP-2A and the HLA-B*0801 restricted epitope (FLRGRAYGL) from EBNA-3A. The CD8+ T cell response to an HLA-A*0201 restricted immunodominant peptide (NLVPMVATV) from pp65 of human cytomegalovirus (CMV) was studied as a control for anti-viral MHC class I restricted CD8 T-cell responses.
One x 106 PBMCs were stained with 10 µl of PE conjugated-pentamers alone, washed with PBS and then stained with anti human CD3 APC Alexa e780 (eBioscience Inc., San Diego, CA) and CD8 ECD (Beckman Coulter, Brea, CA). Cells were also stained for dead cell exclusion (Fixable Dead Cell Stain Kits, Invitrogen, Life Technologies, Paisley, UK). The samples were acquired on a CyAN ADP cytometer (Beckman Coulter) and analysed by FlowJo software (Tree Star, Ashland, OR). Frequencies of pentamer+ cells below 0.02% of CD3+ T cells were considered as background staining as indicated by the manufacturer. An example of the gating strategy used to identify pentamer+ cells is shown in Figure S1.
Thirteen cerebral tissue blocks from 7 MS cases (MS79, MS92, MS121, MS154, MS180, MS234, MS342) who died in the secondary progressive phase of MS and were characterized by substantial brain inflammation were analyzed in this study. Tissues were provided by the UK Multiple Sclerosis Tissue Bank at Imperial College in London after collection via a prospective donor program with fully informed consent. Based on the available clinical documentation, all MS patients were in the progressive phase of the disease, and no immunotherapy is reported in the 6 months before death. Control tissues for BZLF-1 immunohistochemistry included fixed-frozen brain sections from one control subject who died for cardiac failure (obtained from the UK MS Tissue Bank), paraffin sections of a brain with tuberculous meningo-encephalitis and of an EBV-negative brain B-cell lymphoma (kindly provided by Dr R. Hoftberger, Clinical Institute of Neurology, Wien), and paraffin sections of a tonsil from a subject with active infectious mononucleosis (kindly provided by Dr G. Niedobitek, Sana Klinikum Lichtenberg/Unfallkrankenhaus, Berlin).
Eight brain tissue blocks (4 cm3 each; 1 snap frozen, 7 fixed frozen) from 5 MS cases (MS92, MS121, MS154, MS180, MS342) were used for immunohistochemical studies. Lesion inflammatory activity was assessed as previously described [63]. Five snap-frozen brain tissue blocks from 4 MS cases (MS79, MS92, MS180, MS342) were used to study BZLF-1 gene expression using quantitative real-time RT-PCR. One snap-frozen control lymph node was obtained from Dr Egidio Stigliano, Policlinico A. Gemelli, Rome.
Brain sections were stained using immunohistochemical and single or double indirect immunofluorescence techniques. Immunohistochemical detection of CD20, MHC class II antigen and myelin-oligodendrocyte glycoprotein (MOG), and immunofluorescence stainings for BFRF1, Ig-A,-G,-M, CD8 and perforin alone or in different combinations were performed as previously described [32]. For BZLF-1 immunohistochemistry, deparaffinised sections from infectious mononucleosis tonsil, cerebral B cell lymphoma and brain with tuberculous meningo-encephalitis and cryosections from PFA-fixed control brain were subjected to antigen retrieval procedure in citrate buffer in microwave for 3 cycles of 3 min each before quenching of endogenous peroxidase activity in PBS containing 0,1% H2O2. Sections were treated with 0.5% Triton X-100 in PBS for 10 minutes and incubated for 1 hour with normal serum 30%+0,25% Triton X-100 and then overnight at 4°C with mouse monoclonal antibody (mAb) specific for BZLF-1 protein (clone BZ-1, kindly provided by Dr J. Middeldorp, VUMC, Amsterdam) diluted 1∶50 in PBS containing 0,25% Triton X-100. Sections were then incubated with biotin-conjugated rabbit anti-mouse antibody (Jackson Immunoreaearch Laboratories, West Grove, PA) for 1 hour at room temperature (RT), ABC-peroxidase complex (Vector Laboratories, Burlingame, CA) for 45 min, and AEC (DakoCytomation, Glostrup, Denmark) or diaminobenzidine (Sigma, St Louis, MO) to reveal the immune reaction.
The EBV-producing B95-8 cells (marmoset B-cell line transformed with EBV) [64] were used as positive control for BZLF-1 immunofluorescence staining. Paraformaldehyde-fixed frozen brain sections from MS cases and one control case were air-dried and post-fixed in 4% PFA for 5 minutes at RT or in iced acetone for 10 minutes at 4°C. Sections were subjected to antigen unmasking, permeabilization steps and block of unspecific binding sites as described above and then incubated for 36 h at 4°C with BZ-1 mAb (diluted 1∶50 in PBS +0,1% Triton X-100). Antibody binding was visualized using tetramethyl rhodamine isothiocyanate (TRITC)-conjugated goat anti-mouse antibody (Jackson Laboratories) containing 5% normal goat serum for 50 minutes at RT. After washing, sections were sealed in DAPI-containing medium or incubated further with FITC-conjugated rabbit anti-human Ig-A-G-M (1∶400; Dako Cytomation) for 1 hour at RT. For double immunofluorescence for BZLF-1 and CD8, sections were stained with BZ-1 mAb and anti-human CD8 rabbit polyclonal antibody (1∶50; Pierce, Thermo Fisher Scientific Inc. Rockford, IL) followed by a mixture of Alexa Fluor 488-conjugated goat anti-mouse and TRITC-conjugated goat anti-rabbit secondary antibodies. Double and triple immunostainings for CD8/granzyme B and CD8/granzyme B/BFRF1 were performed by incubating PFA-fixed cryosections with a mixture of anti-CD8 rabbit polyclonal antibody and anti-granzyme B mAb (Dako), or anti-CD8 mAb, anti-BFRF1 rabbit polyclonal (1∶800) and anti-granzyme B goat polyclonal (1∶50, Santa Cruz) antibodies overnight at 4°C, and then with a mixture of donkey FITC anti-mouse (Invitrogen, Eugene, OR), TRITC anti-rabbit and AMCA anti-goat (Jackson Immunoresearch Lab) secondary antibodies. Sections were sealed in ProLong Gold antifade reagent with 4′,6′-diamidino-2-phenylindole (DAPI) (Invitrogen) or in Vectashield (Vector Laboratories). Images were analysed and acquired with a digital epifluorescence microscope (Leica Microsystem, Wetzlar, Germany). Negative control stainings were performed using Ig isotype controls and/or pre-immune sera.
Snap-frozen brain tissue blocks from 4 MS cases (MS79, MS92, MS180, MS342) and control lymph node were used for laser capture microdissection and subsequent RNA analysis. For each tissue block, the integrity and quality of total RNA extracted with the SV Total RNA Isolation System (Promega, Madison, WI) were checked on ethidium bromide containing 1% agarose gels in Tris-borate/EDTA buffer. Ten to 20 serial brain sections for each MS case and from control lymph-node were mounted on membrane slides for laser capture microdissection (MMI AG, Glattbrugg, Switzerland) and processed as described previously [33]. Sections before and after these series were stained for CD20 and Ig to identify B cell- and plasma cell-containing regions in the inflamed meninges and white matter lesions. Using a laser microdissector SL Cut (MMI AG) equipped with a Nikon Eclipse TE2000-S microscope, we isolated areas containing meningeal infiltrates and B-cell follicles, lesioned grey matter, B cell-enriched perivascular cuffs in white matter lesions, lesioned white matter surrounding inflamed blood vessels, and normal-appearing white matter. The same brain areas were cut in 3 to 10 serial sections and pooled in a single cap. B cell follicles were isolated from the lymph-node. The isolated tissue fragments were collected in 50 µL of lysis buffer (PicoPure RNA isolation kit, Arcturus Engineering), incubated at 42C° for 30 minutes and centrifuged at 800× g for 2 minutes. Lysates were stored at −80 C° until use.
DNase-treated total RNA was extracted from 20-µm-thick brain sections or microdissected areas from 4 MS brains and 1 control lymph node, as previously described [32]. RNA samples were reverse-transcribed with oligo (dT) and random hexamers using the Murine Leukemia Virus Reverse Transcriptase (Invitrogen Life Technologies, Carlsbad, CA). PreAMP Master Mix Kit (Applied Biosystems, Foster City, CA) was used to enrich for both cellular and viral gene transcripts. The cDNAs obtained from whole brain sections and microdissected brain and lymph node samples were preamplified according to the manufacturers' instructions using 90 nmol/L of each primer in a mix containing the same forward and reverse primers for GAPDH, CD19 and BZLF-1 used for real time RT-PCR. Quantitative PCR assays were performed in triplicate, as previously described [33]. cDNA from EBV-positive P3HR-1 cells and human primary B cells were included in each run as positive controls for BZLF-1 and CD19 gene expression, respectively. Sample values were normalized by calculating the relative quantity of each mRNA to that of GAPDH using the formula 2−ΔCt, where ΔCt represents the difference in cycle threshold (Ct) between target mRNA and GAPDH mRNA. The following primer pairs were used in this study: GAPDH_for ACAGTCCATGCCATCACTGCC; GAPDH_rev GCCTGCTTCACCACCTTCTTG [33]; BZLF-1_for GTTGTGGTTTCCGTGTGC; BZLF-1_rev AACAGCTAGCAGACATTGGTG [65]; CD19_for AGAACCAGTACGGGAACGTG; CD19_rev CTGCTCGGGTTTCCATAAGA [33].
Differences between categorical variables were evaluated by Pearson's chi-squared test while differences between continuous variables were analysed by unpaired t-test with 95% confidence intervals.
|
10.1371/journal.pntd.0001636 | Development of a Humanized Antibody with High Therapeutic Potential against Dengue Virus Type 2 | Dengue virus (DENV) is a significant public health threat in tropical and subtropical regions of the world. A therapeutic antibody against the viral envelope (E) protein represents a promising immunotherapy for disease control.
We generated seventeen novel mouse monoclonal antibodies (mAbs) with high reactivity against E protein of dengue virus type 2 (DENV-2). The mAbs were further dissected using recombinant E protein domain I-II (E-DI-II) and III (E-DIII) of DENV-2. Using plaque reduction neutralization test (PRNT) and mouse protection assay with lethal doses of DENV-2, we identified four serotype-specific mAbs that had high neutralizing activity against DENV-2 infection. Of the four, E-DIII targeting mAb DB32-6 was the strongest neutralizing mAb against diverse DENV-2 strains. Using phage display and virus-like particles (VLPs) we found that residue K310 in the E-DIII A-strand was key to mAb DB32-6 binding E-DIII. We successfully converted DB32-6 to a humanized version that retained potency for the neutralization of DENV-2 and did not enhance the viral infection. The DB32-6 showed therapeutic efficacy against mortality induced by different strains of DENV-2 in two mouse models even in post-exposure trials.
We used novel epitope mapping strategies, by combining phage display with VLPs, to identify the important A-strand epitopes with strong neutralizing activity. This study introduced potential therapeutic antibodies that might be capable of providing broad protection against diverse DENV-2 infections without enhancing activity in humans.
| Dengue virus (DENV) infection remains a serious health threat despite the availability of supportive care in modern medicine. Monoclonal antibodies (mAbs) of DENV would be powerful research tools for antiviral development, diagnosis and pathological investigations. Here we described generation and characterization of seventeen mAbs with high reactivity for E protein of DENV. Four of these mAbs showed high neutralizing activity against DENV-2 infection in mice. The monoclonal antibody mAb DB32-6 showed the strongest neutralizing activity against diverse DENV-2 and protected DENV-2-infected mice against mortality in therapeutic models. We identified neutralizing epitopes of DENV located at residues K310 and E311 of viral envelope protein domain III (E-DIII) through the combination of biological and molecular strategies. Comparing the strong neutralizing activity of mAbs targeting A-strand with mAbs targeting lateral ridge, we found that epitopes located in A-strand induced stronger neutralizing activity than those located on the lateral ridge. DB32-6 humanized version was successfully developed. Humanized DB32-6 variant retained neutralizing activity and prevented DENV infection. Understanding the epitope-based antibody-mediated neutralization is crucial to controlling dengue infection. Additionally, this study also introduces a novel humanized mAb as a candidate for therapy of dengue patients.
| Dengue is the most important arthropod-borne viral disease in humans and an increasing public health concern in tropical and subtropical regions of the world. Approximately 50–100 million cases of dengue fever (DF) and 500,000 cases of dengue hemorrhagic fever (DHF) occur every year, and 2.5 billion people are at risk of dengue infection globally [1], [2]. Dengue infection may lead to fever, headache and joint pain in milder cases but may also lead to the more severe life-threatening DHF/dengue shock syndrome (DSS) has plasma leakage, thrombocytopenia, and hemorrhagic manifestations, possibly leading to shock [3], [4].
Dengue virus (DENV) is positive-sense single-stranded RNA virus of approximately 11 kb genome of the genus Flavivirus, a family Flaviviridae. It has four genetically and antigenically related viral serotypes: DENV-1, -2, -3 and -4. Flaviviruses encode a single polyprotein processed by host and viral protease to produce three structural proteins, including capsid (C) protein, precursor membrane/membrane (prM/M) and envelope (E) protein, and seven nonstructural proteins: NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5 [5]. The E protein, a 53 kDa glycoprotein important for attachment, entry, and viral envelope fusion, can bind to cellular receptors and induce neutralizing antibodies [6], [7].
The DENV consists of an icosahedral ectodomain, containing 180 copies of the E protein [8]. E protein monomer contains three structural and functional domains [9], [10]. E protein domain I (E-DI) is a central β-barrel structure. E protein domain II (E-DII) is organized into two long finger-like structures and contains the flaviviruses conserved fusion loop. E protein domain III (E-DIII) has an immunoglobulin-like fold and may mediate interactions between the virus and the receptors on the host cell [11]. Studies of the biological characteristics and epitope specificities of mouse monoclonal antibodies (mAbs) have elucidated the antigenic structure of flavivirus E proteins [12]–[15]. Serotype-specific mAbs with neutralizing activity against DENV-2 have been found to be located on the lateral ridge of E-DIII and the subcomplex-specific mAbs recognized A-strand of E-DIII [14], [16], [17]. Antibody-mediated neutralization has been found to alter the arrangement of viral surface glycoproteins that prevent cells from viral attachment [16]. Binding of an antibody to the viral surface can interfere with virus internalization or membrane fusion [6].
Primary DENV infection is believed to provide lifelong immunity against re-infection with the same serotype [18], [19]. However, humoral immune responses to DENV infection are complex [20]–[22], and may exacerbate the disease during heterologous virus infection [18], [19]. Antibody-dependent enhancement (ADE) in dengue pathogenesis results from the increase in the efficiency of virus infection in the presence of non-neutralizing or sub-neutralizing concentrations of anti-E or anti-prM immunoglobulins [21], [23]. The attachment of antibody-virus complex to such Fcγ receptor-bearing cells as monocytes and macrophages can lead to an increased virus replication [18], [24], [25].
A better understanding of the neutralizing epitopes may facilitate the generation of new antibody-based therapeutics against DENV infection. In this study, we generated several mAbs against DENV-2. We found that serotype-specific anti-E-DIII mAbs played an important role in the neutralization of virus infectivity. Studies of the neutralizing epitopes found the strongest mAbs to be DB32-6 and DB25-2, both DENV-2 serotype-specific antibodies. These two mAbs recognized the A-strand of E-DIII at residues K310 and E311, respectively. Humanized DB32-6 mAb efficiently neutralized DENV-2 infection in a therapeutic mouse model and its variant version prevented enhancing activity.
BHK-21 cells were grown at 37°C with 5% CO2 in Minimal Essential Medium (MEM, Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS, Gibco) and 100 U/ml penicillin, 100 µg/ml streptomycin, 0.25 µg/ml amphotericin B (Antibiotic-Antimycotic, Gibco). Aedes albopictus C6/36 cells were grown at 28°C in 1∶1 Mitsuhashi and Maramorosch (MM) insect medium (Sigma-Aldrich)/Dulbecco's modified Eagle's medium (DMEM, Gibco) containing 10% FBS and 100 U/ml penicillin, 100 µg/ml streptomycin, 0.25 µg/ml amphotericin B (Antibiotic-Antimycotic, Gibco). The four DENVs (DENV-1 Hawaii, DENV-2 16681, DENV-3 H87 and DENV-4 H241) were provided by Dr. Duane J. Gubler from the Centers for Disease Control and Prevention, Fort Collins, U.S.A. The various DENV-2 strains including New Guinea-C (NGC), NGC-N (mouse-adapted neurovirulent), PL046 and Malaysia 07587 were used in this study [26], [27]. These viruses were passaged in C6/36 cells.
Anti-DENV-2 mAbs were generated according to previously described procedures [28], [29]. Female 4- to 6-week-old BALB/c mice were immunized with 107 plaque-forming units (pfu) of DENV-2 (16681). The DENV-2 was purified from viral culture supernatant using 4G2 (an anti-E protein mAb)-coupled protein G-Sepharose 4 Fast Flow gel. After four inoculations with the same concentration of antigens, the splenocytes from the immunized mouse spleen were harvested and then fused with mouse myeloma NS-1 cells. Fused cells were cultured in DMEM supplemented with 15% FBS, HAT medium and hybridoma cloning factor (Roche) in 96-well tissue culture plates. Two weeks after fusion, culture supernatants were screened by ELISA. Selected clones were subcloned by limiting dilutions. Hybridoma clones were isotyped using a commercially isotyping kit (Southern Biotech) by ELISA. Ascites fluids were produced in pristine-primed BALB/c mice. mAbs were affinity-purified by standard protein G-Sepharose 4 Fast Flow (GE Healthcare Bio-Sciences) according to manufacturer's directions.
C6/36 cells at 80% confluency in 96-well plates were infected with DENV-1 to -4 to produce viral antigens. These cells were then harvested 5–7 days after infection. One µg/ml mAbs was added to the plates and incubated at room temperature (RT) for 1 h. After washing with PBS, horseradish peroxidase (HRP)-conjugated anti-mouse IgG (Jackson ImmunoResearch Laboratories) was incubated at RT for 1 h. Finally, plates were incubated with peroxidase substrate o-phenylenediamine dihydrochloride (OPD; Sigma-Aldrich). Reaction was stopped with 3N HCl and optical density was measured using a microplate reader set at 490 nm.
C6/36 cells were harvested after viral infection. Lysates or expression proteins were collected. Cell extracts were mixed with sample buffer (Bio-Rad Laboratories). Protein samples were separated by SDS-PAGE and transferred to nitrocellulose membrane (Hybond-C Super). Nonspecific antibody-binding sites were blocked with 5% skimmed milk in PBS, and membranes were incubated with primary antibody. Blot was then treated with horseradish peroxidase-conjugated goat anti-mouse immunoglobulin (Jackson ImmunoResearch Laboratories) and then developed with enhanced chemiluminescence reagents (ECL, Thermo Fisher Scientific).
BHK-21 cells at 80% confluency were infected at a multiplicity of infection (MOI) of 0.5 with DENV-2 (16681). After 2 days infection, the cells were fixed with 1∶1 methanol/acetone for 10 min at −20°C. Cells were blocked using PBS supplemented with 1% BSA for 1 h at RT. Primary anti-DENV antibodies or control antibodies (normal mouse IgG, Jackson ImmunoResearch Laboratories) were diluted (1∶250) in block solution for 1 h at RT. Secondary antibody, FITC-conjugated goat anti-mouse IgG (Jackson ImmunoResearch Laboratories) was diluted to 1∶250 and supplemented with DAPI (Invitrogen) diluted 1∶2,000 for 1 h at RT. The binding activity of antibodies to the DENV-2-infected or transfected cells were observed and photographed through a fluorescence microscope.
The expression constructs of E-DI-II and E-DIII were cloned into the pET21a vector (Merck). The E-DI-II, comprising amino acids 1–295 of the E protein, was tagged to flag and hexahistidine at the C terminus for affinity purification. The E-DIII, comprising amino acids 295–400 of the E protein, was tagged to flag and hexahistidine, too. The plasmids were expressed in Escherichia coli strain BL21 (DE3). The recombinant proteins E-DI-II and E-DIII were analyzed using 12% SDS-PAGE by Western blot analysis. The DNA fragments corresponding to E-DI-II and E-DIII were also cloned into a mammalian expression vector, pcDNA3.1 (Invitrogen). The expression constructs of DENV-2 C, prM, prM-E, E, NS1, NS2A, NS2B, NS2B-3, NS3, NS4A, NS4B and NS5 were obtained from Dr. Y.-L. Lin [30]. Transient expression of DENV-2 proteins in BHK-21 cells was transfected by PolyJet (SignaGen Laboratories) according to manufacturer's recommendations and then to test specificity of mAbs.
(i) For the plaque reduction neutralization test (PRNT), eight 3-fold serial dilutions of mAbs (from 200 µg/ml to 0.1 µg/ml) were mixed with an equal volume of 200 pfu of DENV-2 (16681) and incubated at 4°C for 1 h. The final concentration of mAbs at the PRNT ranged from 100 to 0.05 µg/ml. Antibody-virus mixtures (100 µl) were added to BHK-21 cells at 80%–90% confluency in 12-well plates. After absorption of virus for 2 h, BHK-21 cells were washed and 2 ml of 1% (w/v) carboxyl methyl cellulose (Sigma-Aldrich) in MEM plus 2% (v/v) FBS was layered onto the infected cells. After incubation at 37°C for 5 to 7 days, the viral plaque that had formed on the cell monolayer was fixed by 1 ml 3.7% formaldehyde (Sigma-Aldrich) at RT for 1 h. The cells were then stained with 1% crystal violet. Percentage of plaque reduction was calculated as: %Inhibition = 100−[(plaque number incubated with mAb/plaque number without mAb)×100]. (ii) For flow cytometry, serial dilutions of DB32-6 mAb were incubated with DENV-2 (16681, NGC, PL046 and Malaysia 07587) at MOI of 0.5 at 4°C 1 h before adding BHK-21 cells. After 2 h absorption, the monolayers were washed and incubated with MEM (Gibco) plus 2% (v/v) FBS at 37°C for 2 days. The cells infected with DENV-2 were washed and fixed with 3.7% formaldehyde at 4°C for 10 min. They were then permeabilized in PBS supplemented with 1% FBS, 0.1% saponin (Sigma) at 4°C for 10 min. For staining, cells were incubated with 4G2 at a concentration of 1 µg/ml at 4°C for 30 min. After two washes, R-Phycoerythrin (PE)-conjugated AffiniPure F(ab′)2 fragment goat anti-mouse IgG (H+L) (Jackson ImmunoResearch Laboratorie) diluted 1∶250 was then added at 4°C for 30 min followed by two washes and analyzed by flow cytometry. % Infection = (the intensity of cells incubated with mAb/without mAb)×100.
This study was carried out following strict guidelines from the care and use manual of National Laboratory Animal Center. The protocol was approved by the Committee on the Ethics of Animal Experiments of Academia Sinica. (Permit Number: MMiZOOWH2009102). The mice were killed with 50% CO2 containing 50% O2. All efforts were made to minimize suffering. (i) Breeder mice of the ICR strain were purchased from the Laboratory Animal Center National Taiwan University College of Medicine. Purified mAbs at doses of 1, 10 and 100 µg/ml were incubated with 1×104 pfu (25-fold LD50) of DENV-2 (16681) at 4°C for 30 mins. Two-day-old suckling mouse brain was inoculated with 20 µl of the reaction mixture by intracranial (i.c.) injection. Survival rate and signs of illness, including paralysis, were observed daily for 21 days following challenge. In post-exposure therapeutic experiments, mice were passively injected with 5 µg of mAb via i.c. route after 1 day of infection. (ii) Stat1-deficient mice (Stat1−/−) [31] were bred in the specific-pathogen-free animal facility at the Institute of Biomedical Sciences, Academia Sinica. Mice were challenged intraperitoneally with 1×105 pfu (300-fold LD50) of DENV-2 (NGC-N) in 300 µl of PBS and simultaneously injected intracranially (i.c.) with 30 µl of PBS. In prophylaxis experiments, antibodies (100 µg per mouse, intraperitoneally) were administered 1 day before infection and administered on day 0, 1, 3, 5 and 7 after infection. In postexposure therapeutic experiments, antibodies (100 µg per mouse, intraperitoneally) were administered on day 1, 3, 5 and 7 after infection.
The phage display biopanning procedures were performed according to previous reports [28], [32]. Briefly, an ELISA plate was coated with mAbs at 100 µg/ml. Samples of 100 µl diluted mAb were then added to wells and incubated at 4°C for 6 h. After washing and blocking, the phage-displayed peptide library (New England BioLabs, Inc.) was diluted to 4×1010 pfu of phage and incubated for 50 mins at RT. After washing, bound phage was eluted with 100 µl 0.2 M glycine/HCl (pH 2.2) and neutralized with 15 µl 1 M Tris/HCl (pH 9.1). The eluted phage was amplified in ER2738 for subsequent rounds of selection. The phage was titrated onto LB medium plates containing IPTG and X-Gal. The biopanning protocol for the second and third rounds was identical to the first round except for the addition of 2×1011 pfu of amplified phage for biopanning.
An ELISA plate was coated with 50 µl mAbs 50 µg/ml. After washing and blocking, amplified phage diluted 5-fold was added to coated plate and incubated at RT for 1 h. After washing, 1∶5000 diluted HRP-conjugated anti-M13 antibody (GE Healthcare) was added at RT for 1 h. OPD developed and was terminated with HCl. Optical density was measured at 490 nm.
We used the recombinant expression plasmid pCBD2-2J-2-9-1 [33] to generate VLP mutants. Various VLP mutants were generated by site-directed mutagenesis derived from pCBD2-2J-2-9-1 as a template. PCR was performed using pfu ultra DNA polymerase (MERCK) and all mutant constructs were confirmed by sequencing. BHK-21 cells at 80%-90% confluency in 48-well plates were transfected with plasmids of various VLPs. After two days transfection, the cells were washed with PBS supplemented with 1% FBS, fixed with 3.7% formaldehyde at 4°C for 10 min, and then permeabilized in PBS supplemented with 1% FBS, 0.1% saponin (Sigma-Aldrich) at 4°C for 10 min. For staining, cells were incubated with mAbs at 4°C for 30 min, DB32-6, DB25-2, 3H5 and mix mAbs (4G2, DB2-3, DB13-19, DB21-6 and DB42-3) at a concentration of 0.1, 1, 1 and 1 µg/ml, respectively. After being washed twice, R-Phycoerythrin (PE)-conjugated AffiniPure F(ab′)2 fragment goat anti-mouse IgG (H+L) (Jackson ImmunoResearch Laboratories) diluted to 1∶250 was then added at 4°C for 30 min and analyzed by flow cytometry. Relative recognition was performed according to previously described procedures and calculated as [intensity of mutant VLP/intensity of WT VLP] (recognized by a mAb)×[intensity of WT VLP/intensity of mutant VLP] (recognized by mixed mAbs) [34].
Total RNA was extracted from hybridoma cells using the TRIzol reagent (Invitrogen) and mRNA was isolated with the NucleoTrap mRNA Mini Kit (Macherey-Nagel GmbH & Co. KG.). Purified mRNA was reverse transcribed using oligo (dT) as a primer in a ThermoScript RT-PCR system (Invitrogen). The variable heavy- and light-chain domains (VH and VL) were amplified from the cDNA product by PCR with a variety of primer sets [35], [36]. The PCR products were cloned using the TA kit (Promega) and the VH and VL sequences were determined by DNA sequencing. Software Vector NTI was used for sequence analysis. From these sequences, the framework regions (FRs) and complementarity-determining regions (CDRs) were analyzed by comparing them with those found in the Kabat database and the ImMunoGeneTics database [37].
Two human genes, GenBank accession DI084180 and DI075739, were 94.7% and 92.2% identical to DB32-6 VH and VL, respectively. Humanized DB32-6 VH consisted of the modified FR1 to FR4 from the accession DI084180 gene, and the CDR1 to CDR3 of the DB32-6 VH, respectively, while humanized DB32-6 VL consisted of the modified FRs from the accession DI075739 gene and the CDRs of the DB32-6 VL. Both were synthesized (GENEART) and amplified by PCR using pfu Turbo DNA polymerase (EMD Bioscience). The resulting VH was cloned into modified expression vector pcDNA3.1 (Invitrogen) with a signal peptide and human IgG1 constant region, while the VL was cloned into modified expression vector pSecTag (Invitrogen). We generated a variant of humanized DB32-6 (hDB32-6 variant) in which leucine residues at positions 1.2 and 1.3 of CH2 domain were substituted with alanine residues [38]. The VH and VL plasmids were cotransfected into CHO-K1 cells and selected by G 418 and puromycin for 2–3 weeks. Transformed cells were limit diluted in 96-well plates. After two weeks, stable clones produced humanized antibodies in the McCoy's 5A medium (Sigma-Aldrich), as identified by ELISA. Humanized antibodies were produced by CELLine AD 1000 (INTEGRA Biosciences) according to manufacturer's directions.
Murine and humanized DB32-6 mAbs affinity analysis for E-DIII of DENV-2 was performed by surface plasmon resonance (BIAcore X, Biacore, Inc). Purified E-DIII (50 µg/ml) was immobilized on a CM5 sensor chip (Biacore, Inc) and injected at a flow rate of 10 µl/min. The mAbs were diluted to 4, 2, 1, 0.5, 0.25 and 0 nM in HBS-EP buffer (Biacore, Inc). mAbs were injected at a flow rate of 30 µl/min for 3 min and then allowed to dissociate over 1.5 min. Regeneration of the surface was achieved with an injection of 10 mM glycine HCl/0.2 M NaCl (pH 3.0) before each mAb injection. The data were analyzed by the BIAevaluation software with a global fit 1∶1 binding model.
Serial dilutions of mAbs were mixed with DENV-2 (16681) at MOI of 1 at 4°C for 1 h. The 100 µl mixture were incubated with 5×104 K562 cells [39] in 96-well plates at 37°C for 2 h. After infection, the cells were washed and incubated with RPMI (Gibco) plus 2% (v/v) FBS at 37°C for 2 days. The cells were washed with PBS supplemented with 1% FBS, fixed with 3.7% formaldehyde, and permeabilized in PBS supplemented with 1% FBS, 0.1% saponin (Sigma) at 4°C for 10 min. For staining, cells were incubated with DB42-3 at a concentration of 3 µg/ml at 4°C for 30 min. After two times washes, R-Phycoerythrin (PE)-conjugated AffiniPure F(ab′)2 fragment goat anti-mouse IgG (H+L) (Jackson ImmunoResearch Laboratories, West Grove, PA) diluted 1∶250 was then added at 4°C for 30 min follow by two times wash steps and analyzed by flow cytometry.
Survival rate was expressed using Kaplan-Meier survival curve, and log rank test was used to determine the significant differences. For body weight change experiments, paired t-test was used to determine the significant differences, * P<0.05, ** P<0.01.
Seventeen mAbs with high reactivity against E protein of DENV-2 were generated after immunization of mice with DENV-2 strain 16681. We identified 17 mAbs belonging to the IgG isotype that reacted with DENV-2-infected cells but not with mock-infected cells using immunofluorescence assay (IFA) (Figure S1) and ELISA (Figure 1A). 4G2 was a pan-flavivirus mAb that could recognize the fusion loop of E-DI-II, and 3H5 (ATCC HB46) was a DENV-2 serotype-specific mAb that could recognize the lateral ridge of E-DIII [12], [17], [40]. Both 4G2 and 3H5 were used as positive controls (Figure 1). The specificities of the mAbs recognized as the four DENVs were further confirmed by ELISA and Western blotting (Figures 1A–1B and Table 1). Based on our Western blot analysis using a nonreducing condition, 14 of the mAbs recognized E protein (53 kDa) (Figure 1B). Three mAbs could not be identified by Western blotting. In order to identify the target proteins of these mAbs, we prepared BHK-21 cells transfected with plasmids expressing DENV-2 C, prM, prM-E, E, NS1, NS2A, NS2B, NS2B-3, NS3, NS4A, NS4B and NS5 (Figure S2). Results indicated that three mAbs (DB21-6, DB22-4 and DB36-2) recognized E protein (Figure 1C). The identification and characterization of the 17 mAbs are summarized in Table 1.
To characterize the antigenic structure of the DENV E protein and to study the relationship between epitopes and their neutralizing potency, we constructed and expressed the recombinant E-DI-II and E-DIII from DENV-2 in E. coli and mammalian expression systems. Western blot analysis and IFA showed that, of the 17 mAbs recognizing E protein, 10 mAbs (DB2-3, DB9-1, DB13-19, DB21-6, DB22-4, DB23-3, DB27-3, DB33-3, DB39-2 and DB42-3) targeted to E-DI-II and 2 mAbs (DB25-2 and DB32-6) recognized E-DIII (Figures 1D–1E and Table 1). However, 5 mAbs could not be identified by these two assays.
We evaluated the ability of mAbs to inhibit DENV-2 infection in BHK-21 cells using a plaque reduction neutralization test (PRNT). Ten mAbs had neutralizing activity with 50% PRNT (PRNT50) concentrations ranging from 0.14 µg/ml to 33 µg/ml (Table 1). DB32-6 was found to be a DENV-2 serotype-specific mAb against E-DIII (Figures 1A, 1B and 1E) and was the most efficient at neutralizing DENV-2 infection at a PRNT50 concentration of 0.14 µg/ml (Figure 2A). In addition, it could completely inhibit the infection at a lower concentration of 1.2 µg/ml (Figure 2A). The mAb DB25-2 was found to be a DENV-2 serotype-specific mAb against E-DIII (Figures 1A, 1B and 1E) and to neutralize DENV-2 at a PRNT50 titer of 1.2 µg/ml (Figure 2A). These findings indicate that serotype-specific mAb DB32-6 against E-DIII was the most potent in neutralizing DENV infection. Some serotype-specific mAbs, such as DB2-3 and DB23-3 against E-DI-II and DB25-2 against E-DIII showed strong neutralizing activity. Many complex reactive mAbs showed moderate-to-poor neutralizing activity (Table 1).
Two different mouse models were used to assess whether DB32-6 could efficiently protect mice against DENV-2 challenge. Protection assay of neutralizing mAbs was performed with ICR strain 2-day-old suckling mice [41]. Mice were inoculated intracerebrally with 20 µl of DENV-2-mAb mixture containing 1×104 pfu (25-fold LD50) of DENV-2 with neutralizing mAbs at concentrations of 1, 10 or 100 µg/ml. Generally, the non-neutralizing antibody normal mouse IgG (NMIgG) treated group showed paralysis, ruffling, and slowing of activity around 6 to 9 days. This was followed by severe sickness leading to anorexia, asthenia and death within 9 to 17 days (Figures 2B and 2C). In contrast, mAbs DB32-6 at a concentration of 10 µg/ml protected 93% of the mice from the lethal challenge of DENV-2 (Figure 2B). mAbs 3H5, DB23-3, DB2-3 and DB25-2 had survival rates of 75%, 76%, 72% and 71%, respectively. DB42-3 and DB13-19 had survival rates of 46% and 28%, respectively (Figure 2B). The neutralizing mAbs showed a significant delay of the onset of paralysis and death relative to the NMIgG. To evaluate the therapeutic potential of the highly protective mAb DB32-6, we administered 100 µg/ml or 1 µg/ml to infected suckling mice. The survival rates for DB32-6 at 100 µg/ml or 1 µg/ml were 100% and 89%, respectively (Figure 2C). In comparison, 3H5 showed 82% and 40% survival rates at 100 µg/ml or 1 µg/ml, respectively.
Stat1−/− mice, which lack a transcription factor involved in interferons (IFNs) signaling were sensitive to lethality induced by DENV-2 infection [27], [31]. To test the potential therapeutic effects of the strongest neutralizing mAb DB32-6, we challenged Stat1−/− mice at a strict condition with 1×105 pfu (300-fold LD50) of DENV-2 (NGC-N). After 21 days observation, mice showed ruffled fur, mild paralysis and lost approximate 20% of their initial body weight at day 7 after infection (P<0.01), and then they all died within 7–18 days of infection (Figures 2D and 2E). In the prophylaxis experiments, antibodies (100 µg per mouse, intraperitoneally) were administered 1 day before infection and at day 0, 1, 3, 5 and 7 after infection. The DB32-6 prophylactically treated group showed 100% protection (Figure 2D left). Even in the postexposure therapeutic experiments, the DB32-6 treated mice had a survival rate of 50% (Figure 2E left). The mAb DB32-6 had excellent neutralizing activity against different DENV-2 strains (16681 and NGC-N) in two mouse models.
To further evaluate whether the strongest mAb DB32-6 could broadly neutralize the diverse DENV-2 strains, we infected BHK-21 cells with four different DENV-2 Southeast Asian genotype strains, 16681, NGC, PL046 and Malaysia 07587. Remarkably, mAb DB32-6 exhibited effective neutralization against various DENV-2 strains (Figure S3).
Epitopes recognized by neutralizing antibodies have been identified in all three domains of the E protein [42]–[44]. To find out more about the epitopes of these neutralizing antibodies, we used phage display [29], [45] to identify the neutralizing epitopes. After three rounds of phage display biopanning, the phage titers were increased to 85-fold (DB32-6) and 331-fold (DB25-2) compared to the phage display biopanning results from the first round (Figure 3A). Individual phage clones from the third round of biopanning were randomly selected. ELISA was performed to determine whether the mAbs could specifically recognize selected phage clones. Of 20 selected phage clones, 17 and 18 clones had significant enhancement of binding activity to DB32-6 and DB25-2, respectively (Figure 3B). The selected phage clones PC32-6 and PC25-14 were specific and dose dependently bound to DB32-6 and DB25-2, respectively. They did not react with control NMIgG (Figure 3C).
The 17 immunopositive phage clones that were highly reactive with DB32-6 were amplified and phage DNA was isolated for DNA sequencing. All of the phage clones displayed 12 amino acid (aa) residues (Figure 3D left). Phage-displayed peptide sequences selected by DB32-6 had the consensus motifs of histidine (H)-lysine (K)-glutamic acid (E)-tryptophan (W)/tyrosine (Y)-histidine (H) (Figure 3D left). Similarly, 17 immunopositive phage clones selected by DB32-6 using phage library displayed 7 amino acid residues, which contained the consensus motif H-K-E-W/Y-H (Figure 3D left). Interestingly, all phage-displayed peptides selected by DB32-6 and DB25-2 contained lysine (K) and glutamic acid (E), respectively (Figure 3D).
To further confirm the neutralizing epitopes, we developed various E protein epitope-specific variants VLPs and screened loss-of-binding VLP mutants for identification of critical recognition residues. Using this strategy, we found that DB32-6 lost its VLP binding activity when the residue K310 in the A-strand of E-DIII was changed to alanine (K310A) or glutamine (K310Q) (Figure 4A left). Similarly, DB25-2 lost its VLP binding activity when E311 was changed to arginine (E311R) in the A-strand of E-DIII (Figure 4A right). Both the critical recognition residues K310 and E311 were located in the A-strand of E-DIII (Figures 4B and 4C). We found that mAb 3H5 recognized residues K305, E383 and P384 (Figure S4), as previously reported [17], [20]. Notably, even the adjacent residues (K310 and E311) induced antibodies with different levels of neutralizing activity. By comparing the amino acid sequences of E proteins from representing genotypes of DENV-2 (Table S1), we found residues K310 and E311 in E-DIII of the different genotypes (Southeast Asian, West African and American) (Figure S5). Our data further showed epitopes in the A-strand of E-DIII were important for inducing neutralizing antibodies.
Murine mAbs have been shown to have limited clinical use because of their short serum half-life, inability to trigger human effector functions and the production of human anti-murine antibodies (HAMA) response [46]. mAbs have been humanized by grafting their CDRs onto the VH and VL FRs of human Ig molecules [47]. DB32-6 was the most potent mAb against DENV-2 and showed potential as a therapeutic antibody. To develop humanized mAbs, we sequenced VH and VL segment of the neutralizing mAbs from hybridoma cell lines. The CDRs of DB32-6 were grafted onto human IgG1 backbone to create humanized DB32-6 (hDB32-6) (Figure 5A). The hDB32-6 was expressed in CHO-K1 cells and purified from culture supernatants. Both hDB32-6 and mDB32-6 were able to against DENV-2 (Figure 5B). The hDB32-6 maintained the specificity of murine DB32-6 (mDB32-6). Furthermore, we established stable clones of hDB32-6. After selection, mAbs hDB32-6-30, hDB32-6-48 and hDB32-6-51 were found to have highly binding activity (Figure 5C). Comparing to these mAbs, we found hDB32-6-48 to have the highest production rate in cells. mAb hDB32-6-48 was dose-dependent against DENV-2 and E-DIII (Figure 5D). The affinity was analyzed by surface plasmon resonance. The mDB32-6 and hDB32-6-48 bound to E-DIII of DENV-2 with a similar affinity (0.12 nM and 0.18 nM, respectively) (Figure 5E). The results revealed that hDB32-6 maintained the same binding affinity to the E protein as mDB32-6.
We established a suckling mice model to determine the protective activity of mDB32-6 and hDB32-6. To evaluate therapeutic effect of mAbs, we administered 5 µg of mAb at day one after 1×104 pfu (25-fold LD50) of DENV-2 (16681) infection. Through 21 days of observation, groups treated with mDB32-6, hDB32-6-48 and 3H5 mAbs were found to have survival rates of 96%, 94% and 56%, respectively (Figure 6). However, none of the mice in control antibody normal human IgG (NHIgG)-treated group survived (Figures 6). These results demonstrate that both mDB32-6 and hDB32-6 have excellent neutralizing activity against DENV-2.
When developing the antibody-based therapy, ADE phenomenon is a major cause for concern in dengue pathogenesis because it might enhance DENV infection. Modification of Fc structure in an antibody can prevent Fcγ receptors binding and lead to eliminate ADE [38], [39], [48]. We generated a variant of humanized DB32-6 (hDB32-6 variant) to prevent Fcγ receptors binding while maintaining DENV neutralizing capability without enhancing infection (Figure 7). The hDB32-6 variant retained the same neutralizing activity as unmodified mAb mDB32-6 at high concentrations (100 µg/ml and 10 µg/ml) but was completely devoid of enhancing activity at low concentrations (1 µg/ml and 0.1 µg/ml) (Figure 7). The hDB32-6 variant eliminated the ADE phenomenon and holds great potential for being developed into therapeutic antibodies for the prevention and treatment of DENV-2 infection.
mAbs of DENV have served as powerful research tools for antiviral development and pathological investigations. Here, we newly generated and characterized 17 mAbs with high reactivity against E protein of DENV-2. Several mAbs had potent neutralizing activity. The neutralizing epitopes were identified using a combination of strategies, including phage display, computational structure analysis [49], and high-throughput epitope mapping of VLPs. From these results, the A-strand of E-DIII was found to be important in neutralizing DENV-2 than the lateral ridge of E-DIII. mAb DB32-6 which had the strongest neutralizing activity against various strains of DENV-2 was humanized and modified to abrogate the ADE phenomenon. The mAb DB32-6 was demonstrated to increase the survival rate in two mouse models even after DENV-2 infection.
Based on previous epitope mapping results, several epitopes have been shown to elicit strong neutralizing antibodies against individual flaviviruses that situated in E-DIII [14], [50]. Investigation of neutralizing epitopes on the E proteins may provide the framework for a detailed understanding of both specific mechanisms of the viral infection as well as the identification of the specific DENV domain that attaches to a cellular receptor. Phage display is useful in the identification of B-cell epitopes, including linear [32], [51] and conformational epitopes [29], [45]. However, these epitopes need to further elucidation using other methods. Combining different strategies provided a fast and reliable evidence for identifying epitopes (Figures 3 and 4). To date, few mAbs possess better neutralizing activity than 3H5, which has been shown to bind to residues K305, E383 and P384 at the lateral ridge of E-DIII [17], [20]. DB32-6 had higher neutralizing activity than 3H5. Neutralizing epitope of DB32-6 was mapped on K310 residue in A-strand of E-DIII (Figure 4). Neutralizing epitope of another mAb DB25-2 was mapped on E311 residue in A-strand of E-DIII, too (Figure 4). These serotype-specific neutralizing epitopes located in the A-strand of E-DIII induced stronger neutralizing activity than those located on the lateral ridge of E-DIII. We aligned different DENV-2 genotypes and found that the K310 and E311 were frequently observed in DENV-2 (Figure S5). The K310 may be important to DENV-2. Thus by binding DB32-6 to K310, it lead to dramatic neutralized DENV-2. To determine whether DB32-6 can neutralize diverse genotypes of DENV-2 is a critical step in evaluating the potential of therapeutic development in the future.
Previous studies have shown that the strongly neutralizing mAb, subcomplex-specific 1A1D-2 and cross-reactive 9F12 recognized residues at K305, K307 and K310 in A-strand [15], [17]. Our mAb DB32-6 is a serotype-specific neutralizing mAb that recognized residue K310 but not residues K305 or K307. Although K310 is considered as a subcomplex-specific epitope, DB32-6 is a serotype-specific mAb. There may be other regions that affect the binding of DB32-6 to DENV-2. We found that by mutating residue I312, DB32-6's binding activity was reduced by 50% (data not shown). Residue I312 may be a minor epitope of DB32-6. Moreover, 1A1D-2 is a temperature dependent mAb due to its needs for dynamic motion on the virion surface to neutralize virus [16]. Different from 1A1D-2, DB32-6 is temperature independent. When DB32-6 was incubated with DENV at 4°C, it still exhibited significant neutralizing activity (Figures 2 and 6). As expected, when incubating the DENV and DB32-6 at 37°C, DB32-6 showed better efficacy than it did at 4°C (data not shown). The residue K310 on the surface of DENV-2 may be accessible to DB32-6 binding. Additionally, DB32-6 had high binding affinity (0.12–0.18 nM) to DENV-2. Based on the above finding, the residue K310 induce serotype-specific mAbs and is crucial in the neutralization of virus infectivity.
Antibodies to E-DI-II tend to be more cross-reactive and less potent in neutralization of dengue infection [39]. However, there are fewer antibody concentrations capable of recognizing E-DIII than there are that recognize E-DI-II in dengue patients [20], [39]. Wahala et al. studied the human immune sera of DENV infection and found the E-DIII binding antibodies to play a minor role in DENV neutralization, similar to West Nile virus-infected human [52], [53]. The mAbs that bind to E-DIII expresses potent neutralizing activity, but only a few of them exist in serum of the patients infected with DENV or WNV. Combining the information from both mice and human mAbs studies of DENV infection is critical to understanding the complex mechanism behind the humoral immunity following natural DENV infection. According to one previous study, the immunoglobulin populations recognizing residues K310, E311 and P364 in dengue fever patients were much larger in IgM than in IgG [20]. The strong neutralizing IgG made up a small proportion of the antibody in dengue patients. de Alwis et al. has conducted an in-depth analysis of the human mAbs derived from memory B-cells of patients infected with primary DENV infections [54]. After the epitope mapping of anti-DENV-2 human mAbs, the strong neutralizing mAb 10.16 was mapped to K305, K310 and E311 in the A-strand. Together, the finding above suggest that the highly protective epitopes K310 and E311 in mouse play a role in humans as well.
We also identified several E-DI-II specific mAbs with high to no neutralizing activity. Serotype-specific mAbs (DB2-3 and DB23-3) with potent neutralizing activity were found to recognize E-DI-II of DENV-2 (Figure 2 and Table 1). Some studies have identified highly neutralizing and protective antibodies against JEV and DENV located in E-DI [55], [56] Currently, we are in the process of identifying the neutralizing epitopes of DB2-3 and DB23-3. mAbs that broadly cross-react with other flaviviruses are in E-DII near the fusion loop, which is immunodominant antigenic [20], [34], [42]. Binding an antibody to DENV can change the rearrangement of the E protein, which may neutralize or enhance viral infection [16], [57]. The high or no neutralizing activity of our mAbs can be used help identify neutralizing or immunopathogenic epitopes in the E protein. Studies that explore the mAbs mediated neutralization mechanism and mAbs dependent enhancement are currently underway.
The mouse models for dengue infection developed to date do not represented the entirety of the pathogenesis of human dengue infection [58]. Developing of mouse models to studying its pathogenesis is important but challenging. We used two models, suckling mice protection assay and Stat1-deficient (Stat1−/−) mouse model with different DENV-2 strains through intracerebral or intraperitoneal inoculation to evaluate the neutralizing activity of DB32-6 mAb (Figures 2 and 6). Our findings suggested that mAb DB32-6 might effectively block virus entry. However, disease manifestation of suckling mice is not relevant to dengue disease in humans since DENV infections in humans rarely involve the nervous system [58]. The Stat1-deficient mice are genetically mutated and not immunocompetent, hence they are not representative of the wild types' immune response to DENV. However, their survival rates might reflect the therapeutic potential of these mAbs. The results from these mouse models showed that the therapeutic potential of this newly generated mAb DB32-6 is worth further investigation.
In the absence of an effective dengue vaccine, neutralizing antibodies can be used as a passive immunotherapeutic strategy for treating dengue. Previous studies of humanized antibodies against DENV were derived from two chimpanzee Fab fragments: humanized IgG1 1A5 cross-neutralizing DENV-1 and DENV2 and humanized IgG1 5H2 specific against DENV-4 [42], [48], [56]. Our newly generated hDB32-6 was derived from murine mAb. However, when developing antibody-based therapy, ADE phenomenon is a major concern. Modification of Fc structure in an antibody can prevent Fcγ receptors binding and inhibit ADE (Figure 7) [39], [48].
Our studies show that the serotype-specific mAbs targeting the A-strand of E-DIII could serve as a dramatic neutralization determinant. Through testing in different mouse models, we have successfully generated a mAb hDB32-6 variant with high therapeutic potential against diverse DENV-2 strains without inducing ADE. Such an antibody-based therapy may help control severe dengue in the future.
|
10.1371/journal.pgen.1000774 | A Major Role of the RecFOR Pathway in DNA Double-Strand-Break Repair through ESDSA in Deinococcus radiodurans | In Deinococcus radiodurans, the extreme resistance to DNA–shattering treatments such as ionizing radiation or desiccation is correlated with its ability to reconstruct a functional genome from hundreds of chromosomal fragments. The rapid reconstitution of an intact genome is thought to occur through an extended synthesis-dependent strand annealing process (ESDSA) followed by DNA recombination. Here, we investigated the role of key components of the RecF pathway in ESDSA in this organism naturally devoid of RecB and RecC proteins. We demonstrate that inactivation of RecJ exonuclease results in cell lethality, indicating that this protein plays a key role in genome maintenance. Cells devoid of RecF, RecO, or RecR proteins also display greatly impaired growth and an important lethal sectoring as bacteria devoid of RecA protein. Other aspects of the phenotype of recFOR knock-out mutants paralleled that of a ΔrecA mutant: ΔrecFOR mutants are extremely radiosensitive and show a slow assembly of radiation-induced chromosomal fragments, not accompanied by DNA synthesis, and reduced DNA degradation. Cells devoid of RecQ, the major helicase implicated in repair through the RecF pathway in E. coli, are resistant to γ-irradiation and have a wild-type DNA repair capacity as also shown for cells devoid of the RecD helicase; in contrast, ΔuvrD mutants show a markedly decreased radioresistance, an increased latent period in the kinetics of DNA double-strand-break repair, and a slow rate of fragment assembly correlated with a slow rate of DNA synthesis. Combining RecQ or RecD deficiency with UvrD deficiency did not significantly accentuate the phenotype of ΔuvrD mutants. In conclusion, RecFOR proteins are essential for DNA double-strand-break repair through ESDSA whereas RecJ protein is essential for cell viability and UvrD helicase might be involved in the processing of double stranded DNA ends and/or in the DNA synthesis step of ESDSA.
| Deinococcus radiodurans bacterium is among the best-known organisms found to resist extremely high exposures to desiccation and ionizing radiation, both causing extensive DNA double-strand breaks. Because a single unrepaired DNA double-strand break is usually lethal, DNA double-strand breaks are considered as the most severe form of genomic damage. The extreme radioresistance of D. radiodurans is linked to its ability to reconstruct a functional genome from hundreds of chromosomal fragments. Genome reconstitution occurs through a two step process: (i) an extended synthesis-dependent strand-annealing process (ESDSA) that assembles genomic fragments in long linear intermediates that are then (ii) processed through recombination to generate circular chromosomes. Here, we demonstrate the essential role of key components of the D. radiodurans RecF pathway in ESDSA. We show that (i) inactivation of only one exonuclease (RecJ) results in cell lethality; (ii) cells devoid of RecF, RecO, or RecR display greatly impaired growth; (iii) RecF, RecO, or RecR proteins are essential for radioresistance through ESDSA; and (iv) UvrD helicase has an unexpected crucial function in DNA double-strand-break repair through ESDSA.
| The bacterium Deinococcus radiodurans is extremely resistant to treatments such as ionizing radiation and desiccation. This resistance can be correlated with the ability of D. radiodurans to reconstruct a functional genome from hundreds of radiation or dessication-induced chromosomal fragments, while the genomes of most organisms are irreversibly shattered under the same conditions. The rapid reconstitution of an intact genome is dependent on extended synthesis-dependent strand annealing (ESDSA) and recombination [1],[2]. It was proposed that, following severe DNA damage, the fragmented DNA end is recessed in a 5′–3′ direction, liberating single stranded 3′ overhangs which, through RecA- and RadA-mediated strand invasion, prime DNA synthesis on overlapping fragments [2]. DNA synthesis is initiated by Pol III and elongated by Pol I or by Pol III and the newly synthesized single-strands anneal to complementary single stranded extensions forming long double stranded DNA intermediates which are assembled into intact circular chromosomes by RecA-mediated homologous recombination [2]. Though the dependence of ESDSA on RecA, Pol I, and Pol III activities is well documented [1],[2], little is known about the cellular factors required for the first steps of this process (i.e. the formation of the single stranded 3′ overhangs which promote RecA/RadA - dependent strand invasion to prime DNA synthesis).
Three enzymatic activities are required for presynaptic processing of double stranded DNA ends in the model bacterium Escherichia coli: a helicase, a 5′-3′exonuclease, and a mediator function for efficient RecA filament formation onto ssDNA (see for reviews [3]–[5]). All these activities are carried out by the RecBCD complex (or its functional homolog AddAB) which is the major component for initiation of recombinational repair of DNA double-strand-breaks (DSB) in wild-type cells. However, if RecBCD is inactivated, an alternate pathway, the RecF pathway, promotes recombinational DSB repair [6]–[10] in cells containing mutations in sbcB (suppressor of recBC), which encodes the 3′-5′ exonuclease I, and in sbcC (or sbcD) [11]. This pathway comprises the 5′-3′ single-strand DNA exonuclease RecJ, the RecQ helicase and the RecF, RecO and RecR proteins that act together to promote loading of RecA onto single stranded DNA.
Whereas examination of the phylogenetic distribution of RecBCD and AddAB complexes revealed that one or the other complex is present in most sequenced bacteria, D. radiodurans is naturally devoid of these two complexes but does encode a RecD homologue [12]. RecD protein was shown to be present in the absence of RecBC not only in D. radiodurans, but also in firmicutes and Streptomyces [13]. The deinococcal RecD protein is expressed and active as a DNA helicase [14]. Further work is required to assign RecD protein to a specific DNA repair pathway because conflicting data have been published concerning the in vivo role of RecD in radioresistance [15]–[16]. D. radiodurans possesses homologs of the key components of the RecF pathway: RecJ (DR1126), RecQ (DR1289), RecF (DR1089), RecO (DR0819), and RecR (DR0198) suggesting that the RecF pathway is the main recombinational repair pathway in this organism, as observed in other bacteria that lack RecBCD homologs [13]. D. radiodurans also lacks homologs of the SbcB nuclease, an inhibitor of the RecF pathway in E. coli. Moreover, it was shown that expression in trans of the SbcB protein from E. coli renders D. radiodurans cells radiation-sensitive [17].
In this paper, we investigate the role of the D. radiodurans proteins belonging to the RecF pathway in ESDSA and/or homologous recombination. We demonstrate that RecJ exonuclease is an essential protein for cell viability. We show that the RecF, RecO, RecR proteins as well as the RecA protein are absolutely required for massive DNA synthesis during DSB repair whereas RecQ appears to be substituted by the UvrD helicase to play a role in this process. We propose that RecJ, in conjunction with UvrD, could generate the single stranded tails on which RecFOR will stimulate RecA loading. Interestingly, an intact genome could be slowly reconstituted in the absence of RecA, RecF, RecO or RecR, suggesting alternate DSB repair through non-homologous end joining (NHEJ) and/or single-strand annealing (SSA).
To determine the importance of the RecFOR pathway in DSB repair and radioresistance in D. radiodurans, we replaced the coding regions of key genes belonging to this pathway (recJ, recQ, recF, recO, and recR) with an antibiotic resistance cassette. The deletion-substitution alleles were constructed in vitro using the tripartite ligation method [18] and introduced by transformation into D. radiodurans to replace the corresponding wild-type alleles via homologous recombination. Because D. radiodurans contains from 4 to 10 genome equivalents [19],[20], the transformants were extensively purified on selective media in order to obtain the mutant homogenotes whose purity was verified by PCR. Whereas only few rounds of purification on selective antibiotic plates sufficed to obtain ΔrecQ, ΔrecF, ΔrecO and ΔrecR homogenotes (see Figure S1), in the case of recJ, the wild-type allele was present together with the ΔrecJ allele even after seven rounds of purification of three independent candidates (Figure 1), suggesting that RecJ protein is essential for cell viability.
To obtain positive evidence for the essentiality of the recJ gene, we used the new diagnostic assay described by Nguyen et al [21]. For this purpose, the recJ gene was cloned onto a prepUTs vector thermosensitive for replication in D. radiodurans [21]. The sequence of DR1126 (recJ) in strain ATCC 13939 (GenBank, accession number QG856645) was found to differ from the DR1126 published sequence [22]. An additional G was found 7 nucleotides upstream the published putative GTG initiation codon and another additional G was found 58 nucleotides before the published putative TGA STOP codon giving rise to a RecJ protein containing 705 aa (versus 684 aa in the RecJ protein predicted from the previously published sequence) with 64 additional amino acids in the N-terminal domain and 43 aa missing in the C-terminal domain of the protein. The predicted sequence of the RecJ protein in strain ATCC 13939 displays a better alignment with the published protein sequences of the E. coli, Deinococcus geothermalis and Thermus thermophilus RecJ proteins (Figure S2). The recombinant plasmid was introduced into a recJ+ recipient and the chromosomal copy of recJ in the resulting merodiploid strain was inactivated (Figure 1). If recJ is an essential gene, the cells will die upon loss of the complementing plasmid at the non permissive temperature. As can be observed in Figure 2 (lanes 1–3), the ΔrecJ (prepUTs-recJ+) bacteria grew normally at 28° (the permissive temperature for the plasmid) but lose viability at 37° (the non-permissive temperature for the plasmid), demonstrating the essentiality of the recJ gene.
The ΔrecF, ΔrecO, and ΔrecR mutants, though viable, showed a greatly impaired growth. Indeed, the mutants had a generation time 4-fold longer than the wild-type (5 hours for the mutants versus 80 min for the wild-type) and comparable to that of a ΔrecA mutant. Furthermore, cells devoid of RecF, RecO or RecA proteins had a 10-fold reduced plating efficiency as compared to the wild-type strain and this defect was even more pronounced in the ΔrecR mutant, displaying a 30-fold reduced plating efficiency (Figure 3).
In E. coli, the RecQ helicase initiates DSB repair via the RecFOR pathway by unwinding duplex DNA in the 3′-5′ direction, while the single stranded DNA exonuclease RecJ hydrolyzes the 5′ strand to provide a DNA-substrate for RecA loading onto the 3′ strand [4],[23].
We found that inactivation of the RecQ helicase in D. radiodurans had no effect on radioresistance, because the knockout mutant was as resistant to γ-irradiation as the wild-type strain (Figure 4A). This result suggests that other helicase(s) might be involved in the initiation step of DSB repair in this organism. We tested the RecD and UvrD helicases for putative roles in DSB repair. We found that a ΔrecD deletion mutant was as radioresistant as the wild-type strain, whereas a ΔuvrD mutant showed a reduction in survival that ranged from 5-fold at 11.6 kGy to more than 100-fold at 17.8 kGy (Figure 4A). However, the mutant still retained significant radioresistance as compared to a repair-deficient ΔrecA strain (Figure 4B), suggesting that other helicase(s) may overlap in function with UvrD and thus lessen the effect of a ΔuvrD mutation. To test this hypothesis, we investigated whether the combined absence of UvrD and RecQ or UvrD and RecD proteins results in a more dramatic effect on radio-resistance. As seen in Figure 4A, the ΔuvrD ΔrecQ double mutant bacteria were not more sensitive to γ-rays than a ΔuvrD single mutant. In contrast the ΔuvrD ΔrecD double mutant bacteria were slightly more sensitive to γ-rays than a ΔuvrD single mutant, suggesting that the RecD helicase may have a partial back-up function in the absence of UvrD.
To investigate the possible role(s) of the UvrD helicase in DSB repair, we examined whether the ΔuvrD mutant was affected in two key steps of the ESDSA pathway: (i) the reassembly of broken DNA fragments and (ii) the associated massive DNA synthesis. Cells were exposed to 6.8 kGy γ-irradiation, a dose that introduces approximately 200 DSB per genome equivalent in a D. radiodurans cell [24]. Recovery from DNA damage was monitored by the appearance of the complete pattern of the 11 resolvable genomic DNA fragments generated by NotI digestion [25] and de novo DNA synthesis was measured by labelling DNA with a 15 min 3H-TdR pulse at different times post irradiation. As seen in Figure 4, ΔrecQ and ΔrecD cells repaired DSB with the same kinetics as the wild-type strain, reconstituting an intact genome within 3 h post-irradiation (Figure 5A). In contrast, in ΔuvrD bacteria, this process required approximately 8 h (Figure 5A), the kinetics of DSB repair had an increased latent phase (240 min in the mutant versus 90 min in the wild-type) during which DNA degradation took place and a slower rate of fragment assembly. Moreover, resumption of DNA synthesis was delayed in ΔuvrD mutant bacteria and its rate was 2-fold lower than that observed in wild-type bacteria (Figure 5B). These results suggest that UvrD plays a major role in DSB repair through ESDSA.
The ΔrecF, ΔrecO and ΔrecR mutants were as radiosensitive as a ΔrecA mutant (Figure 4B). The radiosensitivity of the ΔrecF and ΔrecO mutants was fully complemented by a plasmid expressing RecF or RecO proteins in trans, whereas, in the case of the ΔrecR mutant, bacteria expressing recR+ in trans only recovered 90% of wild-type survival after γ-irradiation (Figure 4B). Because recR belongs to a putative operon, the radiosensitivity of the knock-out mutant may be due in part to a polar effect of our construct on a downstream gene or to a sub- or overoptimal plasmid-based expression of the RecR protein.
The kinetics of DNA double-strand-break repair in the three mutants was very similar to that observed in a ΔrecA mutant (Figure 6A). There was a slight and progressive reassembly of the radiation-induced DNA fragments that culminates at 24h post-irradiation incubation in the restitution of a complete pattern of the 11 NotI resolvable fragments (Figure 6A). However, only very faint bands of reconstituted chromosome were observed 24h post-irradiation incubation suggesting that a complete genome was only present in a small subpopulation of the mutant cells. The initial degradation of the damaged DNA that can be seen in the wild-type during the first hour of post-irradiation incubation (Figure 5A) was also markedly reduced in the three recFOR mutants (Figure 6A), as was previously observed for a ΔrecA mutant [2]; Figure 6A).
The reconstitution of the complete genomic NotI pattern in the irradiated recFOR mutants did not result from the multiplication of rare survivors, because there was no observable increase in the number of CFU during 24 hours of incubation of irradiated cells (data not shown). Pulses of 3H-TdR showed that no DNA synthesis was observed during the 6 hours following γ-irradiation (Figure 6B) nor during the late fragment assembly in ΔrecF, ΔrecO and ΔrecR bacteria (data not shown), as observed previously in ΔrecA bacteria [2]. Moreover, the late genome reconstitution in these mutants is not sufficient to ensure cell survival.
In conclusion, our results suggest that RecF, RecO and RecR proteins, like RecA protein, play a central role in Deinococcal radioresistance, probably because they are absolutely required for loading RecA onto its DNA substrate to perform efficient double-strand-break repair via ESDSA and recombinational repair pathways.
Recent studies have shown that the ability of D. radiodurans to cope with the DNA-shattering effect of elevated doses of γ-irradiation or dessication involves a robust DNA repair process called extended synthesis-dependent strand annealing (ESDSA) in which long tracts of newly synthesized DNA are made [1],[2]. Whereas the dependence of massive DNA synthesis on Pol I, Pol III and RecA (or its homolog RadA) in ESDSA is well documented [2], little is known about the cellular factors required for the initial steps of this process: (i) the formation of the single stranded 3′ overhangs and (ii) the loading of RecA on this recombinogenic substrate to prime DNA synthesis.
D. radiodurans is naturally devoid of RecB and RecC proteins but contains homologs of key proteins of the E. coli RecF pathway: RecJ, RecQ, RecF, RecO and RecR. We found that cells devoid of RecF, RecO or RecR proteins were as radiosensitive as cells devoid of RecA. The ΔrecF, ΔrecO and ΔrecR mutants, as previously shown for a ΔrecA mutant [2], supported a slow and progressive reassembly of the radiation-induced DNA fragments. As in ΔrecA cells, genome reassembly was not accompanied by significant DNA synthesis, suggesting that cells devoid of RecF, RecO or RecR proteins are deficient for ESDSA, with repair of DSB probably mediated by RecA-independent pathways, such as single-strand annealing (SSA) or non-homologous end joining (NHEJ). The mutants also showed an important lethal sectoring during normal growth, similar to that observed in a ΔrecA mutant, in which about 90% of the visible cells failed to give rise to colonies [26]. The similarity of the ΔrecFOR and ΔrecA phenotypes supports the hypothesis that RecA activity in D. radiodurans is totally dependent on a functional RecF pathway.
Following exposure of D. radiodurans to ionizing radiation, there is a rapid and extensive degradation of chromosomal DNA that plays an important role in the repair process in this species (reviewed by [24]). The initial degradation of damaged DNA can be observed using pulsed-field electrophoresis as a reduction of the amount of the double stranded DNA fragments during the first 90 min of post-irradiation incubation in the wild-type cells, prior to the onset of fragment assembly. Slade et al observed that DNA degradation is markedly reduced in a ΔrecA mutant, leading the authors to propose that RecA itself regulates maturation of double-strand ends by controlling both DNA degradation and DNA synthesis [2]. We found that DNA degradation was also reduced in ΔrecF, ΔrecO or ΔrecR mutants, as well as in a ΔrecA mutant. RecA may play a regulatory role in the control of expression of nuclease-like activities in response to DNA damage, while RecFOR proteins may be indirectly involved in DNA degradation by facilitating the formation of the RecA filament on single stranded DNA. It would be interesting to analyse DNA degradation in the Deinococcal recA424 mutant, which retains the RecA coprotease activity while remaining deficient in recombination activity [27].
Biochemical studies using RecFOR proteins from E. coli indicate that these proteins act together as mediators of the formation of the pre-synaptic RecA filament onto single stranded DNA. Current models agree on the formation of two complexes, RecFR and RecOR. RecOR is generally thought to be responsible for rendering SSB-coated ssDNA accessible to RecA. RecFR targets dsDNA or dsDNA-ssDNA junctions and is responsible for the targeting of RecA to the ssDNA region of gaps [28]–[31]. More recently, it was proposed that RecR is the key component with which RecA interacts, whereas the RecO protein can displace SSB and bind to single stranded DNA independently of RecR, yet does not load RecA until RecR is added [32],[33]. When RecF is present, a RecFOR loading pathway, independent of RecO-SSB interactions, is preferred [33].
Recent X-ray structural analysis of RecO and RecR proteins from D. radiodurans confirms the existence of a RecOR complex in this organism. RecR molecules form a ring structure that can encircle both dsDNA and ssDNA [34],[35]. The structure of the RecF protein from D. radiodurans has also recently been elucidated, showing that the RecF protein exhibits extensive structural similarity with the head domain of the eukaryotic Rad50 protein [36]. More recently, a model of recognition of the ds-DNA ss-DNA junction in D. radiodurans through a DNA-protein and protein-protein interaction was proposed: RecR interacts with ssDNA coated by RecO-SSB, which leads to the elevation of the local concentration of RecR and stimulates RecF binding in the adjacent ds-DNA [37].
While inactivation of RecA or RecFOR proteins in D. radiodurans reduced cell viability, inactivation of RecJ resulted in a fully lethal phenotype. In other bacterial species, mutations in recJ are highly synergistic with those in recBCD. In E. coli, recBC recJ mutants are recombination deficient, extremely UV-sensitive and highly growth disrupted [11],[38]. In Salmonella typhimurium, recB recJ mutants also display a similar phenotype [39]. More recently, it was shown that a recJ knock-out is colethal with recBCD or recD deletions in Acinetobacter baylyi [40]. The strongly reduced viability (or lethality) of recBC recJ bacteria was attributed to severe deficiencies in repair of spontaneous DNA damage and inactivated replication forks [39],[40]. It should be noted that, whereas E. coli and S. typhimurium contain at least three 5′-3′ exonucleases [RecJ, Exo V (RecBCD), Exo VII (XseAB)], the genome of A. baylyi encodes only Exo V and RecJ, and that of D. radiodurans encodes only RecJ and one of the two subunits of Exo VII. We propose that Exo VII has some back-up activity in E. coli or S. typhimurium when RecJ and ExoV are inactivated, an activity that is missing in A. baylyi and D. radiodurans.
In E. coli, RecJ and RecFOR were proposed to be required to restore DNA synthesis after UV-induced damage [41],[42]. The mechanism by which lesion-blocked replication forks recover in E. coli is thought to involve the formation of reverse replication fork intermediate stabilized by RecA and RecF and degraded by the RecQ-RecJ helicase-nuclease when RecA or RecF are absent [41]. The fork regression allows DNA repair enzymes to remove the blocking lesion, thus restoring processive replication. In the absence of RecJ, the recovery of replication is significantly delayed and both replication recovery and cell survival become dependent on translesion synthesis by DNA polymerase V [42]. D. radiodurans does not encode a bypass DNA polymerase belonging to the Y family, and under these conditions RecJ may be essential for restoration of replication forks after arrest, even in cells not treated by DNA damaging agents. Frequent DNA double-strand-breaks were thought to arise spontaneously ranging from 0.2–1 per genome replication in E. coli [4],[43]. However, a more direct quantification of DNA double-strand-breaks indicated that the rate of spontaneous breakage is 20 to 100-fold lower than predicted, only one percent of the cells having one or more DNA double-strand-breaks per genome replication [44].
Because cells devoid of RecA or RecFOR are viable, the ΔrecJ lethal phenotype cannot be only due to a possible deficiency in DSB repair, leading us to postulate that RecJ is required in D. radiodurans for more than one cellular process and that inactivation of all of these processes (DSB repair, fork reversion, restoration of a fork structure after regression …) may be lethal for the cell.
In E. coli, the RecJ exonuclease has been mainly associated with the RecQ helicase in recombination and repair (see, for review, [4]). The RecQ protein from D. radiodurans shows unusual domain architecture with three tandem HRDC (Helicase RNase D C-terminal) domains in addition to the conserved helicase and RQC (RecQ C-terminal) domains. The tandem arrangement of HRDC domains regulates the specificity of the binding of RecQ to DNA substrates [45],[46]. Here, we found that ΔrecQ mutants displayed a wild-type level of resistance to γ-irradiation, exhibiting the same kinetics as the wild-type strain for fragment reassembly and DNA synthesis after irradiation. In another report, a recQ knock-out mutant was shown to be highly sensitive to H2O2 and slightly more sensitive than the wild-type strain to elevated γ-irradiation doses [46]. It was recently proposed that recQ deletion, by causing transcriptome alteration, would generate ROS accumulation and Fe and Mn alterations [47]. Our findings suggest that the RecQ helicase in D. radiodurans plays only a minor role in DSB repair, probably as consequence of redundant functions provided by other helicase(s). Mutants devoid of RecD behave like ΔrecQ mutants in that they show wild-type radioresistance and repair capacity. The Deinococcal RecD protein has been characterized in vitro as a helicase with 5′-3′ polarity (opposite to that of RecQ) and low processivity [14].
In contrast, we found that inactivation of UvrD (helicase II) markedly reduced Deinococcal radioresistance and severely delayed the kinetics of DSB repair. UvrD has been largely characterized for its role in nucleotide excision repair (NER) and mismatch repair (MMR) in E. coli (reviewed by [48]). However, the altered kinetics of repair and the radiosensitivity of ΔuvrD bacteria are unlikely to result from a deficiency in the NER pathway because uvrA deficient mutant bacteria display a wild-type survival pattern following exposure to ionizing-radiation (Figure S3). The ΔmutS bacteria deficient for MMR were also shown to be as radioresistant as wild-type bacteria [18]. Interestingly, the delayed kinetics of DSB repair in cells devoid of UvrD coincided with DNA synthesis (albeit significantly less extensive than that observed in the wild-type cells) suggesting that ESDSA repair could take place but only inefficiently in this mutant. We propose that UvrD is involved in ESDSA and that the redundant activity of other helicase(s) is responsible for the residual DNA repair capacity observed in the ΔuvrD mutant. The fact that the ΔrecQΔuvrD and the ΔrecDΔuvrD double mutant bacteria were not as radiosensitive as ΔrecA bacteria suggests that neither RecQ nor RecD can solely fulfil this role, and that other helicase(s) may be involved. Helicase IV (HelD) has been implicated as partner of the RecJ exonuclease in the RecF pathway in E. coli, together with Helicase II and RecQ [49]. Mutational inactivation of Helicase IV has no effect on the radioresistance of D. radiodurans [50]. Alternatively, RecA itself, by binding to double stranded DNA ends, could unwind DNA and provide a DNA substrate for RecJ or another 5′-3′ exonuclease. Indeed, in vitro, the D. radiodurans RecA protein binds preferentially to double stranded DNA [51].
In E. coli, the UvrD protein was not shown to be required for DNA double-strand-break repair. In contrast, it was shown to possess an anti-recombination activity, which has been related to its capacity to disrupt the RecA nucleoprotein filament [52],[53]. This activity is conserved among many species [54]. Thus, as in other species, the D. radiodurans UvrD protein might not be involved directly in the maturation of DNA double-strand ends. Several observations suggest that E. coli UvrD may be involved in DNA replication [55]–[58] and it was shown to be required for DNA replication of several different rolling-circle plasmids in E. coli [59]. Thus, the D. radiodurans UvrD protein might also act in the DNA synthesis step of ESDSA.
Taking into account our results and those of others [2],[60],[61], we propose a model for the role of the proteins of the RecF pathway in ESDSA (Figure 7). In this model, RecJ or an as-yet unidentified exonuclease associated with the UvrD helicase, could generate 3′ single stranded DNA ends required for priming of massive DNA synthesis. Alternatively, RecA itself, by binding to double stranded DNA ends, could unwind DNA and provide a DNA substrate for RecJ or another exonuclease.
Analysis of the transcriptome of D. radiodurans revealed a large group of genes that are up-regulated in response to either desiccation or ionizing radiation [62]. The deinococcal specific ddrA (DR0423) and ddrB (DR0070) genes were found among the most highly induced in response to each stress and their inactivation promotes sensitization of the mutant cells to ionizing radiation [62]. The DdrA protein is involved in protection of 3′ DNA single stranded ends [60] and presumably ensures long-lived recombinational substrates [63]. The DdrB protein binds single stranded DNA but not duplex DNA and is the prototype of a new bacterial SSB family [61]. The induction of an alternative SSB following irradiation has potentially broad significance for efficient genome reconstitution. We propose that during initiation of ESDSA, DdrA protects the 3′ DNA ends whereas SSB or the SSB-like DdrB binds to single stranded DNA.
Our results supporting the idea that RecA activity in D. radiodurans is totally dependant on a functional RecF pathway, lead us to propose that RecFOR renders SSB or DdrB- coated single stranded DNA accessible to RecA and favors formation of a RecA nucleoprotein filament required for invasion of a double stranded homologous DNA. Finally, as described previously [2], Pol III and Pol I can promote DNA synthesis, eventually with the help of the UvrD helicase.
Moreover, the compact D. radiodurans nucleoid structure that remains unaltered after high-dose γ-irradiation may passively contribute to radioresistance by preventing the dispersion of free DNA ends [64],[65]. Such a condensed genome may provide suitable scaffolds for DNA repair through ESDSA, recombinational and/or DNA end joining processes.
In conclusion, we demonstrate the essential role of key components of the D. radiodurans RecF pathway in ESDSA. We show for the first time that (i) inactivation of only one exonuclease, RecJ, results in cell lethality (ii) cells devoid of RecF, RecO or RecR display greatly impaired growth (iii) RecF, RecO or RecR proteins are essential for radioresistance through ESDSA (iv) UvrD helicase has an unexpected crucial function in DNA double-strand-break repair through ESDSA.
Bacterial strains and plasmids are listed in Table 1 and Table 2, respectively. The Escherichia coli strain DH5α was used as the general cloning host and strain SCS110 was used to propagate plasmids prior to introduction into D. radiodurans via transformation [66]. All D. radiodurans strains were derivatives of strain R1 (ATCC 13939). D. radiodurans was grown in TGY2X (1% tryptone, 0.2% dextrose, 0.6% yeast extract) or in TGYA (0.5% tryptone, 0.2% dextrose, 0.15% yeast extract) at 30°C with aeration or on TGY1X plates solidified with 1.5% agar. E. coli strains were grown in Luria-Bertani (LB) broth (Gibco Laboratories). When necessary, media were supplemented with the appropriate antibiotics used at the following final concentrations: chloramphenicol 3 µg/mL for D. radiodurans; kanamycin 6 µg/mL for D. radiodurans; tetracycline 2.5 µg/mL for D. radiodurans; hygromycin 50 µg/mL; spectinomycin 40 µg/mL for E. coli and 75 µg/mL for D. radiodurans. Transformation of D. radiodurans with PCR products, genomic DNA, or plasmids was performed as previously described [26].
Plasmid DNA was extracted from E. coli using the QIAprep spin miniprep kit (Qiagen). Chromosomal DNA of D. radiodurans was isolated as previously described [67]. Amplification of plasmid or genomic DNA by PCR was performed with DyNAzyme EXT DNA polymerase (Finnzyme) or Extensor Hi-Fidelity PCR enzyme Mix (ABgene). Oligonucleotides used are listed in Table S1.
The recF, recO, recR, recA, uvrD, recD, recQ, recJ disruption mutants were constructed by the tripartite ligation method [18]. The mutated alleles constructed in vitro were then used to transform D. radiodurans to replace their wild-type counterpart by homologous recombination. The genetic structure and the purity of the mutants were checked by PCR using primers described in Table S1.
Plasmid p11869 is a derivative of the thermosensitive plasmid p13840 [21]. To construct p11869, the recJ gene was amplified by PCR using the primer pair (PS441/PS442) and the product was cloned into plasmid p13840 between the NdeI/XhoI sites.
Plasmids p11862, p11860 and p11870 carrying the recF, recO, recR genes, respectively, under the control of their natural promoter were used to express the recF, recO, recR genes in a ΔrecF, ΔrecO, ΔrecR background. To construct plasmid p11860, the recO gene was amplified by PCR using the primer pair (PS402/PS403) and the resultant product was cloned into plasmid p11520 [68] between the SacI/BamHI sites. Plasmid p11870, containing the recR gene, was constructed in a similar manner using the primer pairs PS414/PS415. The recF gene was cloned into plasmid p11520 between the SpeI/BglII sites in a similar manner using the primers PS410/PS411 to obtain p11862. All constructions were verified by DNA sequencing.
Plasmid p11562 [63], expressing recA from a PSpac promoter, was used to transform GY12968: ΔrecAΩkan giving rise to strain GY14111. The expression of recA was induced by adding 10 mM IPTG to the media.
Exponential cultures, grown in TGY2X (supplemented with spectinomycin when necessary), were concentrated to an A650 = 10 in TGY2X and irradiated on ice with a 137Cs irradiation system (Institut Curie, Orsay, France) at a dose rate of 44.7 Gy/min. Following irradiation, diluted samples were plated on TGY plates. Colonies were counted after 3–4 days incubation at 30°C.
The essentiality of genes was evaluated in a growth experiment in which the strains grown at 28°C in liquid medium with spectinomycin, were serially diluted, plated on TGY agar and incubated at 28°C or 37°C in the presence or absence of spectinomycin [21].
Non-irradiated or irradiated (6.8 kGy) cultures were diluted in TGY2X to an A650 = 0.2 and incubated at 30°C. At different post-irradiation recovery times, culture aliquots (5mL) were removed to prepare DNA plugs as described previously [60]. The agarose embedded DNA plugs were digested for 16 h at 37°C with 10 units of NotI restriction enzyme. After digestion, the plugs were subjected to pulsed field gel electrophoresis as described previously [69].
The rate of DNA synthesis was measured according to a modified protocol from Zahradka et al [1]. Exponential cultures, grown in TGYA, were concentrated to an A650 = 20 in TGYA and irradiated as described previously. Non-irradiated or irradiated cultures (6.8 kGy) were diluted in TGYA to an A650 = 0.2 and incubated at 30°C. At different time 0.5mL samples were taken and mixed with 0.1mL pre-warmed TGYA containing 4.8 µCi [methyl-3H]thymidine (PerkinElmer, specific activity 70–90 Ci/mmol). Radioactive pulses of 15 min were terminated by addition of 2 mL ice-cold 10% TCA. Samples were kept on ice for at least 1 h, and then collected by vacuum filtration onto Whatman GF/C filters followed by washing twice with 5mL 5% TCA and twice with 5mL 96% ethanol. Filters were dried for 10 min under a heat source and placed in 4 mL scintillation liquid. The precipitated counts were measured in a liquid scintillation counter (Packard, TRI- carb 1600 TR).
|
10.1371/journal.pntd.0005743 | Comparison of a new multiplex real-time PCR with the Kato Katz thick smear and copro-antigen ELISA for the detection and differentiation of Taenia spp. in human stools | Taenia solium, the cause of neurocysticercosis (NCC), has significant socioeconomic impacts on communities in developing countries. This disease, along with taeniasis is estimated to infect 2.5 to 5 million people globally. Control of T. solium NCC necessitates accurate diagnosis and treatment of T. solium taeniasis carriers. In areas where all three species of Taenia tapeworms (T. solium, Taenia saginata and Taenia asiatica) occur sympatrically, conventional microscope- and copro-antigen based diagnostic methods are unable to distinguish between these three Taenia species. Molecular diagnostic tools have been developed to overcome this limitation; however, conventional PCR-based techniques remain unsuitable for large-scale deployment in community-based surveys. Moreover, a real-time PCR (qPCR) for the discrimination of all three species of Taenia in human stool does not exist. This study describes the development and validation of a new triplex Taq-Man probe-based qPCR for the detection and discrimination of all three Taenia human tapeworms in human stools collected from communities in the Central Highlands of Vietnam. The diagnostic characteristics of the test are compared with conventional Kato Katz (KK) thick smear and copro-antigen ELISA (cAgELISA) method utilizing fecal samples from a community based cross-sectional study. Using this new multiplex real-time PCR we provide an estimate of the true prevalence of taeniasis in the source population for the community based cross-sectional study.
Primers and TaqMan probes for the specific amplification of T. solium, T. saginata and T. asiatica were designed and successfully optimized to target the internal transcribed spacer I (ITS-1) gene of T. solium and the cytochrome oxidase subunit I (COX-1) gene of T. saginata and T. asiatica. The newly designed triplex qPCR (T3qPCR) was compared to KK and cAgELISA for the detection of Taenia eggs in stool samples collected from 342 individuals in Dak Lak province, Central Highlands of Vietnam. The overall apparent prevalence of taeniasis in Dak Lak province was 6.72% (95% confidence interval (CI) [3.94–9.50]) in which T. solium accounted for 1.17% (95% CI [0.37–3.17]), according to the T3qPCR. There was sympatric presence of T. solium, T. saginata and T. asiatica. The T3qPCR proved superior to KK and cAgELISA for the detection and differentiation of Taenia species in human feces. Diagnostic sensitivities of 0.94 (95% credible interval (CrI) [0.88–0.98]), 0.82 (95% CrI [0.58–0.95]) and 0.52 (95% CrI [0.07–0.94]), and diagnostic specificities of 0.98 (95% CrI [0.94–1.00]), 0.91 (95% CrI [0.85–0.96]) and 0.99 (95% CrI [0.96–1.00]) were estimated for the diagnosis of taeniasis for the T3qPCR, cAgELISA and KK thick smear in this study, respectively.
T3qPCR is not only superior to the KK thick smear and cAgELISA in terms of diagnostic sensitivity and specificity, but it also has the advantage of discriminating between species of Taenia eggs in stools. Application of this newly developed T3qPCR has identified the existence of all three human Taenia tapeworms in Dak Lak province and proves for the first time, the existence of T. asiatica in the Central Highlands and the south of Vietnam.
| Human taeniid tapeworms comprise three species, Taenia solium, Taenia saginata and Taenia asiatica. Taeniasis is a meat-borne zoonosis transmitted by the consumption of cysticerci in raw or undercooked pork for T. solium and T. asiatica (liver) and in beef for T. saginata. Accidental ingestion of T. solium eggs by humans also results in the formation of cysticerci, often in the brain, referred to as neurocysticercosis (NCC). T. solium NCC is a significant cause of morbidity and mortality owing to epilepsy in many resource-poor communities. In animals, ingestion of eggs passed by humans results in organ and/or carcass condemnation and suboptimal economic outcomes for farmers. The accurate diagnosis of T. solium tapeworm carriers is essential to monitor the success of control programs. In areas where all three species of Taenia tapeworms occur together, conventional diagnostic methods are unable to distinguish between the different species of Taenia. In this study, we develop and apply a T3qPCR capable of detecting and discriminating all three-tapeworm species in stools in a rapid and high-throughput fashion, suitable for large-scale community surveys. The newly developed T3qPCR proved superior to previously developed immunodiagnostic and conventional microscopic-based tests in terms of diagnostic sensitivity, specificity and the ability to identify and distinguish human Taenia species. This qPCR assay facilitated the identification of T. asiatica tapeworms in the Central Highlands of Vietnam, for the first time.
| Humans are infected with three species of Taenia—Taenia solium, Taenia saginata and Taenia asiatica [1,2]. While T. saginata has a global distribution, T. solium is distributed mostly in developing countries of Latin America, sub-Saharan Africa and Asia, and T. asiatica is restricted to certain Asian countries [3,4]. Taeniasis is estimated to infect 2.5 to 5 million people globally [5–7]. All three-tapeworm species utilize humans as the definitive hosts (adult tapeworm) due to the ingestion of undercooked and/or raw meat or liver [1,8]. The intermediate hosts of T. solium and T. asiatica are swine whereas the intermediate hosts of T. saginata are cattle (porcine/bovine cysticercosis) [3]. Humans may also become the accidental intermediate hosts of T. solium via the ingestion of food and water contaminated with T. solium eggs, which may result in neurocysticercosis (NCC) when the cysts lodge in the central nervous system. The clinical manifestations of T. solium NCC in humans are varied, ranging from being asymptomatic to severe neurological signs and symptoms such as epilepsy, paralysis, dementia, chronic headache, blindness or even death [9–11]. The symptoms of taeniasis in humans, on the other hand, are mostly subtle and mild, and include abdominal distension, abdominal pain, digestive disorders and anal pruritus (mostly for T. saginata and T. asiatica) [12,13].
Control of NCC necessitates accurate diagnosis and treatment of T. solium taeniasis carriers to break the cycle of transmission to pigs [14]. Although all cases of taeniasis necessitate treatment, species-specific identification from an epidemiological perspective will allow better targeted control programs to be implemented aimed at interrupting the lifecycle specific to each species of Taenia endemic within a community and region. Several diagnostic tools for detecting taeniasis, including microscopy, copro-antigen ELISAs, sero-antibody immunoblot [15–17], and copro-DNA tests have been developed[18,19]. Of these diagnostic tools, microscopy-based examinations of fecal samples have the limitation of poor diagnostic sensitivity owing to intermittent shedding of proglottids/eggs and low specificity owing to the inability to morphologically discriminate between the eggs of Taenia species [20]. Although self-detection of proglottid shedding is valuable for confirmation of individual (but not community-wide) taeniasis, it is however, not always reliable especially when deep pit latrines are being utilized. Moreover, in cases of infection with T. solium, the proglottids are immobile and may be misidentified as other worms [19]. Differentiation of T. saginata from T. asiatica based on proglottid morphology is onerous and challenging [21]. The cAgELISA utilizes polyclonal antigens and more sensitive (0.85, 95% CI [0.62–0.98]) than microscopy-based methods (0.53, 95% CI [0.11–0.97]) [22], is only capable of identification to a genus level [23–25]. More recently, a cAgELISA specific to T. solium was developed [20,26]. Another drawback of the cAgELISA is that it may miss cases of T. asiatica and T. saginata in areas where other human Taenia tapeworms co-exist [19]. Serum antibody detecting immunoblot assays using excretory/secretory or recombinant antigens have been developed for identification of antibodies to T. solium [15,17] and T. asiatica taeniasis [16]. Immunoblot assays were highly sensitive and specific when used in Taenia non-endemic areas, however resulted in a high proportion of false positive results in areas endemic for T. saginata [17]. Copro-PCRs are considered superior tools for the detection of Taenia eggs in fecal samples owing to their high diagnostic sensitivity and the ability to distinguish Taenia species. Currently, the only multiplex-PCR [27,28] and PCR-RFLP [21] that is able to differentiate all human Taenia tapeworms is unsuited for large-scale community surveys owing to labor intensive process of restriction digest and gel electrophoresis. A better suited real-time PCR [29] has been developed but it is only capable of discriminating T. solium and T. saginata. Loop-mediated isothermal amplification (LAMP) [30] has also shown promise as a potential point-of-care diagnostic technique capable of highly sensitive detection (one copy of target gene/reaction or at least five eggs/gram of feces) and differentiation of Taenia spp. in stool [30], however the technology is still far from point-of-care based and is considered equivocal to PCR [19].
This study describes the development and validation of a new multiplex Taq-Man probe-based qPCR for the detection and discrimination of all three species of Taenia in human stools. The diagnostic characteristics of the test are compared with the test characteristics of conventional KK thick smear and the cAgELISA using hyper immune rabbit anti-Taenia IgG polyclonal antibody by applying all three diagnostic tests to detect Taenia spp. in fecal samples collected as part of a community-based cross-sectional study carried out in the Central Highlands of Vietnam.
This study was reviewed and approved by the Behavioural and Social Sciences Human Ethics Sub-committee, the University of Melbourne (reference number 1443512) and conducted under the supervision of the local Center for Public Health, Dak Lak, Vietnam. Participants positive for taeniasis on KK thick smear were treated immediately by local medical officers with praziquantel (Distocide, Shinpoong Daewoo Pharma CO.LTD) at 20 mg/kg as a single dose, whereas those positive by cELISA and qPCR were treated once results were known.
Production of polyclonal antibodies in rabbits used for cAgELISA were obtained from the Ethical Committee for animal experiments of the Institute of Tropical Medicine, Antwerp (Reference number DG012S).
An in-house coproantigen detection ELISA [33] with slight modifications, as described by Mwape et al. (1990) [34], was performed on the stool samples. Briefly, a mixture of an equal amount of Phosphate Buffered Saline (PBS) and stool sample stored in 10% formalin was soaked for one hour with slight shaking and centrifuged at 2000 g for 30 minutes where after supernatant was collected. The Polystyrene ELISA plates (Nunc Maxisorp, Waltham, MA, USA) were coated with capturing hyper immune rabbit anti-Taenia IgG polyclonal antibody diluted at 2.5 mg/ml in carbonate-bicarbonate buffer (0.06 M, pH 9.6) and incubated for 1 hour at 37°C. The plates were washed once with PBS in 0.05% Tween 20 (PBS-T20). All wells were blocked by adding PBS-T20+2% New Born Calf Serum and incubated for 1 hour at 37°C. After washing the stool supernatant of 100 ml was added to all wells and incubated for 1 hour at 37°C followed by washing five times with PBS-T20. One hundred microliter of biotinylated hyper immune rabbit IgG polyclonal antibody diluted at 2.5 mg/ml in blocking buffer was added into each well. Between two cycles of incubating for 1 hour at 37°C followed by washing five times, 100 μl of streptavidin-horseradish peroxidase (Jackson Immunoresearch Lab, West Grove, PA, USA) diluted at 1/10,000 was added. Finally, 100 ml of ortho phenylenediamine (OPD) substrate, prepared by dissolving one tablet in 6 ml of distilled water and adding 2.5 ml of hydrogen peroxide, was added to all the wells, and incubated in a dark for 15 minutes at room temperature before stopping the reaction by adding 50 ml of sulphuric acid (4 N) to each well. The plates were read using an automated spectrophotometer at 490 nm with a reference of 655 nm. To determine the test result, the optical density (OD) of each stool sample was compared with the mean of a series of eight reference Taenia negative stool samples plus 3 standard deviations.
All fecal samples were microscopically examined for the presence of Taenia eggs using a duplicate KK thick-smear technique [35–38] and examined by trained parasitologists at Tay Nguyen University.
The fieldwork was conducted between May to October 2015 in Krong Nang, M’Drak and Buon Don districts in Dak Lak province, Vietnam (Fig 1). M’ Drak is located in the east of Dak Lak province with an average altitude of 400m to 500m above sea level and a tropical monsoon climate typical of the Vietnamese Central Coast. Krong Nang is situated to the north of Dak Lak at 800m above sea level. Buon Don, situated to the west of Dak Lak at approximately 330 m above sea level and had a hot and dry climate. Living standards in each of three districts is poor; open defecation using outdoor latrines is a common practice and livestock access to these latrine areas is, for the most part, unrestricted. Moreover, consumption of raw or undercooked blood, meat, organ tissues from domesticated and wild pigs and cattle is common. The practice of non-confinement of pigs and cattle is common in rural regions with slaughter activities commonly carried out in backyards without official meat inspection. Meat inspection is only carried out at abattoirs or slaughter-points, which serve the district level and/or clusters of large villages [32].
The objective of this study was to estimate the true prevalence of human taeniasis in each of the three study districts. Based on the previous study of Van De at al. [13], the true prevalence of human taeniasis was assumed to be 7% and there was 95% certainty that this estimate was within 5% of the true population value (i.e. the true prevalence ranged from 2% to 12%). Ignoring the tendency for taeniasis to cluster within households we estimated that a total of 100 stool samples were required. We then assumed the average number of individuals eligible for the study per household was at least three and the between household (cluster) variance in taeniasis prevalence was around 1.14 times greater than the within household (cluster) variance, returning an intra-class correlation coefficient of 0.07 [39]. Our revised sample size, accounting for the tendency of taeniasis to cluster within households, was 114 for each district and 342 for the whole province.
Within each district, three villages were selected at random. Sixty-seven households were randomly sampled from within each of the selected villages. Within each household a maximum of three individuals over the age of seven years were selected at random and asked if they would like to take part in the study. Those that consented were then asked to provide informed written consent for stool donation. Participants under the age of 18 years were recruited with assent of themselves as well as written consent from their parents or legal guardians. During the participant recruitment phase investigators informed study participants that they could withdraw from the study at any stage.
A labeled stool container was delivered to each individual. Fecal samples collected were divided into three aliquots. Approximately 5 g of each fecal sample was fixed in 5% potassium dichromate (w/v), and transported to the University of Melbourne, Australia for molecular analysis. The second and the third aliquots were stored in 5% formalin, shipped to the Institute of Tropical Medicine, Belgium, and kept at Tay Nguyen University, Vietnam, for cAgELISA testing and microscopy examination (KK), respectively.
Statistical analyses were carried out using R (R Development Core Team 2017) [40], Microsoft Excel 2007 and Microsoft Access 2007. Cohen’s kappa test statistic (K) was used for comparison of multiplex qPCR to Kato-Katz, and multiplex qPCR to Coproantigen ELISA.
The diagnostic sensitivity and specificity of multiplex qPCR, cAgELISA and KK were estimated using a Bayesian approach. The Bayesian approach utilised prior information based on both previous literature [22] and expert opinion (Table 3). Prior information for the diagnostic sensitivity and specificity for the T3PCR was elicited from six independent molecular parasitology diagnosticians. Prior information for the sensitivity and specificity of cAgELISA and KK derived from Praet et al. (2013) [22] (Table 3). Prior information for the true prevalence of taeniasis in the three study districts was obtained from previous studies carried out in the same region [41].
The model of estimation of diagnostic sensitivity and specificity for three tests applied to individuals from the same population with no gold standard test was based on Branscum et al. (2005) [42] with minor modification of the assumption that the three tests were independent [43]. The posterior distribution of diagnostic sensitivity and specificity were obtained using Markov chain Monte Carlo (MCMC) techniques. The MCMC sampler was run for 100,000 iteration and the first 1,000 ‘burn in’ samples were discarded. The point estimate and 95% credible interval (CrI) for diagnostic sensitivity and specificity were reported as the median and 2.5% and 97.5% quantiles of the posterior distribution of sensitivity and specificity. Parallel chains were run using diverse initial values to ensure that convergence was achieved to the same distribution [44]. Confirmation that the posterior estimates of the monitored parameters had converged to a stable distribution was achieved by plotting cumulative path plots for each variable [45,46] and quantified using the Raftery and Lewis convergence diagnostic [47,48].
Designed primers/probes for the T3qPCR were highly species-specific and failed to cross-amplify non-target parasites including other species of Taenia. Multiplexing the assay had no effects on the sensitivity and efficiency of the qPCR (Fig 2).
Of 342 stool samples collected in Dak Lak province that were tested using T3qPCR and KK, 23 (6.7%, 95% CI [3.9–9.5]) and 7 (2%, 95% CI [0.9–4.4]) were positive for Taenia spp. For cAgELISA using a cut off of ≥ 0.2 OD, 21 (6.1%, 95% CI [3.9–9.4]) samples were classified positive for Taenia spp., whereas 3 (0.9%, 95% CI [0.2–2.7]) samples were classified positive for T. solium using a higher cut off of OD ≥ 0.55. The summarized test results are provided in Table 4.
Seven individuals that were positive for taeniasis by KK that were administered praziquantel, eliminated at least one adult Taenia worm. In all cases, the species of Taenia identified was consistent with that detected by the T3qPCR, including one individual that harboured co-infection with T. saginata and T. asiatica. Amplification and DNA sequencing of the individual adult tapeworms recovered from this individual showed that one of them had 99% similarity with GenBank sequence LM146668 of T. asiatica and the other to have 100% similarity with GenBank sequence JN986712 of T. saginata.
Of the 23 samples positive for Taenia spp. using the T3qPCR, single infections of T. solium and T. saginata were identified in 3 (0.87%) and 14 (4.09%) individuals, respectively. No cases of a single infection with T. asiatica were identified. Mixed infections of T. solium and T. saginata occurred in one individual, and of T. saginata and T. asiatica in five individuals (Fig 3).
The T3qPCR was more sensitive for detection of Taenia spp. eggs in human stools compared with KK. Twenty three of 342 individuals (6.72%, 95% CI [3.9–9.5]) were positive for Taenia egg using T3qPCR compared with seven of 342 individuals (2.05%, 95% CI [0.9–4.4]) positive using KK (Table 4). All the KK positive samples (7 of 342) were confirmed T. saginata and/or T. asiatica using the T3qPCR. No cases of T. solium detected by qPCR were positive by KK. Cohen’s Kappa test showed moderate agreement between the two methods (Table 5). The estimated diagnostic sensitivities of T3qPCR and KK for detecting Taenia spp. were 0.94 (95% CrI [0.88–0.98]) and 0.52 (95% CrI [0.07–0.94]), respectively, while the diagnostic specificity of the two tests were 0.98 (95% CrI [0.95–1.00]), and 0.99 (95% CrI [0.96–1.00]) (Table 6), respectively.
A cut off value for the cAgELISA of 0.2 was obtained by calculating the mean OD value of a series of 8-reference Taenia negative stool samples ± 3 SD. The cAgELISA identified 21 (6.14%) samples as positive for taeniasis however failed to detect 15 of 342 samples positive by T3qPCR. The agreement between the two diagnostic tests was fair with a Kappa value of 0.32 (95% CI [0.21–0.42]) (Table 5).
A cut off value of ≥0.55 OD on the cAgELISA was assumed to be more specific for the detection of T. solium antigens. Three (0.87%) of 342 individuals were positive for T. solium by this method in agreement with T3qPCR resulting in an apparent prevalence of 1.16% (95% CI [0.0–0.03]). The cAgELISA failed to detect any additional cases of T. solium detected by T3qPCR. There was very good agreement between the two diagnostic tests with Kappa of 0.86 (95% CI [0.75–0.96]) (Table 5). The diagnostic sensitivity and specificity of the cAgELISA was estimated to be 0.82 (95% CrI [0.58–0.95]) and 0.91 (95% CrI [0.85–0.96]), respectively (Table 6).
The T3qPCR developed in this study demonstrated superior ability to both the KK thick smear and cAgELISA for the specific detection and discrimination of all three taeniid tapeworms found in humans, T. solium, T. asiatica and T. saginata. The assay proved highly specific with no cross-reactivity observed for 13 different intestinal parasites as well as with each other. The cross-reactivity of the T3qPCR with Echinococcus multilocularis and Echinococcus granulosus was not evaluated as DNA of these parasites are not expected to shed in stools. The innate advantage of this T3qPCR over other diagnostic methods is its ability to detect mixed cases of Taenia spp. infection, a necessary tool that provides a more comprehensive understanding of the epidemiology of taeniasis and Taenia cysticercosis in both humans and animals in regions where all three species are sympatric [49]. For example, in communities such as those in the Central Highlands of Vietnam, where taeniasis is highly endemic, use of the T3qPCR allows the true prevalence and risk factors for T. solium taeniasis carriers to be determined. Use of these diagnostic tools to inform taeniasis control programs will not only reduce the risk of T. solium NCC in humans, but will also assist in breaking the transmission cycle to backyard and free-roaming pigs, reducing economic losses to the small-holder farmers due to carcass condemnation [50–53]. Using the Bayesian approach described in this study, the T3qPCR had an estimated sensitivity of 0.94 (95% CrI [0.88–0.98]) compared to 0.83 (95% CrI [0.57–0.98]) for a similar Taq-Man probe based multiplex qPCR described by Praet et al. (2013) [22] for the specific detection and differentiation of T. solium and T. saginata. The specificity of the T3qPCR of 0.98, 95% CrI [0.95–1.00] was equivalent (0.99, 95% CrI [0.98–0.99]) to the multiplex qPCR developed by Praet et al. (2013) [22].
Treatment and recovery followed by morphological and genetic identification of two adult tapeworms from a single individual supported the T3qPCR results to confirm the presence of T. asiatica in the Central Highlands of Vietnam for the first time, extending the known distribution of this tapeworm to southern Vietnam. This assay also confirmed the co-existence of T. solium, T. saginata and T. asiatica in the Central Highlands region. The elimination of two adult tapeworms from a single individual also confirmed the ability of the T3qPCR to detect both species of worms as opposed to infection with hybrid tapeworm. In comparison to the KK thick smear, the T3qPCR offered a significantly greater sensitivity (0.94, 95% CrI [0.88–0.99]) for the detection of Taenia eggs in stool samples. There were 16 Taenia spp. T3qPCR-positive samples not detected by KK (Table 4). The innately lower sensitivity of KK to detect eggs in feces may explain this [54]. However, the unsuitability of microscopic-based techniques for the detection of Taenia spp. in faeces may also be compounded by the inability of Taenia spp. to actively shed/lay eggs through a uterine pore like Pseudophyllidians do so [55]. As opposed to microscopy-based techniques, the T3qPCR may be in addition to any released eggs, detecting sloughed tapeworm DNA present in the stool, however this needs to be confirmed. Most significantly, KK missed all four T. solium positive individuals confirmed by T3qPCR in this community. As a result, it is our opinion that microscopic based fecal examination (KK) is neither suitable nor recommended for screening for taeniasis. Therefore, real-time PCR or cAgELISA at a higher OD cut off is recommended for screening T. solium carriers in community-based surveys in South East Asia where sympatric infection of all three human Taenia tapeworms is common [26,4].
The estimated sensitivity of the cAgELISA assay with an OD cut off value at 0.55 using polyclonal antibodies from hyper-immunized rabbits against excretory/secretory tapeworm antigens in this study was estimated at 0.82 (95% CrI [0.58–0.95]), twofold higher than that of KK. There was a strong agreement with the T3qPCR, in which all three cAgELISA-positive samples were confirmed as T. solium infection by T3qPCR. The sensitivity of the cAgELISA in this study was lower than that reported by Allan et al. (1996) [56] who determined test sensitivity on purgation of a subset of coproantigen-positive cases. Using a Bayesian approach, Praet et al. (2013) [22] Showed that the diagnostic sensitivity and specificity of cAgELISA was 0.85 (95% CrI [0.62–0.98]) and 0.92 (95% CrI [0.85–0.96]), respectively, similar to that estimated for this study. In total, thirteen of 21 cAgELISA OD 0.2-positive samples (Table 5) were not confirmed to be positive for Taenia by T3qPCR and microcopy. This suggests that the cAgELISA may cross-react with other parasites rather than Taenia spp [56]. cAgELISA shows excellent potential for a large scale epidemiological community-based survey in regions where sympatric Taenia spp. exist as it is capable of specifically identifying T. solium carriers when the cut off OD is set at ≥0.55. Therefore, choosing an appropriate OD cut off value for cAgELISA seems to be an important diagnostic consideration when applying the assay for the diagnosis of T. solium taeniasis in different communities.
In conclusion, this study describes the development and test characteristics of a new T3qPCR for the detection and differentiation of all three human Taenia species in stool. The absence of a diagnostic gold standard necessitated the use of a Bayesian approach to estimate diagnostic test performance. In future, the usefulness of the T3qPCR for large-scaled epidemiological studies can be confirmed using a larger sample size of individuals, ideally in a community in which the prevalence of taeniasis is higher. The assay also has the potential use in anthelmintic efficacy trials targeting individual species of tapeworms, thereby directly allowing assessment of species-specific susceptibility to drugs as well as non-chemotherapeutic interventions.
|
10.1371/journal.pgen.1005406 | Tempo and Mode of Transposable Element Activity in Drosophila | The evolutionary dynamics of transposable element (TE) insertions have been of continued interest since TE activity has important implications for genome evolution and adaptation. Here, we infer the transposition dynamics of TEs by comparing their abundance in natural D. melanogaster and D. simulans populations. Sequencing pools of more than 550 South African flies to at least 320-fold coverage, we determined the genome wide TE insertion frequencies in both species. We suggest that the predominance of low frequency insertions in the two species (>80% of the insertions have a frequency <0.2) is probably due to a high activity of more than 58 families in both species. We provide evidence for 50% of the TE families having temporally heterogenous transposition rates with different TE families being affected in the two species. While in D. melanogaster retrotransposons were more active, DNA transposons showed higher activity levels in D. simulans. Moreover, we suggest that LTR insertions are mostly of recent origin in both species, while DNA and non-LTR insertions are older and more frequently vertically transmitted since the split of D. melanogaster and D. simulans. We propose that the high TE activity is of recent origin in both species and a consequence of the demographic history, with habitat expansion triggering a period of rapid evolution.
| Transposable elements (TE) are stretches of DNA that propagate autonomously within genomes, but it is not clear whether TEs are moving at a constant rate or if TE activity is variable. Determining the genome-wide TE content of two closely related Drosophila species, we show that transposition rate heterogeneity is abundant. Since TE insertions are frequently associated with a selective advantage, we suggest that the observed high TE activity may have served a central role facilitating the adaptation of the two species to their novel environments after the recent out of Africa habitat expansion.
| Transposable elements (TE) are stretches of DNA that selfishly spread within genomes. Without any force counteracting their spread, TE numbers would exponentially grow within hosts until the accumulated TE burden causes extinction of host populations. Two mechanism have been proposed that could lead to a stable equilibrium of TE copy numbers within hosts, at which the number of insertions gained by transposition equals the number of TEs lost by purifying selection [1]. Either the effective transposition rate (i.e. number of new insertions less the number of excised TEs) may be a decreasing function of TE copy numbers or the strength of negative selection against TE insertions may be increasing with TE copy numbers [1]. One important outcome of strong negative selection is that most TE insertions in D. melanogaster are segregating at low population frequencies (transposition-selection balance model) [2, 3, 4]. Alternatively, TE families in D. melanogaster may not yet have attained a stable equilibrium. In this case, the predominance of low frequency insertions is thought to be due to recent activity (transposition burst model) [3, 5, 6]. In particular, families that recently invaded a novel host, like the P-element, may not yet have reached an equilibrium state [6, 7]. Nevertheless, given sufficient time all TE families are expected to eventually attain an equilibrium between the gain of new insertions by transposition and elimination of insertions facilitated by negative selection. The dynamics of TEs after reaching this equilibrium are not well understood. One possible outcome is that the equilibrium is stable, which results in vertical transmission as frequently seen for non-LTR transposons [8, 9]. Alternatively, the evolution of host factors [10, 11] could modulate transposition rates over time. Such fluctuations in TE activity could result in vertical extinction, especially if transposition rates reach low levels. Alternatively, a gradual and irreversible accumulation of deleterious mutations may inevitably lead to vertical extinction of some TE lineages [12, 13]. Horizontal transmission (HT) of active copies to a novel host may be a necessary step to ensure long-term maintainence of these lineages [14, 12]. While all these processes have been inferred from the analysis of TEs in extant populations, it is clear that the long-term evolution of TEs can only be understood if intraspecific TE dynamics can be connected between species that are sufficiently diverged to recognize differences, but also sufficiently close to make informative comparisons. We investigated the TE content in natural D. melanogaster and D. simulans populations, two closely related species which diverged about 2–3 million years ago [15, 16]. Using empirical TE insertion frequency estimates from Pool-Seq we show that, like in D. melanogaster (f ≤ 0.2; 87%), most TE insertions in D. simulans segregate at low frequencies (f ≤ 0.2; 80%). We propose that this is likely due to a high activity of more than 58 TE families in both species. This high TE activity may be of recent origin in both species, triggered by habitat expansion. Interestingly, retrotransposon families were more active in D. melanogaster while DNA transposons were more active in D. simulans.
We compared the TE abundance in natural populations of the two closely related species D. melanogaster and D. simulans to determine the patterns of long-term transposon activity. The comparison of TE abundance in the two species has been complicated by markedly different qualities of the reference genomes and the associated TE annotations. To avoid bias that might arise from using genomes assemblies of different quality, we pursued the following strategies: (i) using an improved D. simulans reference assembly [17], (ii) restricting the TE abundance comparison to orthologous regions, i.e. regions present in the assemblies of both species (iii) using the same de novo TE annotation pipeline in both species [annotating TEs in all currently available D. simulans assemblies [18, 19, 17]; see Material and Methods] and (iv) employing a TE calling method that is independent of the presence of a TE insertion in the reference genome. Our pipeline also takes sequence variation between insertions of TE families into account by mapping reads to the consensus TE sequences as well as to all sequence variations of a TE family found in the reference genome(s). From each species we analyzed isofemale lines collected 2013 in Kanonkop (South Africa). By sequencing pooled individuals (Pool-Seq) [20] we obtained an average coverage of at least 320-fold using Illumina paired end reads, which corresponds to an average physical coverage of 145 at TE insertion sites. We estimated TE abundance using PoPoolationTE [5]. The impact of the various steps in our pipeline is detailed for every TE family in S1 Table.
A comparison of de novo annotated TEs in D. melanogaster with the reference annotation [FlyBase; v5.53; [21, 22]], indicated that our pipeline for annotating TEs has a high sensitivity as well as a high specificity (S1 Text). The high quality of our TE annotation is further supported by the very similar sets of TE insertions identified in a D. melanogaster population [5] using our pipeline and either the de novo annotation of TE insertions or the reference annotation (77–91% overlap; S1 Text). Moreover the population frequency estimates and number of TE insertions in the South African D. melanogaster population were highly similar to the ones in a European population [5] despite that the latter one was based on the reference TE annotation (population frequency estimates: Spearman’s rank correlation, rS = 0.82, p < 2.2e − 16, insertion numbers: Spearman’s rank correlation, rS = 0.81, p < 2.2e − 16; S1 Text). As final validation of our annotation pipeline we compared the genomic TE distribution in natural populations obtained from our pipeline to an independently acquired data set. Vieira et al. (1999) estimated the abundance of 36 TE families in D. melanogaster and D. simulans populations by in situ hybridization. We obtained a reasonable correlation between the estimates of both methods (D. melanogaster: Spearman’s rank correlation, rS = 0.85, p = 3.6e − 9; D. simulans: rS = 0.62, p = 0.0002; S1 Text), confirming the robustness of our method. In agreement with these indicators of reliable TE identification, recent computer simulations indicated that the software used for estimating TE abundance (PoPoolationTE) has a high sensitivity [23] and TE insertions identified with this software were validated with PCR [24].
The number of TE insertions differs markedly between the two species (Fig 2) with a larger number of TE insertions in a D. melanogaster population than in a D. simulans population (Dmel = 18,382, Dsim = 13,754, Chi-squared test, χ2 = 666.5, p < 2.2e − 16; physical coverage = 145; minimum count = 3; orthologous regions). Analyzing only TE insertions for which population frequencies could be estimated (S2 Table) and excluding INE-1, an old and abundant TE family [25, 26], we found that this observation also holds when comparing the average number of TE insertions per haploid genome (Dmel = 1,275, Dsim = 1,172, Chi-squared test, χ2 = 4.3, p < 0.037; including INE-1: Dmel = 2,459, Dsim = 2,531, χ2 = 1.04, p < 0.31). A lower number of TE insertions in D. simulans than in D. melanogaster has been reported previously using in situ hybridization [27, 28, 29]. We found that the number of fixed insertions (f ≥ 0.9, allowing for some error) is very similar between the two species (Dmel = 1,574, Dsim = 1,639, Chi-squared test, χ2 = 1.315, p = 0.215) and that the different TE abundance between populations of the two species is mostly due to low frequency insertions (f ≤ 0.2, Dmel = 14,789, Dsim = 10,203, Chi-squared test, χ2 = 841.5, p < 2.2e − 16). We confirm the previously reported predominance of low frequency insertion in D. melanogaster [30, 31, 2, 5, 32] and show that the same pattern, albeit to a slightly lesser extent (f ≤ 0.2, Dmel = 87.5%, Dsim = 80.2%; Fig 1) is present in D. simulans. In agreement with this, the average population frequency of TE insertions is higher in D. simulans (0.199) than in D. melanogaster (0.146). As heterochromatic regions may contain substantial fractions of TE insertions (S1 Table) and the two reference genomes include different amounts of heterochromatin, the absence of insertions of a TE family in a comparison of orthologous regions (Fig 2), does not necessarily imply that this family is truly absent. Despite these limitations, we do not find species specific TE families. All 121 investigated TE families are present in both D. simulans and D. melanogaster (with the exception of Stalker3, which may be missing in D. simulans; S1 Table). Analyzing the different TE classes separately, we uncovered pronounced differences in TE abundance between the two species. D. melanogaster (i.e. the D. melanogaster population from South Africa) has markedly more Long Terminal Repeat (LTR; Dmel = 7,252, Dsim = 3,222; Fisher’s Exact Test p < 2.2e − 16) and non-LTR (Dmel = 5,723, Dsim = 2,902; Fisher’s exact test p < 2.2e − 16) insertions, whereas D. simulans has more Terminal Inverted Repeat (TIR) insertions (Dmel = 5,021, Dsim = 7,258; Fisher’s exact test p < 2.2e − 16). Many RNA transposon families (LTR and non-LTR) have more insertions in D. melanogaster whereas DNA transposon families (TIR) are more abundant in D. simulans (Fig 3). The unexpected presence of the P-element in D. simulans [Fig 3; [33, 34, 29]] is discussed elsewhere [24]. Despite these differences, the TE abundance is very similar between D. melanogaster and D. simulans (Spearman’s rank correlation of TE copy numbers for every family; rS = 0.57, p = 2.3e − 11; Fig 3). The similarity is higher for fixed TE insertions (Spearman’s rank correlation of fixed, f ≥ 0.9, insertions; rS = 0.73, p < 2.2e − 16) than for low frequency insertions (Spearman’s rank correlation of low frequency, f ≤ 0.2, insertions; rS = 0.52, p = 7.5e − 10). This high similarity of the abundance of fixed insertions is not unexpected as fixed insertions are highly enriched for insertions shared between D. melanogaster and D. simulans (Fisher’s exact test; p < 2.2e − 16; S2 Text), which likely predate the split between these two species about 2–3 million years ago [15, 16].
To test if the observed differences in the TE abundance between the two species could be caused by heterogenous transposition rates, we performed computer simulations. For each TE family we tested whether the observed interspecific differences in copy number (Fig 3) deviate significantly from expectations under drift using an equilibrium model in which we assume that the transposition rate and the selective effects are the same in both species. Our simulations considered each TE family separately and relied on a fitness function in which fitness decreases exponentially with insertion numbers, a necessary condition for obtaining stable equilibria [1]: w i = 1 − x g i t, where wi is the fitness of a given individual, x the selective impact of a TE insertions, gi the number of TE insertions found in a given individual and t the degree of synergism between TE insertions (needs to be > 1.0 for stable equilibria). We refrained from simulating other models that would also lead to stable equilibria, which either require that the transposition rate decreases or that the excision rate increases with insertion numbers [1], as there is little support for these models [10]. Given the strong influence of population size on TE dynamics [35, 36] (S3 Text), we used a population size ratio in our computer simulations that reflects the ratio of the population variation estimator π (πDsim/πDmel = 0.0113/0.0074 = 1.519; S4 Text). These simulations provide the probability (p) that the observed difference in TE copy numbers between D. simulans and D. melanogaster is compatible with the null hypothesis of an equilibrium model with genetic drift, constant transposition rates, and equal negative selection against TE insertions. In about 50% (46/93) of the TE families the number of insertions deviated significantly from expectations under drift after accounting for differences in population size (Fig 3; see Fig 4 for an illustration of the procedure used for identifying significant deviations). This result was robust with respect to a wide range of different population sizes (N ≥ 10,000; S3 Text). Also when assuming an equal population size of the two species (e.g. [37]) substantial deviations from expectations under drift were identified (Fig 3). Furthermore, our results are robust to recombination rates allowing even higher ones than those reported for D. melanogaster [as may be found in D. simulans [38]] as well as over a wide range of other parameters (t ≥ 1.3, x ≥ 0.0004; S3 Text). Relaxing these parameters further (e.g. t < 1.3) quickly results in conditions under which purifying selection against TEs is too weak to maintain stable TE copy numbers, leading to extinction of the host population (S3 Text). The fraction of families with heterogenous transposition rates is roughly similar for all three TE orders [LTR 51% (25/49), non-LTR 42% (11/26), TIR 55% (10/18)]. RNA transposons (LTR and non-LTR) are significantly more active in D. melanogaster while DNA transposons (TIR) are more active in D. simulans (Dsim RNA = 7, DNA = 7; Dmel RNA = 29, DNA = 3; Fisher’s exact test; p = 0.0045). It is important to note that these results are based on the assumption that TE families evolve in transposition-selection balance [3] which, although probably true for most TE families [7], may not hold for families that recently invaded a novel host, like the P-element [7]. Especially LTR transposons could be of very recent origin and thus not yet in transposition-selection equilibrium [6, 5]. Therefore, we separately analysed TE families that are likely vertically transmitted as a conservative set of TE families in transposition-selection balance. We identified families with at least one shared TE insertion between D. simulans and D. melanogaster [only high frequency insertions, f ≥ 0.8, were considered as the strong insertion bias of some TE families may lead to shared low frequency insertions [24]; Fig 3], suggesting vertical transmission since the split of the two species. In total 28 families had at least one shared insertion, with TIRs having the most and LTRs the least [TIR 50% (9/18), non-LTR 42% (11/26), LTR 16% (8/49)]. For about 57% (16/28) of vertically transmitted families the TE abundance between the two species significantly deviated from expectations under drift (Fig 3).
The large number of species specific TE activity patterns encouraged us to evaluate the distribution of TEs between two D. melanogaster populations from South Africa and Portugal. We observed substantial differences in TE abundance for two families (R1A1-element, gypsy2; S1 Text). This pattern is in agreement with previous observations [39, 29] suggesting that the TE composition of local Drosophila populations can differ markedly despite little differentiation among cosmopolitan D. melanogaster populations [40, 41].
The age distribution of TE insertions is an important parameter describing the dynamics of TEs. A direct approach to determine the age of TE insertions is based on the number of mutations after insertion [6, 9, 42, 43], but this method cannot be applied to Pool-Seq data. Nevertheless, the previously demonstrated strong correlation between sequence divergence of TEs and their frequency in a natural D. melanogaster population [5] suggests that population frequencies of TE insertions are good age estimators, with young insertions mostly segregating at low population frequencies while old insertions frequently have higher population frequencies. We further scrutinized this relationship by reasoning that young TE insertions are more likely to be expressed. Using RNA-Seq data from D. simulans [17] we found a significant negative correlation (Spearman’s rank correlation, rS = −0.34, p = 0.00024;S1 Fig) between population frequency and expression intensity. By contrast, we show that fixed TE insertions are mostly old as we found them to be enriched for insertions predating the split between D. simulans and D. melanogaster (see above and S2 Text). Overall, our analyses suggested that the population frequency of TE insertions provides a rough, but suitable estimator for the age of TE insertions. Based on this estimator we suggest that low frequency insertions are mostly due to recent TE activity. Hence, the predominance of low frequency insertions in D. melanogaster and D. simulans is due to recent activity of multiple TE families in both species (f ≤ 0.2, Dmel = 87.5%, Dsim = 80.2%), where 58 families (62%; 58/93) in D. melanogaster and 64 (68%; 64/93) families in D. simulans have more than 10 low frequency insertions. The five families with the lowest population frequency, and thus likely the most recently active TE families, in D. melanogaster are: P-element, Tirant, R2-element, copia and mdg1; and in D. simulans: P-element, R2-element, gypsy, G6 and accord2 (see S3 Table for full data set). This inference could be confirmed for the P-element, which invaded both species only within the last few decades [44, 24]. In both species LTR insertions have, on the average, the lowest population frequency whereas TIR insertions have the highest (Dsim LTR = 0.11, non – LTR = 0.13, TIR = 0.26; Dmel LTR = 0.07, non – LTR = 0.08, TIR = 0.33) suggesting that in both species LTR insertions are mostly of recent origin. This is in agreement with previous work which showed that LTR insertions in D. melanogaster are mostly young [6, 45]. We found that the average population frequency of TE families is correlated between D. simulans and D. melanogaster (Spearman’s rank correlation for families having at least one insertion in both species; rS = 0.57, p = 5.0e − 10). This correlation is strongest for TIR transposons and weakest for LTR transposons (LTR rS = 0.43, p = 0.001; non-LTR rS = 0.59, p = 0.0004; TIR rS = 0.81, p = 7.0e − 05), which suggests that the timing of activity is most similar between TIR families and the least between LTR families. We propose that this could be the outcome of different modes of transmission of TEs. Previous studies suggested that non-LTR transposons may be preferentially vertical transmitted [8, 9]. In agreement with this we found a high fraction of vertically transmitted TE families (estimated as families sharing one high frequency insertion between the two species; see above) for non-LTR but also for TIR transposons. LTR transposons had the smallest fraction of vertically transmitted families [LTR = 16% (8/49), non – LTR = 42% (11/26), TIR = 50% (9/18); Fig 3]. Conversely, in a scan for evidence of horizontal transfer of TEs between D. simulans and D. melanogaster, Bartolome et al. [46] found putative HT for many LTR families but only for a few non-LTR and TIR families [S1 Table from [46]; Ks < 0.04; LTR = 81% (26/32), non – LTR = 23% (3/13), TIR = 33% (1/3); Fig 3]. It is thus possible that vertical transmission is more frequent for TIR and non-LTR transposons, while HT is more frequent for LTR transposon. This could account for the weak correlation of average age of TE insertions (as measured by population frequency) of LTR families and the strong correlation of non-LTR and TIR families, as vertical transmission may result in more predictable temporal development of TE activity than HT, which is a highly stochastic process (e.g. [24]).
In this report, we provide the first genome-wide characterization of TE abundance in large population samples of the two closely related species D. simulans and D. melanogaster. Consistent with previous reports [29, 48], we found considerable differences in TE composition between the two species.
We show that in both species, D. simulans and D. melanogaster, most TE insertions segregate at of low population frequencies. We propose that this predominance of low frequency insertions is most likely due to a high activity of multiple (> 58) TE families in both species, which raises the important question whether this high activity is continuously maintained, e.g. since the split of the two species, or is of recent origin. Based on the observation that TE abundance in ancestral African populations of D. melanogaster is lower than in populations of other continents and because of the generally high heterogeneity of TE abundance in D. simulans populations, Vieira et al. [29] suggested that the recent habitat expansion of D. simulans and D. melanogaster may have triggered bursts of TE activity in these two species [49, 50]. Colonization of new environments may trigger increased TE activity by two, not mutually exclusive mechanisms: either stress associated with new environments disturbs systems that guard against TE proliferation, such as piRNA, or the habitat expansion may bring species into contact, that not co-existed previously. In combination with horizontal transfer of TEs, this could result in activity of a TE in a new host [34, 51]. One classic example for this scenario is the transfer of the P-element from D. willistoni to D. melanogaster, which invaded the territory of D. willistoni in South America [34]. After the horizontal transfer, the P-element rapidly spread in D. melanogaster populations worldwide [52]. Moreover, previously dormant TE families may also become reactivated upon the activation of a single TE family, as has been noted during hybrid dysgenesis [53, 54], where DNA damage mediated stress seems to be causative [54, 55, 53].
The hypothesis of a recent increase in TE activity in both D. melanogaster and D. simulans is supported by several lines of evidence. First, based on computer simulations we find transposition rate heterogeneity in 50% (46/93) of TE families. Since our test is designed to detect differences between the two species and at least some TE families have recently increased their transposition activity in both species it is likely that the phenomenon of transposition rate heterogenetiy is even more common than our data suggests. For example, the P-element has a high, albeit unequal, activity in both D. simulans and D. melanogaster, but it only invaded both species within the last 100 years [44, 24]. Another example is the I-element, which has about equal activity in both species, but it was suggested that active copies of the I-element were lost in D. melanogaster some time ago, and that active copies only recently reinvaded extant populations [56] (Fig 3). Furthermore, differences in TE composition are not only recognized in between-species comparisons, but can be also detected between two D. melanogaster populations (S1 Text). These differences are unlikely to result only from demographic events since these should affect all TE families equally, whereas we only found marked differences for two TE families. Such differences in TE abundance between populations have also been observed in D. simulans [39]. Third, LTR transposons may be of recent origin in D. melanogaster [6, 45]. Based on low population frequencies we suggest that this probably also holds for LTR insertions in D. simulans. Consequently, LTR insertions may be of very recent origin in both species. Fourth, HT of TEs, one mechanism by which habitat expansion could trigger bursts of TE activity, has been reported to be abundant in D. melanogaster especially for LTR transposons [46, 57].
In summary, we conclude that the TE composition in D. simulans and D. melanogaster is probably dynamic and changes quickly, such that inter-population differences can also be detected. It is therefore conceivable that the high TE activity in D. melanogaster as well as in D. simulans is of recent origin. With TE insertions frequently contributing to adaptation to novel environments [58, 5], increased transposition rates may be an important component of successful habitat expansions.
Since it is well understood that the distribution of TE insertions is strongly affected by population size [1, 59], any comparison of TEs in two closely related species needs to account for heterogeneity in genetic drift due to different population sizes in both species. Our computer simulations suggest that the observed differences in copy numbers could not be explained by genetic drift for about half of the TE families. Nevertheless, differences in TE abundance may either be due to differences in transposition rates or strength in purifying selection removing TE insertions. Since population size [59] and the recombination rate [60], the major factors modulating the strength of selection against TE insertions, affect all families similarly, our data are not compatible with unequal purifying selection. The observation that some families are more abundant in D. simulans while other families are more abundant in D. melanogaster strongly suggests the presence of family specific factors that evolved heterogeneously in the different lineages. As family specific divergence of transposition rates has also been documented previously [11, 61, 53, 54], we propose that heterogenous transposition rates are the most likely explanation for significant differences in TE abundance between the two species. However, our computer simulations made several assumptions about the behaviour of TEs and, like for all models, the conclusions drawn are strongly dependent on the parameters used in the computer simulations. Unfortunately, very little is known about the key parameters determining the dynamics of TEs: i) Which of the three equilibrium models (decreasing fitness, decreasing transposition rate, increasing excision rate) or which combination of these three models [1] reflects reality best? ii) Which fitness function most accurately describes the relationship between TE copy number and fitness? iii) What are the biological realistic values of the parameters entering the fitness functions? iv) Is a model assuming co-dominant, recessive or dominant effects of TE insertions closest to reality? v) What are the exact recombination rates of D. melanogaster and D. simulans? vi) Should differences in recombination rates enter the fitness function and if so which function best describes this effect (for example, due to the deleterious effects of ectopic recombination, it is possible that the selective impact of a given TE insertion depends on the recombination neighborhood)? vii) Are more complex demographic scenarios necessary—for example those involving migration—and if so which is the exact demographic history of the two populations? Since it is not possible to consider all these factors in our computer simulations we decided to rely on commonly used default parameters [co-dominant model with exponentially decreasing fitness function; t = 1.3; x = 0.0004; 0.0 ≥ v ≥ 0.003 (e.g. [1, 62]) and to closely reproduce the genomic landscape of D. melanogaster [68,700,000 million insertion sites in high recombining regions (> 1cM/Mbp) on four chromosome arms; the recombination rate of D. melanogaster [63]]. Finally, our simulations reproduce the sampling properties of our study (145 haploid genomes with a minimum count per TE insertion of 3).
Interestingly, retrotransposon families are more active in D. melanogaster while DNA transposons are more active in D. simulans. This contrast may be the outcome of different propensities for horizontal transfer among the major TE groups (LTR, non-LTR, TIR) in combination with the different colonization times of D. melanogaster and D. simulans. DNA transposons (TIR) and LTR transposons seem to be more prone to horizontal transfer than non-LTR TEs, since their double stranded DNA intermediates may be more stable than the RNA intermediate of non-LTR TEs [64, 9]. Furthermore, the integration of DNA transposons requires only transposase and no specific host factor, which makes these TEs potentially more successful invaders of diverged genomes [64, 51]. The very recent out of Africa habitat expansion of D. simulans [65] about 100 years ago is therefore consistent with the higher activity of DNA transposons. D. melanogaster, on the other hand, colonized Europe already more than 10,000 years ago [66], providing sufficient time for less invasive retrotransposons to colonize a new host. Furthermore, if D. melanogaster experienced a burst of DNA TEs shortly after the colonization, the host defense system (e.g.: the piRNA system [67]) may have matured to control the initially invading DNA TEs. Under this scenario, the genomic TE signature in D. simulans is expected to experience a transition from high activity of DNA transposons to high activity of retrotransposon in the next couple of centuries. However, a high propensity for HT of TIR transposons [64, 51] could be interpreted to counter our observation that many TIR families are vertically transmitted. Nevertheless, TEs like the I-element may invade hosts in multiple waves [56], and HT could therefore be abundant even for vertically transmitted TE families. Families with evidence for both vertical and horizontal transmission, like 412 and jockey (Fig 3), may have experienced multiple waves of invasion.
The likely role of habitat expansions for TE activity raise questions regarding genomic TE distributions in species that remained in their original habitat. Does this imply that TE activity is lower in endemic species? The analysis of ancestral African D. melanogaster and D. simulans populations may help to resolve this question as well as related Drosophila species that remained in their ancestral habitat. Furthermore, monitoring TE abundance in experimentally evolving populations may shed some light on the dynamics of TEs in populations and on the short term evolution of transposition rates. Finally, long read sequencing could provide a better characterization of TE insertions [68], which may help unraveling the phylogenetic relationship of TEs and thus provide some clues on the role of vertical and horizontal transmission in the life-cycle of TEs.
We collected 1,300 isofemale lines of D. simulans and 1,250 isofemale lines of D. melanogaster from Kanonkop (South Africa) in 2013. The lines were kept in the laboratory for 8 generations. We used a single female from 793 (554) isofemale D. simulans (D. melanogaster) lines for pooling. Genomic DNA was extracted from pooled flies using a high salt extraction protocol [69] and sheared using a Covaris S2 device (Covaris, Inc. Woburn, MA, USA).
We used three different protocols to prepare paired-end libraries. One library (BGI-91a; S4 Table) was prepared following a modified version of the NEBNext Ultra protocol (New England Biolabs, Ipswich, MA). For another library (BGI-92a, BGI-92b, BGI-93b; S4 Table) we used the NEXTflex PCR-Free DNA Sequencing Kit (Bioo Scientific, Austin, Texas) with modifications. The third library (BGI-93a; S4 Table) was prepared based on the NEBNext DNA Sample Prep modules (New England Biolabs, Ipswich, MA) in combination with index adapters from the TruSeq v2 DNA Sample Prep Kit (Illumina, San Diego, CA). All protocols made use of barcoding (S4 Table). For each library we selected for a narrow insert size, ranging from 260–340, using agarose gels. A total of five lanes 2x100bp paired-end reads were sequenced on a HiSeq2000 (Illumina, San Diego, CA). In summary we sequenced 364 million paired end fragments for D. melanogaster and 288 million paired end fragments for D. simulans (S5 and S6 Tables). This yields an average coverage of 381 in D. melanogaster and of 327 in D. simulans.
One of the requirements for estimating the abundance of TE insertions with PoPoolation TE [5] is a reliable TE data base. A manually curated high-quality annotation of TE insertions has been generated for D. melanogaster [22, 21], whereas, to our knowledge, so far no TE annotation of comparable quality exists for D. simulans. To avoid any biases that may result from using TE annotations of different qualities we decided to de novo annotate TE insertions in both species with an identical pipeline. The reference sequence of D. melanogaster (v5.53) was obtained from FlyBase (http://flybase.org). We used the reference sequence published by Palmieri et al. [17] for D. simulans, as this assembly is of a higher quality than the previously available one [18] and of similar quality as a recently published one [19]. We also obtained a library containing the consensus sequences of Drosophila TEs (transposon_sequence_set.embl; v9.42; [21]) from FlyBase. To avoid identification of spurious TE insertions we excluded canonical TE sequences not derived from D. melanogaster or D. simulans (Casey Bergman; personal communication). We mapped the consensus TE sequences against both reference genomes with RepeatMasker open-4.0.3 [70] using the RMBlast (v2.2.28) search engine and the settings recommended by [71] (-gccalc -s -cutoff 200 -no_is -nolow -norna -gff -u), yielding a raw annotation of TE insertions. The consensus sequences of several TE families contain microsatellites which may, as an artefact, be annotated as TE insertions [71, 21]. To account for this, we identified microsatellites in both reference genomes with SciRoKo 3.4 [72] (required score 12; mismatch penalty 2; seed length 8; seed repeats 3; mismatches at once 3), converted the output into a ‘gtf’ file and removed TEs from the raw annotation that overlapped with a microsatellite over more than 30% of the length using bedtools (v2.17.0; intersectBed -a rawannotation.gff -b microsatellites.gff -v -f 0.3) [73]. Overlapping TE insertions of the same family were merged and disjoint TE insertions of the same family were linked using an algorithm that, similar to dynamic programming, maximizes the score of the linked TE insertions (match – score = 1, mismatch – penalty = 0.5). We resolved overlapping TE insertions of different families by prioritizing the longest TE insertion and iteratively truncating the overlapping regions of the next longest insertions. Finally we filtered for TE insertions having a minimum length of 100 bp.
Estimating the abundance of TE insertions with PoPoolation TE requires paired end sequences from natural populations, a reference sequence, an annotation of TE sequences and a hierarchy of the TE sequences [5]. We extracted the hierarchy of TE sequences from the database of consensus TE sequences (v9.42; see above). We extracted the sequences of the annotated TE insertions from the reference genomes into a distinct file and subsequently masked these TE sequences within the reference genome with the character ‘N’. We than concatenated the individual fasta records of (i) the consensus sequences of TE insertions, (ii) the TE sequences extracted from the reference genome and (iii) the repeat masked reference genome into a single file, which we call TE-merged-reference. Short read mapping software usually only allows for a few mismatches between read and reference genome which may lead to underestimating the abundance of some TE insertions, especially when the TE sequences are highly diverged [5]. Such a high divergence between reads and the reference sequences may also result when the consensus sequences of TE families are derived from a different species. This could lead to underestimating the abundance of TE insertions in D. simulans when using consensus sequences that are mostly derived from D. melanogaster. Therefore, we improved the sensitivity of our pipeline for D. simulans by including TE sequences extracted from the assemblies of Begun et al. [18], Palmieri et al. [17] and Hu et al. [19] (using the same TE annotation pipeline as described above) into the TE-merged-reference of D. simulans.
We mapped 364 million PE fragments of D. melanogaster and 288 million PE fragments of D. simulans (see above) to the respective TE-merged-reference with bwa (v0.7.5a) [74] using the bwa-sw algorithm [75] (S5 and S6 Tables). We used ‘samro’ to restore the paired end information [5]. We estimated the abundance of TE insertions with PoPoolation TE similarly as described in [5] using the following settings: identify-te-insertions.pl –te-hierarchy-level family, –min-count 3, –min-map-qual 15, –narrow-range 100; crosslink-te-sites.pl –min-dist 85, –max-dist 300; estimate-polymorphism.pl –te-hierarchy-level family, –min-map-qual 15; Subsequently we filtered for TE insertions located on the major chromosome arms (X, 2L, 2R, 3L, 3R, 4) and for TE insertions having a minimum physical coverage of 30 (physical coverage as defined here is the sum of paired end fragments that either confirm the presence or the absence of a TE insertion). An unbiased comparison of the abundance of TE insertions between different species requires similar physical coverages in all species. We therefore iteratively subsampled paired-end fragments and repeated TE identification with PoPoolation TE, until we obtained similar physical coverages in both species (S7 Table). The full information about the effect of each step of the pipeline used for estimating TE abundance is enclosed in S1 Table. This file shows for every TE family the number of mapped reads, the number of paired-end fragments supporting a TE insertion, and the TE insertions finally identified during various filtering steps.
We estimated genome-wide levels of nucleotide diversity in the two natural populations using Pool-Seq data and PoPoolation [76]. First, we aligned all reads to the respective reference genome (unmodified) with bwa aln (0.7.5a) [74] and the following parameters: -I -m 100000 -o 1 -n 0.01 -l 200 -e 12 -d 12; Duplicate reads were removed with Picard (v1.95; http://picard.sourceforge.net/). Reads with a mapping quality lower than 20 or reads not mapped as proper pairs were removed with samtools (v0.1.19) [77]. We created a pileup file for each population with samtools (v0.1.19) [77] and the following parameters: -B -Q 0; As alignments spanning indels are frequently unreliable and may lead to spurious SNP calls we removed regions flanking indels (5bp in each direction; minimum count of indel 4) from the pileup with PoPoolation [76]. Subsequently we subsampled the pileup to a uniform coverage of 175 with PoPoolation [72] and the following parameters: –max-coverage 1400 –min-qual 20 –method withoutreplace; Finally we calculated π for windows of 100kb with PoPoolation and the following paramters: –min-count 4 –min-coverage 165 –max-coverage 175 –min-covered-fraction 0.6 –min-qual 20 –no-discard-deletions –pool-size 1300;
To measure the expression level of different TE families in D. simulans we obtained previously published RNA-seq reads [17], derived from a mix of several developmental stages of D. simulans strain M252. The reads were trimmed with PoPoolation v1.2.2 (trim-fastq.pl) [76] using the following parameters: –fastq-type illumina, –quality-threshold 20, –min-length 40; We mapped the RNA-seq reads to a database consisting of the repeat masked reference genome of D. simulans [17] and the library of TE sequences derived from all three assemblies of D. simulans (see above). Reads were mapped with bwa (v0.7.5a) [74] using the bwa-sw algorithm [75]. Subsequently we counted the number of reads mapping to each TE family and normalized counts by the length of the consensus sequence (transposon_sequence_set.embl; v9.42; see above).
The assemblies of D. melanogaster and D. simulans are of different quality, for example varying in the amount of assembled heterochromatin. An unbiased analysis of TE abundance should therefore be restricted to genomic regions being present in the assemblies of both species. We identified these regions by aligning the genomes of D. melanogaster (v5.53) and D. simulans [17] with MUMmer (v3.23; nucmer) [78]. To avoid spurious alignments we masked all sequences derived from TEs in both reference genomes (see above) prior to the alignment. Coordinates were extracted with the ‘show-coords’ tool [78] and only alignments of the major chromosome arms (X, 2L, 2R, 3L, 3R, 4) were considered. Due to the masking of TE sequences these raw alignments contain a plenitude of gaps where the TE insertions actually causing the gaps are not found in genomic regions that are present in the alignment. To mitigate this we linked these gaps by merging alignments not separated by more than 20,000bp in both species. This threshold of 20,000bp was arbitrarily chosen because only six of the masked regions in the repeat-masked genome of D. melanogaster have a size larger than 20,000bp.
We performed forward simulations for estimating the variance of TE abundance in natural populations expected under an equilibrium model. The simulations aimed to capture conditions found in D. melanogaster and accordingly we (i) simulated diploid organisms, (ii) used a genome with a similar size and number of chromosomes as D. melanogaster and (iii) used the recombination rate of D. melanogaster. We obtained the recombination rate from the D. melanogaster recombination rate calculator v2.2 [63] for windows of 1000kb. We excluded the X-chromosome and low recombining regions (< 1cM/Mbp)- including the entire chromosome 4—from the analysis (for both the simulations and the actual data to which the simulation results are compared to). In summary we performed our simulations with T = 68,700,000 TE insertions sites (distributed over the following genomic regions 2L:300,000–16,600,000, 2R:3,900,000–20,700,000, 3L:900,000–17,400,000, 3R:6,600,000–25,700,000) where every insertion site may either be empty or occupied. In our model, every TE insertion has a constant probability of transposing to a novel site v and excision events (u = 0) were not considered. Novel TEs were randomly inserted in any of the T insertion sites at any of the two haploid genomes. If an insertion site was already occupied the transposition event was ignored. For any individual i in a population of size N the fitness wi can be calculated as w i = 1 − x g i t, where gi is the number of TE insertions, x is the selective disadvantage of each insertion and t represents the interactions between the insertions [1]. This is a model where all TE insertions exert a semi-dominant effect [1].
Per default we used x = 0.0004 and t = 1.3 in our simulations. We furthermore used fecundity selection, where any individual has a probability of mating pi that linearly scales with fitness wi (p i = w j / N w ‾; w ‾ is the average fitness; after [79]).
We simulated evolving populations with non-overlapping generations, proceeding at every generation in the following order: First N random pairs were picked according to the mating probability pi, where selfing was excluded. Second, each parent contributed a single gamete to the offspring wherein crossing over events were introduced according to the specified recombination rate (see above). Third, fitness of the offspring wi was calculated from the abundance of TE insertions in the resulting genome of the offspring. And fourth, transposition events were introduced according to the transposition rate v. Note that the novel TE insertions will only contribute to fitness in the next generation. This could for example be interpreted as TE activity in the germline which will mostly also only effect the next generation (i.e.: the offspring). In all simulations, we performed forward simulations for 10,000 generations. We noted that if a stable equilibrium could be reached (e.g.: no increase in the number of fixed insertions), it took less than 5,000 generations. To match the analysis of natural populations we also sampled 145 haploid genomes after the 10,000 generations and required a minimum count of 3 to identify a TE (see above).
|
10.1371/journal.ppat.1007536 | Extracellular vesicles from Kaposi Sarcoma-associated herpesvirus lymphoma induce long-term endothelial cell reprogramming | Extracellular signaling is a mechanism that higher eukaryotes have evolved to facilitate organismal homeostasis. Recent years have seen an emerging interest in the role of secreted microvesicles, termed extracellular vesicles (EV) or exosomes in this signaling network. EV contents can be modified by the cell in response to stimuli, allowing them to relay information to neighboring cells, influencing their physiology. Here we show that the tumor virus Kaposi’s Sarcoma-associated herpesvirus (KSHV) hijacks this signaling pathway to induce cell proliferation, migration, and transcriptome reprogramming in cells not infected with the virus. KSHV-EV activates the canonical MEK/ERK pathway, while not alerting innate immune regulators, allowing the virus to exert these changes without cellular pathogen recognition. Collectively, we propose that KSHV establishes a niche favorable for viral spread and cell transformation through cell-derived vesicles, all while avoiding detection.
| The role of extracellular vesicles (EV) has received considerable attention in recent years. The contents of these cell-derived vesicles have been shown to be modulated upon challenge by a virus or neoplastic transformation, and can influence the behavior of recipient cells. Here we demonstrate that purified EV from an AIDS-associated, virus-driven lymphoma induces unique cellular signaling, motility, and gene expression reprogramming in recipient endothelial cells. This was accomplished without activation of innate immune activation, even after prolonged exposure to these EV. Collectively our results point toward a model in which tumor-derived EV condition neighboring cell physiology, while avoiding detection by immune regulators.
| Extracellular communication is pivotal to maintain organismal homeostasis and a disease state. One of the mechanisms by which cells communicate to their surroundings is through extracellular vesicles (EV). Cells package a number of biological molecules into EV such as proteins, nucleic acids, lipids, and metabolites. EV are released into the microenvironment as well as into the circulation via the blood and lymphatic vessels. All vessels are lined with endothelial cells (EC), which are permanently exposed to EV, and uptake the EV-loaded cargo. This may induce changes in differentiation, metabolism, migration, and gene expression (reviewed in [1], for further examples see [2]). Ample experimental evidence has connected EV to cancer metastasis, immune signaling and the response to invading pathogens, though the molecular details of EV biology are far from established and tend to differ dramatically among experimental systems [1]. As in many aspects of modern biology it is difficult to come up with experimental approaches that are both tractable and physiologically relevant. Kaposi Sarcoma (KS) and Kaposi-sarcoma-associated herpesvirus (KSHV) lymphomas, specifically primary effusion lymphoma (PEL), represent one such system to study EV biology.
KS is one of the most angiogenic cancers in humans, and was one of the first identifiable markers for AIDS (reviewed in [3]). KSHV is the etiological agent of KS and PEL, which induce a unique tumor microenvironment that remodels tumor and lymphatic vasculature [4–6]. Of note, the transdifferentiation induced by the virus is as dramatic in uninfected neighboring cells as in virus-infected cells. We had shown that KSHV-infected cells release EV containing all viral micro RNAs (miRNA), but not the virus itself, and that the viral miRNAs are present at high concentrations in KSHV-lymphoma derived EV (KSHV-EV) in culture and in patients [7]. This establishes the KSHV-EV:EC interaction as physiologically relevant, experimentally robust, and as we show here, highly tractable.
EV membranes are enriched for phosphatidylserine (PS), which is recognized by Annexin-V and plays a role in EV adsorption. Tetraspanins, such as CD9, CD81, and CD63, are found on the majority of EV and have been used to affinity-purify EV [8–12]. Alix and Flotillins-1 and 2 are additional molecules that define EV (see http://www.exocarta.org/). Multiple EV purification methods have been devised [13]. These purification schemes yield comparable preparation of EV though the resultant fractions can be quite heterogeneous and need to be carefully characterized in each experiment. Exosomes are a subtype of EV that is defined by their intracellular biogenesis. Exosomes originate from the inward budding of the late endosome into the multivesicular body and traffic from there to the plasma membrane where they are released. When studies use primary patient material and/or cell culture supernatant, the origin of the EV cannot unequivocally be attributed to the multivesicular body; and the term EV rather than exosome is used. To date, no EV-specific receptors have been defined; evidence for tissue-specific uptake is limited to specific scenarios such as neuronal or immunological synapses [14]. Unlike viruses, EV are believed to be able to enter all cell types. Viruses that modulate EV maturation and content include Human and Simian immunodeficiency virus (HIV and SIV), vaccinia virus, hepatitis C virus, and herpesviruses [1, 7, 12]. This led Gould et al. to propose the trojan horse hypothesis [15], whereby viruses use EV to modulate the cell physiology of neighboring cells in order to further infection.
In this study, we sought to characterize how EV taken from the HIV-associated PEL, either cell culture or primary patient fluid, influence EC behavior. We discovered that affinity purified EV mediated cell migration, proliferation, and secretion of human interleukin-6 (IL-6) through the extracellular signal related kinases (ERK1/2) pathway. This was accomplished without tripping of innate sensors such as interferon regulatory factor 3 (IRF3), stimulator of interferon genes (STING) [16–21], or nuclear factor kappa B (NF-κB). It allows KSHV to modulate its immediate environment without alerting innate immune signaling pathways. Chronic exposure to PEL-derived EV, mimicking the pleural environment, resulted in a reprogramming of the recipient cells’ mRNA profile, without activating innate immune signaling cascades and/or interferon stimulatory genes. Collectively, these results provide a new paradigm for EV function in modifying the phenotype of recipient cells and highlight a previously unknown method of virus-induced cellular reprogramming without alerting viral sensors.
EV research is still a developing field. Functional studies are dependent on consistent and pure EV preparations. Hence, a carefully controlled purification strategy was established and validated. EV from a KSHV-negative B-cell lymphoma (BJAB) and a KSHV-positive primary effusion lymphoma (PEL) cell line (BCBL1) were compared to each other and to EV from primary KSHV-tumor effusions and normal plasma. To allow for large scale EV purification, an initial concentration step using polyethylene glycol (PEG) was used [13]. No discernable size or concentration differences were observed between EV isolated from BJAB and BCBL1 cell supernatant. The mean and mode sizes were <100 nm (Fig 1A and 1B), the particle concentrations were similar (Fig 1C), and acetylcholine esterase (AchE) activity was comparable across all preparations (Fig 1D). The biophysical properties of EV from PEL were similar to those of the Epstein-Barr virus (EBV)-negative B cell lymphoma BJAB, and the EBV-positive B cell lymphoma cell line Namalwa (S1 Fig).
Following established standards [22] the EV markers CD63, CD81, CD9, and Flotillin-2 were evaluated by Western blot (Fig 1E). All proteins were present at comparable levels in EV and none were detected in EV-depleted media. The KSHV latency-associated nuclear antigen (LANA) was present in BCBL1 lysate, but not BJAB lysate and not in EV. GAPDH was present in whole cell lysates, but not EV. MS/MS confirmed the presence of additional, prototypical EV markers, but no viral proteins. There was no difference in EV marker protein content as defined by exocarta.org among EV from BJAB vs BCBL1 (S2 Fig). KSHV-encoded miRNAs are incorporated at high concentrations into EV and provide a high sensitivity marker to trace BCBL1 derived EV [7]. The miRK-12-5 was present in BCBL1 cells and BCBL1-derived EV, but not BJAB cells or BJAB-derived EV (Fig 1F). Negative staining electron microscopy (EM) of EV showed small, rounded vesicles of ~40–70 nm diameter (Fig 1G). After PEG enrichment, EV were further purified by either ultracentrifugation or column purification. Both procedures yielded comparable results (S3 Fig and S4 Fig). In sum, this experimental approach resulted in clean and concentrated EV preparations.
To link the cell line studies to primary patient material, EV from healthy donor (HD) plasma and primary PEL fluid were isolated. The EV exhibited the same characteristic size distribution across different donors (S5 Fig). Their concentrations were at times higher than from culture supernatant. AchE activities, likewise, were comparable. These experiments demonstrate that EV isolated by this method from virus-negative and virus-positive cell supernatants were biophysically and biochemically similar to each other and to normal plasma or primary PEL fluid.
To exclude the presence of virion particles in the EV preparation, EV were subsequently affinity-purified using antibody conjugated beads. Recapitulating our prior results [7], this step retained EV-associated proteins (S6A Fig) and miRNAs (S6B Fig) while removing KSHV particles, as measured by DNA content (S6C Fig). NTA analysis of the affinity-purified EV from BJAB, BCBL1 PEL, HD, and primary PEL yielded consistent characteristics across (S6D–S6F Fig), and EM detected the presence of cup-like vesicles typically of EV, but no evidence of virion particles (S6G and S6H Fig). Through this method, we recovered all but one of the KSHV-encoded miRNAs in the CD63+ fraction of EV from BCBL1 cells. Neither the viral transcript LANA nor the cellular encoded GAPDH mRNA were present, demonstrating enrichment for the viral miRNAs, with primary PEL CD63+ EV serving as our positive control (Fig 2A). To show that the samples contained similar fractions of CD63+ EV, the EV were incubated with Dil, a membrane-intercalating dye, ExoGreen, an internal esterase-substrate, or both, bound to antibody-coated beads and subjected to flow cytometry (S7 Fig). Results were consistent between CD63 (Fig 2B), CD9 (Fig 2C and S8 Fig), and CD81 beads (Fig 2D and S9 Fig). This demonstrated that PEL derived EV carry at least three tetraspanin markers (CD63, CD81, CD9), which are suitable for positive affinity purification and that our three-step approach (PEG > column > affinity-bead) yielded a highly purified, virus and DNA-free, EV fraction.
The principal target of transformation by KSHV are endothelial cells (EC, reviewed in [3]). EC are also exposed to the highest concentration (~ 1011/mL) of EV via lymphatic and blood vessels and thus can be considered a bona-fide target of systemically circulating EV [23]. For these reasons, hTERT-immortalized human umbilical vein endothelial cells (hTERT-HUVEC) were used to investigate EV uptake. To establish that the purified EV were endocytic, 1010 CD63+ EV were labeled with Dil (1,1’-dioctadecyl-3,3,3’3’-tetramethylindocarbocyanine perchlorate, red) and ExoGreen (green) and added to 105 hTERT-HUVECs (MOI = 105) and after 12 hours, fixed and analyzed by fluorescence microscopy. Uptake was equivalent for EV from all sources: BJAB, BCBL1, HD, and primary effusion fluid (Fig 3). The red and green labels co-localized in the cytoplasm. Not every incidence of red and green EV labels co-localized after cells were exposed for 12 hours. This was likely due to differences in the recycling nature of the biomolecules stained (lipids vs. proteins). As the time course was extended (S10 Fig and S11 Fig) the lipid dye Dil redistributed throughout the cell, in contrast to the ExoGreen protein dye which remained in punctate structures. EV uptake was blocked by Annexin-V, but not heparin sulfate at a concentration, which reliably blocks KSHV entry [24] (Fig 4). This establishes that, after extensive purification, the EV retained endocytic activity, which was dependent on phosphatidylserine (PS) but not heparin. Henceforth, this material is referred to as KSHV-EV.
To test the hypothesis that KSHV-EV act as chemoattractants and paracrine growth factors for EC, cell migration was monitored continuously using the xCelligence system. The xCelligence system allowed us to evaluate a number of physiological phenotypes that are associated with EV reprogramming. First, KSHV-EV served as a chemoattractant for EC migration, whereas EV from human donor blood (HD) did not. This chemoattractant property was phenocopied by EV isolated from primary PEL fluid (Primary PEL EV) (Fig 5A and 5B). Second, we asked if synergy existed between KSHV-EV and KS-relevant cytokines (VEGF, IL-6, PDGF-β, or SDF1-α). The hTERT-HUVECs were exposed to EV and plated into the top chamber and monitored for migration into the bottom chamber, which contained the cytokine in serum-free media. KSHV-EV significantly enhanced cell migration in response VEGF and IL-6, but not PDGF-β or SDF1-α (S12 Fig). Third, a scratch assay was performed. The hTERT-HUVECs were grown in the presence of EV, the confluent monolayer disrupted, and closure monitored over time. KSHV-EV and primary PEL EV enhanced migration, EV from HD did not (Fig 5C). The positive control, VEGF, alone or added to EV from HD yielded the same degree of scratch closures as induced by KSHV-EV (Fig 5D and 5E). Fourth, to test the hypothesis that EV induced cytokines, which could amplify or mediate the migration phenotype, supernatants were analyzed for IL-6, IL-10, which are implicated in KS biology, as well as the immune response cytokines IL-18, IL-1β, and interferon alpha (IFN-α). Only IL-6 was significantly induced in response to KSHV-EV and primary PEL EV (Fig 5F and S13 Fig). These experiments establish that KSHV-EV, but not EV circulating in healthy patients contributes to the pathophysiology of KS by modulating EC function and inducing human IL-6.
To test whether KSHV-EV elicited an innate immune response in EC, a number of known innate immune signaling pathways were examined. On the one hand, such a phenotype would be expected as PEL and KS represent an inflammatory microenvironment (reviewed in [3]); on the other hand, pro- as well as anti-inflammatory phenotypes have been reported for infected-cell derived EV from different viruses (reviewed [1]). The innate immune response involves membrane-bound receptors, such as toll-like receptors (TLRs), as well as cytoplasmic RIG-like receptors, such as RIG-I and MDR-5. Both pathways ultimately converge onto interferon regulatory factor 3 (IRF3) and nuclear factor kappa B (NF-κB). IRF3 and NF-κB are normally sequestered in the cytoplasm. Upon stimulation, they translocate to the nucleus. KSHV-EV, or primary PEL EV did not induce nuclear translocation of IRF3 (Fig 6A and 6B) and did not induce IRF3 phosphorylation (Fig 6C). This was in contrast to infection with West Nile Virus or stimulation with Polyinosinic:polycytidylic acid (PolyI:C). KSHV-EV did not inhibit the phosphorylation of IRF3 in response to PolyI:C (Fig 6D), suggesting the KSHV EV did not actively inhibit signaling. KSHV-EV, or primary PEL EV did not induce nuclear translocation of NF-κB either (S14 Fig). Likewise, targeted transcriptional profiling of ninety NF-κB-regulated genes showed no response to KSHV-EV (S15 Fig). The IRF3 and NF-κB transcription factors represent the endpoint of a multitude of RNA sensing pathways. As neither was activated, it is unlikely that any of the upstream receptors (TLR, RLR) were activated to the degree that authentic viral infection would.
KSHV is a DNA virus, and its reactivation from latency is curbed by cGAS/STING [25]. To test for the induction of cGAS-STING signaling by KSHV-EV, we monitored the induction of interferon-beta (IFN-β). KSHV-EV did not change IFN-β transcript levels, and KSHV-EV did not modulate the cGAS/STING response to Interferon Stimulatory DNA (ISD) or Poly I:C (Fig 7A). As a second, independent measure of cGAS/STING activation we measured TANK binding kinase (TBK). TANK is phosphorylated upon recognition of cytosolic nucleic acids in a MAVS dependent manner, and physically interacts with STING. KSHV-EV did not affect phospho-TBK levels alone or in conjunction with the positive inducers (Fig 7B and 7C). In sum, neither TLR, RIG-I, nor cGAS/STING pathways were activated by KSHV-EV; at least under these conditions KSHV-EV did not inhibit the activation of these pathways by physiological triggers either.
To identify molecular pathways that could explain KSHV-EV induced endothelial cell migration, we explored ERK1/2 signaling. ERK1/2 has been implicated in IL-6 signaling as well as EC migration [26, 27]. Treatment of hTERT-HUVEC with KSHV-EV and primary PEL EV, but not HD EV induced ERK1/2 phosphorylation (p-ERK1/2) (Fig 8A). The EV themselves did not contain p-ERK1/2 (Fig 8B). To exclude the possibility that IL-6 induced ERK1/2 phosphorylation as part of a secondary feedback loop, the experiment was repeated in the presence of antagonistic anti-IL-6 receptor antibodies. Despite blocking IL-6 signaling, ERK1/2 became phosphorylated upon KSHV-EV exposure (Fig 8C). Pre-incubation with Annexin-V, which blocks EV adsorption, significantly reduced p-ERK1/2 levels (Fig 8C). This result is consistent with the notion that ERK activation was a direct result of EV exposure.
ERK1/2 is phosphorylated by MEK, which can be targeted pharmacologically. To test the hypothesis that MEK kinase activity was required for p-ERK1/2 in response to KSHV-EV and primary PEL EV, we used AZD6244. Pre-treatment of cells with AZD6244 blocked ERK1/2 activation by primary PEL EV relative to DMSO control (Fig 8D). AZD6244 also reduced primary PEL EV- and KSHV-EV-dependent cell migration (Fig 8E and 8F). To ensure that AZD6244 did not exert off-target effects, we repeated the cell migration assay with a different MEK inhibitor, PD184352. Treatment with PD184352 antagonized the enhanced cell migration phenotype of hTERT-HUVECs treated with KSHV-EV (S16 Fig). These experiments demonstrate that KSHV-EV activate the Ras/Raf/MEK/ERK pathway, which leads to EC activation, proliferation, and migration.
In PEL and KS patients, KSHV-EV are continually released into the microenvironment and systemically into the blood and lymphatic circulation. Next to hemangioma, KS is the most angiogenic cancer in humans (reviewed in [3]). PEL grow as effusions, bathing the cavity walls in KSHV-EV. Thus, a physiological relevant experimental design would expose EC repeatedly to KSHV-EV. Such a design measures long-term cellular reprogramming (Fig 9A). Here, hTERT-HUVECs were exposed to KSHV-EV or BJAB-derived EV over a period of 12 days. Every 24 hours the media was replenished with fresh KSHV-EV or control-tumor EV, and cellular programming was analyzed by RNAseq. The time course was divided into an acute phase (day 2 and day 4), an intermediate phase (day 6 and day 8), and a chronic phase (day 10 and 12). KSHV-EV induced synchronized, progressive, and directional transcription profile changes compared to control (Fig 9B, see S17 Fig for a principal component analysis of hierarchical clustering). To identify continuous transcript changes over time, a likelihood ratio test was used. This identified 67 transcripts that were significantly upregulated and 84 that were significantly downregulated genes over the course of treatment (Fig 9C, and S2 Table). Heatmap representation of these altered genes show distinct contrasts between the treatment groups (Fig 9D–left), which were maintained over the 12 days of KSHV-EV exposure.
To test the hypothesis that KSHV-EV mediate any or all of the transcriptionally changes hitherto ascribed to KSHV miRNA expression within the infected cell, we analyzed transcriptional changes in genes that were previously identified in KSHV-infected EC [6, 28, 29]. Changes in these particular sets of mRNAs signal BEC to LEC transcription upon direct infection of KSHV. These mRNAs remained largely unchanged (S18 Fig). The notable exception was MAF1 mRNA, which was previously shown to be a direct target of multiple KSHV miRNAs with eleven predicted target sites in the 3’UTR [6], which was robustly down-regulated. As a control we analyzed a predefined set of interferon stimulatory genes (ISGs) [30] (Fig 9D, right). These showed induction during the acute phase for a few genes; however, these did not differ between control and KSHV-EV. This induction was not maintained over time, consistent with the idea that KSHV-EV reprogram EC towards a proliferative, activated phenotype, which is different from phenotypic changes induced by inflammation.
Time course analysis allowed for the identification of patterns of transcriptional changes for individual genes. The minimal set of EV-exposure biomarkers was comprised of genes with the highest and most consistent mRNA changes over time (Fig 10A and 10B). Among them were CD9 and JUNB, which we chose to validate at the level of protein expression (Fig 10C). CD9 protein levels were greatly reduced in KSHV-EV treated cells, particularly in the intermediate and chronic time points. As an internal control, we monitored the protein levels of a separate EV protein Tsg101, which remained constant in both treatment groups over time. JUNB levels were very low in the hTERT-HUVEC cells to begin with and further reduced at the intermediate time frame in the KSHV-EV treatment group. Thus, the transcriptional changes in response to KSHV-EV translated into protein level changes and physiological changes.
The advantage of genome-wide transcriptional profiling lies in the identification of signaling networks, rather than individual genes. Hence, the significantly altered genes were clustered into gene ontology (GO) pathways [31]. The preeminent pathways identified herein related to extracellular matrix modulation, cell adhesion, growth and migration (Fig 11). Next, we explored phenotypic changes that would be consistent with the transcriptional pathways that dominated GO analysis. These are summarized in S3 Table. Experiments showing increased cell growth and migration (and thus decreased adhesion) in response to KSHV-EV, but not control-EV were already noted above. CD9 is a tetraspanin, which is involved in cell adhesion, motility and junctional integrity. It is also involved in EV biogenesis. To test whether the KSHV-EV-induced downregulation of CD9 at the mRNA and protein level led to a reduction in EV secretion in the recipient cells, we measured EV in the supernatant of hTERT-HUVEC cells at 24 hours after treatment with KSHV-EV. This experiment was possible because earlier studies had shown that exogenously added EV, analogous to liposomal transfection, are taken up within a few hours after addition to media [12, 32]. There were no changes in the biophysical characteristics of the hTERT-HUVEC-derived EV, but the total amount was reduced by comparison to control (Fig 12 and S18 Fig). To test the hypothesis that KSHV-EV induce some, but not all phenotypes as KSHV infection, morphological differences in the recipient hTERT-HUVEC were evaluated. HTERT-HUVEC were exposed to KSHV-EV, mock, or BJAB-derived, control EV for four consecutive days. Chronically infected HUVECs served as a control. KSHV-EV treatment did not alter Tubulin (Fig 13A) or Actin (Fig 13B) organization. KSHV LANA was present in infected, but not EV-treated cells (Fig 13C). By contrast, the proliferation marker Ki-67 was dramatically induced by KSHV-EV, but not control EV. Ki-67 positivity was similar to KSHV infected cells (Fig 13D and 13E). H&E stain revealed a greater cell density in KSHV-EV compared to mock or BJAB-EV treated cells (S20 Fig). KSHV-infected hTERT cells exhibited greatly increased cell size, as previously described and consistent with mTOR/S6K activation [33]. Overall these results mirror the dramatic dysregulation of infected as well as uninfected EC in KS lesions, where the normal vasculature and extracellular environment is essentially destroyed and slit-like empty spaces develop. This analysis demonstrated that KSHV-EV inducing a long-lasting reprogramming of EC, which results in transcription signatures and pathway alterations consistent with the phenotypic changes observed in KS lesions.
KS is an incredibly angiogenic cancer, second only to hemangioma [3]. It is driven by KSHV-infected EC and defined by a unique molecular mechanism that manifests itself in aberrant EC behavior. Many studies have focused on the cell autonomous roles of KSHV [6, 28, 29, 33–38]. In addition, studies by Mesri and others (reviewed in [39]) have established that the KS phenotype depends to a large degree on paracrine signaling mechanisms to reprogram neighboring uninfected EC. This report establishes that KSHV-EV mediate some of the paracrine phenotypes of KS (summarized in S3 Table).
EV mediate a large variety of phenotypes in the immediate microenvironment as well as at distant sites. EV have established roles in cell differentiation, angiogenesis, cell migration as well as metastasis [40–44]. We had shown earlier that KSHV miRNAs are present in systemically circulating EV in KS patients, PEL fluid as well as in transgenic mice, which carry the KSHV miRNAs, but are not competent to make virions [7]. PEL are of post-GC lineage lymphoma, approaching almost plasmablastic stage. They grow i.p. (in body cavities), unlike Burkitt lymphoma, which are also post-GC lymphoma, but pre-plasmablastic and grow as a solid mass in lymph nodes, not as an effusion. This may explain the prominence that EV have in the biology of PEL and KSHV vis-a-vis other tumors.
Crucial to the study of EV is a well-validated purification pipeline [22, 45]. In the context of virus infections, it is important to exclude viral particles, which tend to co-purify with EV in ultracentrifugation, crowding-agent, and size exclusion chromatography approaches. Hence, we added affinity purification using antibody-coated beads directed against CD63 or other tetraspanins as the final purification step. This step depletes virions to below the limit of detection [7, 12], as it positively retains EV on a column rather than collecting a precipitate. It also reduces the complexity of the EV populations [10, 46] to only those EV that are of narrow size, tetraspanin-positive, and inhibited by Annexin-V. Adding an RNAse and DNAse step as well as size exclusion chromatography eliminated contaminating free RNA and DNA and selected against vesicles that are released non-specifically by dying cells. This has not always been done and may explain reports of EV preparations that induce heavy DNA and RNA-dependent immune stimulation in recipient cells. We believe the final product of our purification pipeline represents biologically-relevant KSHV-EV at or slightly below physiological concentration.
It has been a matter of debate as to whether EV induce or suppress the innate immune response. This phenotype depends largely on the specific virus and target cell. Professional immune cells, such as dendritic cells and macrophages are known to receive and transmit pro-inflammatory signals through EV [47–49]. RNA viruses induce a dramatic innate response and large amounts of secondary messengers, such as IFN-β or cGAMP [48]. DNA viruses, such as herpesviruses may also transmit pro-inflammatory signals through EV, which can be sensed by professional antigen presenting cells [47, 50]. While we cannot exclude that lytically replicating cells or professional antigen presenting cells infected with KSHV behave differently, these experiments demonstrate that EC, the primary target of KSHV infection, do not become activated by EV from KSHV-infected lymphoma cells or by EV from primary PEL fluid. They may become activated by cytokines or small soluble molecules, though a non-activated phenotype of uninfected cells would also be consistent with the biology of KSHV as most primary KSHV infections are clinically asymptomatic and not associated with mononucleosis-like symptoms or autoimmunity as seen with EBV or human cytomegalovirus infection.
These results support a model whereby cellular proteins, cellular and viral miRNAs that are carried in KSHV-EV modulate long-term reprogramming of EC in the immediate microenvironment of the tumor and systemically in the human host. Whereas prior studies focused on the immediate effects of a single bolus of EV, these experiments were designed to mimic continuous KSHV-EV exposure as seen in KS and PEL patients or patients with a high latent virus burden. The concentration of EV in normal blood is ~1010–1011/mL [23], which is ~ 6 orders of magnitude higher than the median concentration of KSHV in the blood of symptomatic, untreated AIDS-KS patients [51]. We added 1010/mL to 106 cells (MOI = 10,000). Based on our prior work and Fig 3, we assume that all EV are endocytic-competent and all KSHV-EV carry the KSHV miRNAs. EV adsorption plateaus within hours of exposure [32]. Using a MOI bolus of 4000 vs. 1000 is unlikely to result in a qualitatively different response after 4 days. The novelty of this experimental design is mimicking chronic exposure, as is the case in the KS microenvironment or for any endothelial cell lining the blood vessels. Whereas PEL and KS, in the context of uncontrolled progression to AIDS, are rapidly fatal, de novo KSHV infection per se is not. HIV-negative endemic, pediatric and classic KS have rapid as well as smoldering clinical progression [52]. KSHV-associated multicentric Castleman’s disease has a waxing and waning presentation, closely associated with IL-6 levels [53]. IL-6, IL-10, VEGF, PDGF and other inflammatory cytokines are elevated in PEL, KS and MCD, and agents, such as pomalidomide, rapamycin and tocilizumab which modulate their levels, modulate disease [54–56]. KSHV-EV consistently induced human IL-6 in uninfected cells. These experiments show that in addition to cytokines EV also transmit pro-growth signals and can reprogram EC.
KSHV-EV induced a much more long-lasting phenotype than acute phase cytokines, which mimics differentiation and trans-differentiation. Whereas it was difficult to identify a single master regulator of this trans-differentiation phenotype, network analysis showed significant changes of transcriptional modules that regulate extracellular matrix remodeling, translation and exosome biogenesis. By comparison, IFN-β and NF-κB transcriptional networks were unaffected. KSHV-EV signaled through MAPK/ERK, which is consistent with MAPK/ERK’s role in modulating EC motility and vascular behavior (reviewed in [57]). Our observations are consistent with a recent study by Yogev et al. [58], who showed that KSHV-EV (derived from infected EC) induce metabolic remodeling of nearby uninfected cells. This represents perhaps the initiating step of trans-differentiation. Afterwards, continued KSHV-EV exposure resulted in continued reprogramming as has been described for KSHV-infected and KSHV-miRNA transfected EC [4–6, 34] and these experiments verified that most KSHV-miRNAs are present in KSHV-EV. Reprogramming here is used to defined an altered state of gene transcription and cell lineage, such as published by Hansen et al. [6], upon transfection of the KSHV miRNAs into EC, or upon infection of EC with KSHV [4, 28, 29, 59–61]. At this point we do not know if this reprogramming will persist after EV exposure has subsided and if not, how quickly the cells return to their normal state. Clinically, KS lesions and KS-associated edema regress as KSHV is cleared by immune restoration upon cART or lowering of immunosuppressive drugs in the context of transplant KS. Hence, we speculate that that the KSHV-EV induced phenotype likewise is transient. This would be in contrast to permanent lineage reprogramming, which is most commonly associated with epigenetic changes to the cellular DNA. Evidence for KSHV-infection induced chromatin remodeling has been published [62–65]. If in addition to transcriptional reprogramming, the viral miRNAs (or other molecules) that are contained in KSHV-EV also alter chromatin accessibility stably and irreversibly is a fascinating hypothesis and the subject of future studies.
The gradual and long-lasting reprogramming of transcriptional networks is consistent with the mechanism of action for miRNAs, which have their most physiological impact in development rather than acute signaling. EBV and KSHV express miRNAs and in infected cells these miRNAs account for as much as 50% of the miRNA pool. KSHV and EBV incorporate the viral miRNAs into EV [7, 66–68]. Both viruses substantially modulate the protein composition of EV [69]. In addition, EBV incorporates the LMP-1 oncogene into EV [70, 71], whereas no KSHV proteins were detected in EV thus far. This phenotype is consistent with the idea that EV are pivotal for establishing local tissue homeostasis and provides a molecular mechanism for it. Further studies are needed, but for the first time there now exists a highly reproducible, physiologically relevant experimental design to study long-term EV-EC interactions.
In conclusion, our findings point toward a novel means of cellular reprogramming by viruses. They pinpoint novel, actionable pathways for intervention and biomarker development. KSHV is able to infect EC, but the larger importance of EV stems from the fact that these vesicles can carry viral components to distant locations and transfer them into cells that the virus cannot enter. This may explain some of the phenotypes that viruses, including HIV, have on uninfected cells and it may explain why clinical sequalae persist long after the virus has been cleared or entered molecular latency.
All cells were grown at 37°C in 5% CO2. hTERT-immortalized HUVECs were cultured in endothelial growth medium (EGM-2 media; Lonza) supplemented with the EGM-2 Bulletkit (Lonza) and 10% EV-depleted fetal bovine serum (FBS). BCBL1 (from the laboratory of Dr. D. Gamen) cells were cultured in RPMI 1640 (Gibco) supplemented with 10% Tetracycline-free, EV-depleted FBS (Clontech), 100 U/mL penicillin G (Gibco), 100 μg/mL streptomycin sulfate (Gibco) and 2 mM L-glutamine (Gibco). BJAB (from the laboratory of Dr. D. Gamen) cells were cultured in RPMI 1640 supplemented with 10% EV-depleted FBS, 100 U/mL penicillin, 100 μg/mL streptomycin sulfate, and 2 mM L-glutamine. Namalwa (EBV-positive, obtained from the ATCC #CRL-1432) were grown in the same conditions as BCBL1 cells.
Total EV were isolated using approximately 400 mL of cell culture supernatant. Cells were pelleted at 4˚C at 800x g for 10 minutes. Supernatant was then passed through a 0.22 μm Nalgene Rapid Flow Filter (Thermo Fisher). Filtered supernatants were aliquoted into individual 50 mL conical tubes (Corning). EV were precipitated with 40 mg/mL PEG-8000 and incubation at 4˚C for >8 hours. Precipitates were then spun down at 4˚C at 1,200x g for 60 minutes. Pellets were resuspended in 500 μL of ice-cold 1X PBS (Gibco). Removal of non-associated molecules were done by (i) ultracentrifugation or (ii) column chromatography. (i) For ultracentrifugation, the volume was increased to ~4 mL with 1X PBS and centrifuged at 4˚C at 120,000x g for 60 minutes using a Beckmann SW32 rotor. The pellet was then resuspended in 4 mL of fresh 1X PBS and centrifuged again. A total number of three washes was done. The final pellet was resuspended in 100 μL of fresh, ice-cold 1X PBS. (ii) For column chromatography, GE Sephadex G-200 was equilibrated with ice-cold 1X PBS for a total of 4 compacted bead volumes (4 mL). The resuspended EV were added to the equilibrated column and allowed to flow through the column by gravity. EV were collected in the first 1 mL of fresh, cold 1X PBS. EV were also isolated from isolated from 50 mL plasma from health donors or 10 mL PEL. Briefly, blood was processed and erythrocytes, leukocytes, platelets, and plasma were separated using Ficoll reagent (GE 17-1440-02) as above.
Samples were enriched for CD63, CD9, and CD81 positive EV using magnetic beads (ThermoFisher 1060D, 10620D, and 10622D, respectively). Briefly, the total EV isolated as above were added to 80 μL of equilibrated, antibody-coated magnetic beads. Non-specific IgG-coated beads were used as a control. EV were bound to beads overnight at 4˚C and beads were washed 3X with 1X PBS. EV were eluted in 100 μL of elution buffer (Invitrogen) or 0.2 M Glycine pH = 2.0 for further analysis. Cell were authenticated by targeted amplification of STR typing loci using Ion Torrent Precision ID GlobalFiler NGS STR Panel and compared against the STR database of the German Collection of Microorganisms and Cell Cultures GmbH.
Total EV were labeled with 1 μM Dil (1,1’-dioctadecyl-3,3,3’3’-tetramethylindocarbocyanine perchlorate; Sigma), 0.5X ExoGreen (SystemBio), followed by G25 column filtration and incubated with 40 μL CD63, CD9, or CD81 beads, washed 3x with PBS and analyzed using the BD Accuri C6 Plus flow cytometer (BD Biosciences). FITC and PE settings were used to detect ExoGreen and Dil with an excitation laser of 488 nm and emission filters of 533 nm and 585 nm, respectively. Unlabeled EV were used to set background fluorescence. Results were analyzed using FloJo 2.0.
hTERT-HUVECs were grown on a cover slip in 6 well plates in a total volume of 3 mL and treated with labeled 109 EV/mL for the indicated time period at 37°C. Cells were then rinsed with PBS and fixed in 4% paraformaldehyde for 10 minutes at RT, washed 3 times with PBS and the cells permeabilized using 0.5% Triton X-100 in PBS for 10 minutes and washed 3x with PBS. For indirect immunofluorescence, coverslips were blocked in a solution of 10% goat serum (Vector Labs) in PBS with 0.2% Triton X-100 and incubated with primary antibodies: anti-IRF3 antibody (Cell Signaling, #4962, 1:100 dilution) anti-P-NF-κB p65 (S536) (clone 93H1, Cell Signaling, 1:100 dilution). Coverslips were washed three times with PBS-0.2% Triton with 2% BSA and incubated with FITC-conjugated anti-rabbit secondary antibody (#FI-1000, Vector Labs Inc. 1:500 dilution). Cells were washed three times with PBS-0.2% Triton X-100 with 2% BSA and stained with 0.2 μg/mL DAPI (Sigma) prior to mounting in VectaShield (Vector Labs). Cells were imaged on a Leica DM4000B microscope with a Q-Imaging Retiga-2000RV camera and HCX-PL-APO 506187 lens at 63x magnification. De-convoluted images (Simple PCI 6 software Metamorph v 7.8.12.0, 10 iterations RB, GB or RGB) were then opened in Imaris V 9.2.0. and background subtraction of all channels was done using recommended settings of 400 um filter width. Localizations of EV-delivered Dil and ExoGreen were done using the “Add Spots” command using spots of different sizes depending on fluorescence intensity. Regions of spot calling were standardized to linear detection ranges using absolute intensity. For nuclei staining, the “Add New Surfaces” command was used.
As a positive control for IRF3 activation, hTERT-HUVECs were infected with West Nile Virus NY99 (WNV) at a MOI of 5 (observed after 36 hours) or Poly I:C at 5 μg/mL (observed after 12 hours).
Pellets (1010 EV or 106 cells) were lysed in 100 μL NP-40 lysis buffer and run on an 8% SDS-PAGE gel, transferred to a nitrocellulose membrane (Hybond) and blocked in 5% dry milk in TBS overnight at 4°C. Antibodies are listed in S1 Table. For detection of tetraspanins, non-reducing conditions were used. To visualize total protein by silver stain, we used the Pierce Silver Stain Kit (ThermoFisher) after which bands were excised and analyzed by mass spectrometry at the UT Southwester core (https://www.utsouthwestern.edu/research/core-facilities/proteomics-core.html).
To block EV entry, EV were incubated with recombinant Annexin-V (2 μg/mL, BD Biosciences) for 30 minutes prior to addition to hTERT-HUVECs. To block virus entry EV were incubated with 50 μg/mL heparin (Lonza). To block autocrine IL-6 feedback, 10 ng/mL IL-6 receptor (sIL-6R, Peprotech #200-06R) was added to cells 24 hours before EV addition. To inhibit MEK and ERK1/2, AZD6244 (SelleckChem) and PD184352 (SelleckChem), respectively were used to treat cells 24 hours prior to addition EV addition at indicated concentrations.
(i) Scratch assays were performed as previously described [7]. Briefly, hTERT-HUVECs were grown in a 24-well plate (Corning) prior to treatment with EV. The wound was initiated using a standard 200 μL pipette tip and the cells were then washed and replaced with fresh media containing one of the following as a chemo-attractant: 10% FBS, 10 ng/mL VEGF (Peprotech), 1 U/mL hIL-6 (Peprotech), 10 ng/mL PDGF-β (Peprotech) or 10 ng/mL SDF-1α (CXCL12) (R&D Systems). The culture was monitored over time. Images were obtained using a Leica DMIL microscope with a HI Plan 10x/0.25 PHI objective and QImaging camera (Cooled color, RTV 10 bit) paired with QCapture imaging software 3.0. Images are shown at 100x magnification and were analyzed using ImageJ software to calculate the percent wound closure at a given time point. (ii) Cell proliferation and migration was analyzed using the xCelligence RTCA DP instrument as previously described [7]. Briefly, hTERT-HUVECs were treated with EV for 24 hours and proliferation measured by conductance. For migration, both sides of the xCelligence CIM Plate 16 (Acea Biosciences) plate membrane were coated with 20 μg/mL fibronectin prior to assembly and media containing FBS or a specified cytokine was placed in the lower chamber as the chemo-attractant. Cells were plated at 15,000 cells per well of the upper chamber. Reads were taken every 2 minutes for a period of 12 hours. The cell index reflects the degree of cellular migration towards the specified chemo-attractant.
Levels of IL-6, IL-8, IL-10, IL-18 and IL-1β were determined by ELISA according to the manufacturer’s protocol (eBioscience, #88–7066 (IL-6), #88–8086 (IL-8), #88–7106 (IL-10), #BMS267INST (IL-18) and #88–7010 (IL-1β). The average of at least three wells is reported for each biological replicate.
Total RNA was isolated using TRI reagent (Molecular Research Center) as previously described (https://www.med.unc.edu/vironomics/services/protocols/), treated with Turbo DNA-free kit (Ambion, Life Technologies) and 100 ng of DNA-free RNA, as determined by Nanodrop, was used as input for High Capacity cDNA synthesis kit (Applied Biosystems, Life Technologies). Custom NF-κB and endothelial lineage real-time qPCR arrays were used previously published [72].
EV were adsorbed on a glow-charged carbon coated 400-mesh copper grids for 2 minutes and then stained with 2% (weight/volume) uranyl acetate in water. Transmission electron microscopy (TEM) images were taken using a Philips CM12 electron microscope at 80 kilovolts. Images were captured on a Gatan Orius camera (2000 x 2000 pixels) using the Digital Micrograph software (Gatan, Pleasanton, CA). Images were then cropped in Adobe Photoshop.
(i) For continuous, variable pairwise T-tests were performed to determine statistical significance for n ≥ 3 biological replicates. (ii) For comparison of mass spectrometry data, a hypergeometric test was used. (iii) For analysis of RNAseq data we used a custom pipeline. The decision to use our particular analysis is discussed at length at https://support.bioconductor.org/p/62684/. In brief, STAR-Aligned BAM files representing table of counts for each samples were processed using DESeq and other Bioconductor packages: https://bioconductor.org/packages/devel/bioc/vignettes/GenomicAlignments/inst/doc/summarizeOverlaps.pdf.
|
10.1371/journal.pcbi.1005870 | Olfactory coding in the turbulent realm | Long-distance olfactory search behaviors depend on odor detection dynamics. Due to turbulence, olfactory signals travel as bursts of variable concentration and spacing and are characterized by long-tail distributions of odor/no-odor events, challenging the computing capacities of olfactory systems. How animals encode complex olfactory scenes to track the plume far from the source remains unclear. Here we focus on the coding of the plume temporal dynamics in moths. We compare responses of olfactory receptor neurons (ORNs) and antennal lobe projection neurons (PNs) to sequences of pheromone stimuli either with white-noise patterns or with realistic turbulent temporal structures simulating a large range of distances (8 to 64 m) from the odor source. For the first time, we analyze what information is extracted by the olfactory system at large distances from the source. Neuronal responses are analyzed using linear–nonlinear models fitted with white-noise stimuli and used for predicting responses to turbulent stimuli. We found that neuronal firing rate is less correlated with the dynamic odor time course when distance to the source increases because of improper coding during long odor and no-odor events that characterize large distances. Rapid adaptation during long puffs does not preclude however the detection of puff transitions in PNs. Individual PNs but not individual ORNs encode the onset and offset of odor puffs for any temporal structure of stimuli. A higher spontaneous firing rate coupled to an inhibition phase at the end of PN responses contributes to this coding property. This allows PNs to decode the temporal structure of the odor plume at any distance to the source, an essential piece of information moths can use in their tracking behavior.
| Long-distance olfactory search is a difficult task because atmospheric turbulence erases global gradients and makes the plume discontinuous. The dynamics of odor detections is the sole information about the position of the source. Male moths successfully track female pheromone plumes at large distances. Here we show that the moth olfactory system encodes olfactory scenes simulating variable distances from the odor source by characterizing puff onsets and offsets. A single projection neuron is sufficient to provide an accurate representation of the dynamic pheromone time course at any distance to the source while this information seems to be encoded at the population level in olfactory receptor neurons.
| A primary goal of olfaction research is to understand how complex olfactory scenes that occur in natural environment are processed by the olfactory system, in particular for orientation behaviors to odor sources. However, until recently natural odor stimuli were not described quantitatively. Hence, most studies of olfactory physiology are restricted to static and white-noise stimuli, limiting our understanding of dynamic olfactory coding. Although realistic olfactory input signals were used to analyze olfactory coding [1–3], they were uncontrolled, so preventing to explore specific features of the olfactory signals and large distances to the odor source. Only a recent paper generated naturalistic odor plumes and described how adaption from olfactory receptor neurons (ORNs) to stimulus mean and variance contribute to encode intermittent odor stimuli [4]. Here we pursue a novel approach to the study of olfactory coding.
For most animals larger than a millimeter, odor cue transport is dominated by turbulence [5–7]. Turbulence is thus a major key to understand animal orientation behavior from olfactory coding to decision dynamics. It prevents formation of gradients pointing towards the source and structures the plume in isolated odor patches that deform spatially as they are transported by the wind. Thus, odor detection is discontinuous. In turbulent conditions, temporal dynamics of odor/no-odor detection is dominated by long-tail statistics [8]. Furthermore, these statistics change with the distance to the odor source. The probability distributions for the duration and spacing in time of odor filaments become more heterogeneous with more short and long odor/no odor events when the distance increases [8]. Therefore, in real environments statistics of odor detection strongly differ from a white-noise pattern and are not characterized by a main frequency. For example, far from the source (~100 m) a detection provides limited information on the time and duration of the next detection.
Male moths tracking plumes of sex pheromone released by their conspecific females is a paradigmatic example of olfactory searches [9, 10]. Successful searches are achieved with starting points at tens to hundreds of meters from the source [11–13]. The statistics of intermittent odor signals play a major role in the dynamics of moth and other insect tracking behaviors. It allows insects to discriminate closely spaced pheromone compounds [14–16] or compounds delivered with millisecond asynchrony [17–19]. Timing is also an important information for the vertebrate olfactory system [20–25]. However, opposite to vertebrates the insect nose is everted and can continuously monitor odor stimuli whose temporal structure is not distorted by respiratory cycles [24–27] and differential sorption into the mucus layer of the olfactory epithelium [28–31]. The rapid responses and rich dictionary of maneuvers of moths in turbulent flows suggest efficient coding of olfactory plume dynamics [32]. In turbulent environments, the flight is empirically described by two dynamical behavioral features: upwind surge, associated to pheromone perception, and zigzagging, a cast maneuver to retrieve odor plumes when lost [33–35]. These features are correlated to odor detection dynamics. Artificial homogenous pheromone clouds, thus unrealistic compared to real environment, lead to inefficient searches [36–39]. Change in the temporal characteristic of odor encounters has immediate effects on the searching dynamics [35, 38, 40–43] and can have more effect than a 1000-fold change in pheromone concentration [44].
Sensory systems efficiently process the stimuli encountered in their natural environment, according to the efficient coding hypothesis [45]. Adaptation of neural coding to the statistics of inputs [46–48] is essential for understanding evolutionary forces driving the properties of nervous systems. Natural stimulations are frequently reported to have broadband distributions with power-law statistics (e.g. for vision [49]). Olfactory signals involve the complex properties of turbulence and their dynamics are characterized by their extreme heterogeneity, long-tail statistics and thus multi-scale variability. In that sense, the olfactory dynamics encountered by individual moths can be “far” from an average statistics induced by the environment. This creates significant challenges in order to properly code the signal. An optimal olfactory coding should be able to reliably encode the plume temporal structure at all relevant distances from the source. Essential characteristics of the puffs-blanks dynamics of the signal required to reach an olfactory source remain unknown. Analyses from experimental trajectories [9, 50] are often performed without knowing the precise olfactory signal (as it is inaccessible). As a consequence, theoretical analyses of orientation strategies have to be performed without knowing the computing capacities of insects.
In this work, we use the quantitative analysis of plume dynamics in turbulent conditions [8] to study moths’ neuronal coding of olfactory plumes with realistic turbulent dynamics and investigate for the first time what information is extracted at large distances from the source. We focus on how ORNs and antennal lobe projection neurons (PNs) encode the temporal characteristics of the pheromone plume in the Noctuid moth Agrotis ipsilon. We show that firing activities of ORNs and PNs are less correlated with the temporal structure of olfactory scenes when they simulate increasing distances to the source. Individual PNs but not individual ORNs encode puff onsets and puff offsets at multiple temporal scales. Thus, PNs can reconstruct the odor sequence at any distance from the source.
In order to study the neuronal coding of dynamic pheromone stimuli in the moth olfactory pathway, we recorded the firing rate of individual pheromone-sensitive ORNs and PNs in whole-insect preparations while presenting time-varying binary sequences of pheromone stimuli. We show the spiking activity of an ORN (Fig 1A–1C) and a PN (Fig 1D–1F) in response to a binary sequence of white-noise stimuli. We observed a good match between stimulus puffs and spike trains. A total of 209 ORNs and 97 PNs were recorded while presenting long duration (15 min and 5 min, respectively) white-noise sequences of pheromone stimuli.
In order to characterize the temporal properties of neural coding of olfactory stimuli we fitted a L-N model to the firing activity of each neuron (Fig 1C and 1F) consisting of a linear filter combined with a nonlinear function. First, we calculated the linear filters (or transfer functions) between white-noise sequences and the instantaneous firing rate with Lasso regularization on short- and long-term components. The linear filters were subdivided in two different time scales: a short-term component (time window -0.5 to 2.5 s) prolonged by a long-term component (2.5 to 320s) with exponential time sampling size. The long-term component takes into account the decrease of evoked firing activity during the stimulation sequence (time constant of the exponential decay is 89 s for ORNs and 15 s for PNs, Fig 2A). The neuronal responsiveness was restored after 3 minutes without stimuli both in ORNs (Fig 2B) and PNs (Fig 2C). The slow decrease of activity is thus a physiological process and did not result from a degradation of the quality of recordings or from a depletion of pheromone in the stimulus pipette. Long-term filters weighted 10.4% for ORNs (median value; bootstrap confidence interval: 9.9 to 12%) and 8.9% for PNs (median value; bootstrap confidence interval: 7 to 10.1%) of the total linear filter. Predictions of firing rates using linear filters were frequently negative indicating discrepancies between the measured neuronal activities and their linear prediction. This was expected since neurons are known to be nonlinear integrators of their inputs. Second, a simple static nonlinearity was used to correct the linear prediction above. We estimated the static nonlinearity with fitting a Hill function between linear prediction and measured activity (sigmoid curve, Fig 1C and 1F). Measured and L-N model-predicted neuronal activities are superimposed for the two example neurons (Fig 1B and 1E). Model predictions reproduced most of the fluctuations of the measured firing rates.
We quantified the performance of the L-N model with the coefficient of determination R2 calculated between the L-N prediction and the measured firing rate. We tested different types of white-noise sequences varying in the pheromone dose (only in ORNs) and in the correlation time (in ORNs and PNs). In ORNs, R2 increased significantly with the pheromone dose (ANOVA, p<0.001), revealing that the relationship between firing rate and stimuli is tighter at high pheromone concentrations (Fig 3A). Furthermore, R2 was lower for the fastest white-noise sequences (time step = 10 ms) both in ORNs (ANOVA, p<0.001; Fig 3A) and in PNs (Wilcoxon’s rank sum test for independent recordings, p<0.01, n = 97; Wilcoxon’s signed rank test for paired samples p<0.05, n = 15; Fig 3B and 3C). Finally, we observed that some PNs did not encode well the temporal fluctuations of the white-noise. Some of these PNs might have been damaged by the dissection and/or when introducing the recording electrode in the antennal lobe. However, we previously described in A. ipsilon the heterogeneity in sensitivity and delay of response across PNs from the cumulus [51]. In the honey bee, some PNs respond faster than others, and the subpopulation of fast PNs may cause rapid odor discrimination at the mushroom body output [52]. Therefore, some PNs with low quality L-N prediction may correspond to weakly sensitive neurons or to neurons that do not encode well the temporal dynamics of stimuli and better encode other properties of the stimulus such as its quality or quantity. In the following analysis of responses to turbulent stimuli we considered only the PNs with R2 > 0.3 (70 out of 97 cells) for white-noise sequences.
The shape of a linear filter characterizes how a neuron integrates time-fluctuating stimuli. The shape of individual linear filters is remarkably homogeneous for ORNs and more diverse for PNs (Fig 4A and 4B). We analyzed the variability in shape of short-term filters with a principal component analysis as performed by Geffen et al. [53]. For ORNs, the first principal component (PC1) explains 79% of the filter variance (Fig 4C). It has a positive phase followed by a negative one and its normalized integrated area is close to 0 (-0.07). Hence, it corresponds to purely phasic responses to olfactory stimuli. The first two principal components (PC1 and PC2) together explain 93.5% of the filter variance; PC2 has essentially one positive phase and a normalized integrated area close to 1 (0.90). Thus, PC2 corresponds to tonic responses to olfactory stimuli. As PC1 and PC2 account for a disproportionate amount of the filter variance, each ORN linear filter can be considered as a linear combination of these two components. Individual linear filters were normalized in amplitude so that each filter corresponds to one dot lying on the n-sphere of radius 1 in the space of the n principal components. A projection of this representation on the 2-D space of PC1 and PC2 is shown (Fig 4E). Most of the dots lie close to the unit circle, indicating that the first two principal components are sufficient to describe the filter shape with the exception of a few ORNs stimulated at the highest pheromone concentration. The relative contribution of the first two principal components to the linear filter can be directly read as the angle φ on that circle. A filter with an angle φ = 0 means that responses were purely phasic, φ > 0 means that responses were phasic-tonic, and φ < 0 observed at high pheromone concentrations means that responses were phasic and then inhibitory. Few cases were close to tonic excitatory responses (φ = π/2).
For PNs the same principal component analysis was performed (Fig 4D and 4F). PC1 explains 79% of the variance and PC1+PC2 92%. PC1 has a positive phase followed by a negative one resulting in a slightly positive normalized area (0.23). PC2 has a negative phase followed by a positive phase with a normalized area of 0.07. The contribution of the negative phase of PC2 cancels the positive phase of PC1 and therefore delays the response from stimulus onset. As a consequence the relative contribution of the first two components (angle θ on the unit circle) can be considered as the temporal phase lag between the neuronal response and the stimulation. Indeed as seen in Fig 4B, the position of the maximum of PN filters is delayed in time with increasing θ. To confirm that the shape of PC2 in PNs reveals the variable delays in the response, we time-shifted the filters according to their onset times and repeated the PC analysis (S1 Fig). The PC2 now accounts for 6.6% of the filter variability and, as for ORNs, consists in only one positive phase which contributes to the tonic component of the response.
The individual long-term filters were distributed along a straight line on a log-log scale plot (Fig 4G and 4H) both for ORNs and for PNs. We thus fitted a power-law function that revealed that both ORNs and PNs had similar scaling exponents (-1.3 for ORNs and -1.1 for PNs).
Linear filters were calculated for individual neurons in response to a sequence of white-noise stimuli. Only if it quantifies a stable cell property over time can it be used to predict neuronal response. We repeated up to 4 frozen sequences of white-noise in PNs interleaved by pauses of 1 to 5 minutes duration. Pseudo-population response is shown in S2 Fig. The model was fitted to each sequence independently and the model coefficients were remarkably similar for the different responses of the same cell confirming the stability of linear filters (Fig 5). For PNs, the cell-specific linear filter calculated from white-noise stimuli was used to predict responses to turbulent stimuli in the same cell. However this was not possible for ORNs since responses to white-noise and turbulent stimuli were recorded from different cells. Instead we estimated individual linear filters with fitting three parameters: the angle φ of the short-term filter, the scaling exponent of the long-term filter β and the relative weight of long- and short-term filters α. The amplitude of the linear filter was not considered since the nonlinear function compensates for it.
We stimulated the antenna with temporal sequences of pheromone stimuli having a temporal structure characteristic of turbulent environments. Turbulent sequences were generated as previously described [8] and consisted in binary sequences of pheromone stimuli with puffs of pheromone and clean air (blanks) between puffs of various durations (Fig 6A). The distribution of the duration of puffs and blanks in these sequences varied with the simulated distance from the odor source (Fig 6B). The distributions were more homogeneous at short distances from the source (at 8 m, 0.1–99.9% quantiles ranged from 126 ms to 8 s for puffs and from 126 ms to 12 s for blanks) and were broader at long distances with both shorter and longer events (at 64 m, 0.1–99.9% quantiles ranged from 16 ms to 22 s for puffs and from 16 ms to 89 s for blanks). These were characteristic long-tail distributions. For comparison we also show the distribution of the white-noise sequences used to fit the L-N model (Fig 6C). Durations of puffs and blanks have an exponential distribution in white-noise stimuli that is more homogeneous (for a correlation time of 50 ms, 0.1–99.9% quantiles ranged from 50 to 550 ms both for puffs and blanks) than the long-tail distributions in turbulent sequences.
L-N models obtained from binary white-noise stimuli were used to predict the response of ORNs and PNs to the turbulent stimuli. To correct statistical errors induced by nonlinearities [55], the non-linear part of the filter was re-estimated for the turbulent stimuli. Examples of responses and L-N predictions are shown in Fig 7A and 7D for two ORNs and one PN. We calculated the coefficients of determination R2 between model predictions and measured responses. In PNs the model fitted with the white-noise sequence predicts the response to turbulent stimuli (median R2 = 0.57; bootstrap confidence interval: 0.52 to 0.62). Removing any component of the model significantly decreased its performance in predicting responses (Table 1). These observations confirm the accuracy of the model and the critical role of the PC1/PC2 ratio.
Due to turbulence, variations of distances from the source during a moth olfactory search not only lead to complex changes in the whiff time dynamics, it also results in large and intricate fluctuations of pheromone concentrations within whiffs, i.e. peak concentration within whiffs does not decrease regularly with the distance to the source, so the highest concentrations are not necessarily all close to the source [8, 56]. Hence, to understand olfactory coding, we varied both the concentration and statistics of temporal dynamics for simulating different distances.
For ORNs, we tested turbulent stimuli simulating 4 different distances at pheromone concentrations over a large range encompassing 7 orders of magnitudes. The performance of the model clearly depended on pheromone concentration (ANOVA, p < 10−10, n = 192 cells) but did not depend on the simulated distance (ANOVA, p>0.3). More specifically, R2 increased linearly from 10−6 ng to 10−1 ng but for the highest concentration (1 ng) decreased back to the level of 10−2 ng (Fig 7C). This effect suggests that the range of concentrations we explored encompasses the physiological range. We represented this space with a phase diagram showing the dependence of olfactory coding performance (R2) on pheromone concentration (dose tested) and simulated distance (Fig 7B).
For each PN we tested up to 4 sequences of stimuli simulating different distances and presented in random order. To benefit from the paired structure of the data, we used specific statistics for testing the dependence of model performance on the temporal sequence. The coefficient of determination R2 was negatively correlated with the logarithm of the simulated distance (permutation test applied on linear regression ramp, p<10−5, n = 39 cells; Fig 7E). Olfactory coding performance in PNs depended on the temporal structure of the turbulent pheromone plume.
We hypothesized that the lower ability of neurons to encode the olfactory signal when the distance is increased may reflect a difference in coding of the long puffs and blanks that are characteristic of large distances. To test this hypothesis we subdivided the responses to turbulent stimuli into different sub-regions of time that we categorized as: (a) onset of puffs (first 0.5 s), (b) tails of puffs (after 2 s from the beginning of puffs), (c) offset of puffs (first 0.5 s following the end of puffs lasting at least 0.1 s) and (d) tails of blanks (after 2 s from the end of puffs) (Fig 8A). As expected, ORN firing rate was the highest at the onset of puffs (median firing rate 19.8 Hz) and reached a steady state after prolonged exposure due to adaptation (7.3 Hz) (Fig 8B). A very low firing rate was observed in the absence of pheromone, both at the offset of puffs (0.02 Hz) and during tails of blanks (0.5 Hz). PN firing rate was the highest at the onset of puffs (median firing rate 21.2 Hz) and decreased during long puffs down to 18.5 Hz (Fig 8C). On average a transient inhibitory phase followed the end of puffs (1.9 Hz) and the firing rate rose up back to a higher level after a prolonged absence of pheromone (7.8 Hz).
To test how neurons encode the different time sub-regions, we repeated twice a frozen sequence of stimuli simulating a distance of 16 m on PNs (13 cells, 31 pairs of sequences, Fig 8D). We evaluated the correlation coefficient ρ between the measured responses to the first and second presentation of the stimulation sequence. ρ was calculated for all the time sub-regions then averaged for each category. It was larger at the onset (median ρ = 0.91) and offset of puffs (median ρ = 0.95), and indistinctively low both during tails of puffs (median ρ = 0) and tails of blanks (median ρ = 0.05). This observation reveals that PNs do not encode olfactory stimuli during prolonged presentation, but they encode well the beginning and the end of puffs which is sufficient to reconstruct the odor sequence.
We then calculated the average correlation coefficient ρ between the L-N prediction and the measured firing rate. We observed the same patterns as for repeated measures: for ORNs ρ was maximal at the puff onset (median ρ = 0.72) and low during tails of puffs (median ρ = 0.17). At the offset of puffs ρ was high again (ρ = 0.82) and was low during tails of blanks (ρ = 0.28). All paired differences were significant with p<0.01 (paired Wilcoxon’s signed rank test, n = 192 cells; Fig 8E). The pattern was similar for PNs: predictability of the response by the model was high at the onset (median ρ = 0.85) and at the offset of puffs (median ρ = 0.74) and low during tails of puffs (median ρ = 0.08) and tails of blanks (median ρ = 0.08). It is noteworthy that there was no difference in the predictability of responses to tails of puffs and tails of blanks (p>0.8, paired Wilcoxon’s signed rank test, n = 46 cells; Fig 8F).
Thus the L-N model predicts only the reliable neuronal responses to turbulent stimulation, and it fails at predicting neuronal activity when the stimulation is poorly encoded.
We then focused on the two sub-regions with high correlation coefficients ρ (onset and offset of puffs) to know if the predictability of responses varied with the duration of the preceding time interval. We observed a clear difference between ORNs and PNs. In ORNs the L-N model predicted well the response to the beginning of isolated puffs, remote from any previous stimulation (time interval > 3.2 s). The prediction became gradually lower as the preceding puff was at a shorter time interval (Fig 8G). In addition, the response to the offset of short puffs (duration < 0.4 s) was better predicted than the response to the offset of longer puffs (Fig 8H). Thus, the coding in ORNs appears adapted for a finite set of temporal profiles: short and isolated puffs. These statistical characteristics are associated to close distance dynamics to the source. The same dependence on the previous pulse duration and on the pulse duration for the prediction of the response to pulse onset and offset, respectively, were observed with the subset of ORNs stimulated with pheromone loads (10−5 and 10−4 ng, see Materials and Methods) that can be directly compared with the one used to stimulate PNs (Fig 8G and 8H). For PNs the prediction of the response to puff onsets was as high for isolated puffs (time interval > 1.6 s) as for puffs immediately following another one (time interval < 0.1 s), and was even larger for puffs with intermediate time intervals (between 0.4 and 0.8 s; Fig 8G). Moreover, responses to puff onsets were always better predicted for PNs than for ORNs and the optimal prediction of PN response to puff offset was extended to puffs lasting up to 0.8 s (Fig 8H).
We repeated this analysis with the subsets of ORN data stimulated with the same pheromone load and with subsets of ORN and PN data stimulated with sequences simulating the same distance (Fig 9). The prediction of ORN responses to puff onsets decreased when lowering the pheromone load (ANOVA, p < 10−10) independently of the duration of the preceding blanks (Fig 9A, middle panel). The prediction of response to puff onset did not depend on the distance (Fig 9B, middle panel; ANOVA, p = 0.22). The coding of the offset of puffs in ORNs also depended on the pheromone load (Fig 9A, right panel; ANOVA, p<0.01) and was more accurate at long distances than at short ones (Fig 9B, right panel; ANOVA, p < 10−10). Similarly, if analyzing only ORNs stimulated with the pheromone loads (10-5 and 10−4 ng) that can be directly compared with the one used to stimulate PNs, ρ depended on distance for puff offsets (Fig 9C, right panel; ANOVA, p<0.01) but not for puff onsets (Fig 9C, middle panel; ANOVA, p = 0.56). For PNs, correlation coefficients were independent of the distance both during the onset of puffs (ANOVA, p = 0.66) and the offset of puffs (ANOVA, p = 0.75; Fig 9D). It then appears that PNs encoded efficiently the beginning and end of puffs over a broad range of temporal scales, irrespective of the temporal dynamic of the stimulation sequence.
Single-unit neuronal firing recorded from ORNs and PNs in response to sequences with white-noise and turbulent temporal statistics were analyzed with L-N models. L-N models were fitted with white-noise stimuli and used for predicting responses to turbulent stimuli. L-N models have proven to successfully predict the response of ORNs [1, 57, 58] and PNs [53] to time-varying olfactory stimuli. In addition, L-N models revealed correlations between time-varying chemosensory signals and orientation behavior both in C. elegans [59] and Drosophila larva [60–62]. Surprisingly, our study demonstrates that L-N modeling is also suited to predict neuronal responses to odorant stimuli with long-tail distributions expected from turbulent transport. As for many sensory systems, the L-N model is unable to predict adequately neuronal responses during long odor puffs but the coding of specific features is preserved, namely the timing of beginning and end of puffs.
ORN responses are generated locally, by detection of odorants and transduction into trains of action potentials. Olfactory detection depends both on pre-neuronal filtering performed by the sensillum cuticle and lymph and on the ORN intrinsic properties [63]. Olfactory information in PNs is inherited from ORN inputs. However, the processing is also refined by intrinsic properties of projection neurons (PNs) and a neuronal network including local neurons (LNs), PNs and modulatory connections from other brain areas [64–66]. We noticed significant differences in the coding properties of turbulent stimuli between ORNs and PNs, especially at large distances from the odor source. The variability of linear filters was higher in PNs than in ORNs. PNs also coded more adequately for the transition between long and short puffs than ORNs. This phenomenon was due to the inhibition phase at the end of PN responses that contrasted with the adapted response and preserved its detectability. In addition, the firing rate was way below the spontaneous activity during the PN inhibition phase, while in ORNs the spontaneous activity was already close to 0. In our model, the inhibition phase was included in the shape of the linear filters, contributing to the difference between ORN and PN filters, in particular in the PC2 component. The inhibition phase results either from local inhibitory circuits [66–68] and/or from intrinsic PN properties [69]. It is noteworthy that inhibitory mechanisms have been hypothesized to enlarge the temporal bandwidth transmitted through ORN to PN synapses [70]. Previous studies reported that ensemble of PNs can very reliably track temporal properties of the stimulus [71]. Here we show that single PNs can encode puff onset and puff offset for a large diversity of temporal structures of stimuli, and thus virtually at any distance from the source, a key information component moths can use for localizing odor sources from large distances.
The statistic of turbulent olfactory signals changes with distance from the source. Adaptation of neural coding to input signal statistics is found in numerous sensory systems including the visual [72, 73], the auditory [74–77], the mechanosensory [78, 79] and the somatosensory systems [80, 81]. Adaptation is not solely desensitization, but it changes how sensory systems process information. In the insect olfactory system, the cellular mechanism of spike-frequency adaptation induces a sparse and reliable encoding of stimuli across neuronal layers [82]. Here, we show that rapid adaptation does not preclude the detection of plume transitions in PNs. Other stimulus properties could still be encoded in the PN adapted response, like odor identity through synchronous activity [83]. There are 3 behaviorally active compounds in the pheromone of A. ipsilon and PNs can be blend-specific, generalist and component-specific neurons [84, 85]. In neuronal systems, adaptation is rarely considered at time scales longer than a few seconds. However recent evidences suggest that adaptation processes still occurs after tens of seconds in pyramidal neurons and this observation can be implemented in L-N models by including an adaptation filter with the shape of a power-law function [86–90]. Here we used a similar approach to compute the long-term filters and we show that the power-law function is also effective in fitting adaptation in insect olfactory neurons. It thus suggests that in insects, olfactory adaptation should be considered at multiple time scales up to tens of seconds. In natural conditions ORNs and PNs are likely never fully adapted but rather oscillate dynamically between multiple adaptation levels.
Analyzing the neuronal coding of turbulent stimuli requires a mean to monitor precisely the time course of olfactory plumes. Two strategies have been used so far: the use of biological (electroantennographic, EAG) [3, 82] or artificial odor detectors [1, 2]. However, both methods are restricted to odor detections over short distances, up to a couple of meters. Moreover, artificial detectors target a specific range of chemicals and EAG signals already result from a neuronal processing of olfactory information. Finally, it is neither possible to precisely control and set the stimulation, nor to ascertain the relative position of the recording site to the center of the plume cone. We chose to overcome these limitations by using an alternative strategy: we delivered olfactory sequences with plume time statistics corresponding to turbulent flows and tested numerous concentrations and long distance transportation. Significant work is still necessary to complete our understanding of the key features extracted by olfactory coding. Fluctuations of concentration with distance and time will also have to be taken into account. The features used by moths to provide information about source position remain unknown. Thus, a question remains on how many features present in real turbulent flows should be simulated in virtual environments. For example, in the current simulations, duration of puffs and blanks were generated to mimic turbulence structures. Yet, finer structures exist in plume dynamics [8]. In long puffs, groups of high concentration peaks are grouped together (clumps). There are relations between the statistics of peak numbers and the distance to the source, thus moths might be able to exploit these peaks to estimate the distance to the source.
The dynamic of odorant reaching the insect antenna depends not only on the turbulence but also on active olfactory sampling behaviors including wing beating, antennal flicking and body displacement [82, 91, 92]. In realistic olfactory searches strong correlations in detection dynamics (time and concentration) are due to the strategy implemented by moths. Thus, a complete study would require effective modelling of the moth search strategies. However, long distance recording of searching moths coupled to the olfactory input is not accessible. Thus, it is unknown whether moths or other insects are really able to perform olfactory searches at very large distances solely based on the specific input of the odor of interest. Nevertheless, our study unveils the basic mechanisms of olfactory processing of the broadband temporal profiles of olfactory plumes in the turbulent realm.
A. ipsilon moths were reared on an artificial diet until pupation. Adults were maintained under a reversed 16h:8h light:dark photoperiod at 23°C and fed with sugar water. Experiments were performed on virgin 4- or 5-day-old (sexually mature) males.
For all recordings, insects were immobilized with the head protruding. For ORN recordings, the antenna was fixed with adhesive tape on a small support and a tungsten electrode was inserted at the base of a pheromone-sensitive sensillum. The electrical signal was amplified (×1000) and band-pass filtered (10 Hz–5 kHz) with an ELC-03X (npi electronic, Tamm, Germany), and digitized at 10 kHz (NI-9215, National Inst., Nanterre, France) under Labview (National Inst.).
For PN recordings, the head was immobilized with dental wax after removing the scales. The cephalic capsule was open and the brain was exposed by removing the mouth parts and muscles. The antennal lobe was desheathed to allow the penetration of the recording electrode, a glass micropipette whose tip was manually broken to a diameter of 2 μm and filled with (in mM): NaCl 150, KCl 4, CaCl2 6, MgCl2 2, Hepes 10, Glucose 5 (pH 7.2). The same solution was used to perfuse the preparation. The pipette was slowly inserted in the cumulus of the macroglomerulus of one antennal lobe, where ORNs sensitive to the major pheromone component project [51], until the appearance of a well-isolated single-unit firing activity. Extracellular recordings from A. ipsilon proved to sample only neurons with a large neurite (PNs but not LNs) in the antennal lobe [51]. Furthermore, the inhibition phase recorded at the end of each individual response is a hallmark of PN responses both in this species [83] and in other insect species such as Manduca sexta [93] and D. melanogaster [94]. The electrical signal was amplified (×1000) and band-pass filtered (10 Hz–5 kHz) with a Cyberamp 320 (Molecular Devices, Union City, CA, USA) and digitized at 10 KHz (NI-9215) under Labview.
We used the major component, (Z)-7-dodecenyl acetate (Z7-12:Ac) of the A. ipsilon sex pheromone blend as stimulus. Z7-12:Ac was diluted in hexane and applied to a filter paper at doses ranging from 10−6 to 100 ng (ORN recordings) and at 10−1 ng (PN recordings). The entire antenna was exposed to a constant charcoal-filtered and humidified air flow (70 L.h-1). For ORN recordings, air puffs (10 L.h-1) were delivered through a calibrated capillary (Ref. 11762313, Fisher Scientific, France) positioned at 1 mm from the antenna and containing the odor-loaded filter paper (10 × 2 mm). For PN recordings, stimuli were delivered to the entire antenna witha Pasteur pipette containing the odor-loaded filter paper (15 × 10 mm) inserted 15 cm upstream of a 7-mm glass tube positioned at 10 mm from the antenna. Responses to a control (hexane) and pheromone puffs of increasing concentrations (10−3 to 101 ng) were first tested. Only PNs exhibiting dose-dependent pheromone responses were included in the analysis.
Control of electrovalves (LHDA-1233215-H, Lee Company, France) was done by custom-made Labview programs. All sequences were executed asynchronously from text files where the states of electrovalves were indicated by a Boolean variable. Sequences were generated using Matlab scripts (The Mathworks, Natick, MA). The time resolution of the sequences was 1 ms. The characteristic response time of the valves, i.e. the time to go from open to closed (and closed to open) is < 5 ms.
White-noise stimuli were approximated by randomly controlling the state (Open-Close) of the electrovalve used to deliver pheromone stimuli. The valve state was updated every 10, 50 or 100 ms (referred to the correlation time). Thus, the spectra were colored for frequencies greater than 16, 3.4 and 1.6 Hz, respectively. There is no significant effect of correlation times on filter extraction.
Time dynamics of turbulent plumes were derived from Celani et al. [8]. The virtual source and positions of the moths were aligned with the wind direction. We tested 4 virtual distances, 8, 16, 32, 64 m. The geometric progression of distances was chosen to emphasize the effect of turbulence on puff/blank statistics. The distribution of puff/blank time was generated from:
p(τp/b)∝(1τp/b)32fp/b(τp/b)
(1)
where the index p is for puff, b for blanks, fp/b is a cutoff function with exponential decrease of rate Tp/b−1 for τp/b ≥ Tp/b. Cutoffs Tp/b are physical properties of the turbulent flow. Using similar notation as in Celani et al. [8], we set U = 1 m.s-1 (average wind velocity), δU = 0.1 m.s-1 (wind fluctuations), a = 0.1 m (size of the pheromone source), χ = 0.4 (intermittency factor), τ=Ua2δU2d,Tp=dU,Tb=Tp(1χ−1). The largest distance used for ORNs, 64 m, imposed a selection on the generated sequences because some sequences exhibited either extremely long stimuli, which led to the complete stop of spiking activity, or almost no puffs. Thus, the statistics for 64 m was biased from the pure turbulence by removal of extremely rare events (puffs > 30 s). A given sequence of turbulent olfactory sequence was never used twice with the same pheromone concentration unless specifically mentioned for tests with frozen sequences.
A single ORN was recorded per insect. For each recorded ORN, only one sequence of stimuli was delivered during 15 min at a constant pheromone concentration. The different stimulation sequences were tested on different ORNs. White-noise stimulations were performed with 3 correlation times (10, 50, 100 ms) and 4 decades of pheromone concentrations. Turbulent plumes were simulated from 4 distances and 7 decades of concentrations. Control experiments were performed with 2 repetitions of 3-min long white-noise sequences interleaved with various (1, 3, 5, 15 min) inter-sequence intervals.
We tested PNs at 0.1 ng of pheromone loaded on the filter paper, a dose chosen in the medium-low range of cell population dose-response curves [51]. Stimuli were delivered to the entire antenna to ensure proper coding in PNs. For each recorded neuron several stimulation sequences were delivered to the antenna. The first sequence was a 4-min long white-noise stimulation followed by 4 min of no stimuli. The white-noise sequence had a correlation time of 50 ms. Then 4 different sequences of olfactory plume statistics simulating 4 distances were presented in random order. Sequence durations were 5 min and inter-sequence intervals were 3 min. Control experiments were performed in the following way: (1) delivering 4 consecutive frozen white-noise sequences with various inter-sequence intervals, (2) delivering 3 consecutive frozen sequences with a temporal dynamic simulating a distance of 16 m and (3) testing 3 white-noise sequences with different correlation times, 10, 50 and 100 ms.
We could not monitor odor delivery with a photoionization detector (PID) due to the too high ionization energy of Z7-12:Ac and the too low sensitivity of PIDs. Therefore, we cannot perfectly characterize the relation between the dynamics of the electrovalve commands and the effective odor release.
Spike sorting was performed with Spike2 (CED, Oxford). Shapes of action potentials were analyzed using principal component analysis and spike clusters were identified using the first 3 eigenvector subspaces. In long duration experiments involving turbulent stimuli, slow drifts of clusters and secondary clusters resulting from collision with spikes from another cell were taken into account. Finally, visual inspection was used to ensure that rare events with high frequency did not prevent efficient spike detections. Firing rates were calculated by Gaussian convolution of 50 ms width then resampled at 100 Hz.
To model neural responses we used the linear–nonlinear (L-N) model [95]. The L-N model was determined separately for each neuron, and for each dose as one single dose of pheromone was applied on each neuron. The firing rate r(t) of neurons is decomposed into a linear kernel K(t) and a nonlinear function H(r). Thus we have
r(t)=H(∫−∞tduK(t−u)s(u))
(2)
with s(t) the olfactory input signal. The kernel is extracted by minimizing the L2 norm between experimental measures and the L-N prediction with a Lasso regularization [96]
K=argminK(‖r¯−Σ¯¯K¯‖2+λ‖K¯‖2+μ|K¯|1)
(3)
with K¯ the Kernel Vector, r¯ the experimental vector response, Σ¯¯ the time shifted stimulus matrix and (λ, μ) the regularization coefficients. The [pxq] matrix Σ¯¯ was built so that the pth line contains the stimulation vector shifted by p time bins, and q is the bin length of the analyzed time window (-0.5 to 2.5 s). Lasso regularization was used to prevent anomalies in K(t) due both to overfitting and to correlations in the input signal s(t). The L2 norm prevents anomalous rise of the Kernel amplitude and L1 norm favors sparse response. The nonlinear function H(r) was extracted by fitting the experimental rate, r(t), as a function of the linear rate,
rL(t)=∫−∞tduK(t−u)s(u)
using a Hill function. The training samples were made by 20% of the points and the test sample by the remaining 80%. The procedure was applied 100 times per sequence with different values of λ and μ, and these parameters were set to optimize the average determination coefficient R2 of the test sample.
In order to sample coding for turbulent environments, recordings were performed over long time periods (up to 15 min). The firing rate was found to decrease during a stimulus sequence in a predictable and reversible fashion. Thus the L-N model with a short-term kernel was completed adding a long-term kernel leading to the effective Kall(t) = Kshort(t) + Klong(t). Both components of the kernel were simultaneously extracted from Eq 2 where the time shifted stimulus matrix was extended with 7 additional time bins with exponentially spaced bin classes. The modified time shifted stimulus matrix had therefore q+7 columns, the first q columns correspond to the short-term Kernel (-0.5 to 2.5 s), and the last 7 columns, corresponding to the long-time kernel, contained the integrated firing rate calculated in time windows with respective edges of 2.5, 5, 10, 20, 40, 80, 160 and 320 s.
The accuracy of the filters was quantified using classical R2 determination coefficients between prediction and measured data. R2 is the proportion of variance in the measured firing rate (independent variable) that is predicted by the model (dependent variable). For data vs data comparison, as well as predicted vs measured data in small time regions, both variables were relatively independent with different mean signals, and thus correlation coefficients, ρ, were used since more informative than R2.
As we used different odor delivery devices when recording ORNs (peri-sensillum device) and PNs (whole-antenna device), we established dose-response curves from ORNs with both devices to compare them (200-ms stimuli; n = 7). We measured the maximum firing frequency during the response and the number of action potentials evoked until the firing frequency returned to baseline. The comparison of pheromone loads and responses indicates that the concentration delivered on a sensillum by the peri-sensillum device is 103 to 104 higher than with the whole-antenna stimulator (S3A Fig). We inferred that 0.1 ng stimuli with the whole-antenna stimulator correspond approximately to 10−5 to 10−4 ng stimuli with the peri-sensillum device. ORN activities (n = 32) were then recorded in response to a frozen white-noise sequence (5 min, time step = 50 ms) delivered consecutively by both devices (10 ng with the whole antenna device and 5 pg with the peri-sensillum device). The order of use of the two devices was randomized. The average short-term filters calculated from the responses to the white-noise sequences delivered by the two devices overlap (S3B Fig), indicating that they deliver stimuli with the same dynamics.
|
10.1371/journal.ppat.1006900 | Timing of host feeding drives rhythms in parasite replication | Circadian rhythms enable organisms to synchronise the processes underpinning survival and reproduction to anticipate daily changes in the external environment. Recent work shows that daily (circadian) rhythms also enable parasites to maximise fitness in the context of ecological interactions with their hosts. Because parasite rhythms matter for their fitness, understanding how they are regulated could lead to innovative ways to reduce the severity and spread of diseases. Here, we examine how host circadian rhythms influence rhythms in the asexual replication of malaria parasites. Asexual replication is responsible for the severity of malaria and fuels transmission of the disease, yet, how parasite rhythms are driven remains a mystery. We perturbed feeding rhythms of hosts by 12 hours (i.e. diurnal feeding in nocturnal mice) to desynchronise the host’s peripheral oscillators from the central, light-entrained oscillator in the brain and their rhythmic outputs. We demonstrate that the rhythms of rodent malaria parasites in day-fed hosts become inverted relative to the rhythms of parasites in night-fed hosts. Our results reveal that the host’s peripheral rhythms (associated with the timing of feeding and metabolism), but not rhythms driven by the central, light-entrained circadian oscillator in the brain, determine the timing (phase) of parasite rhythms. Further investigation reveals that parasite rhythms correlate closely with blood glucose rhythms. In addition, we show that parasite rhythms resynchronise to the altered host feeding rhythms when food availability is shifted, which is not mediated through rhythms in the host immune system. Our observations suggest that parasites actively control their developmental rhythms. Finally, counter to expectation, the severity of disease symptoms expressed by hosts was not affected by desynchronisation of their central and peripheral rhythms. Our study at the intersection of disease ecology and chronobiology opens up a new arena for studying host-parasite-vector coevolution and has broad implications for applied bioscience.
| How cycles of asexual replication by malaria parasites are coordinated to occur in synchrony with the circadian rhythms of the host is a long-standing mystery. We reveal that rhythms associated with the time-of-day that hosts feed are responsible for the timing of rhythms in parasite development. Specifically, we altered host feeding time to phase-shift peripheral rhythms, whilst leaving rhythms driven by the central circadian oscillator in the brain unchanged. We found that parasite developmental rhythms remained synchronous but changed their phase, by 12 hours, to follow the timing of host feeding. Furthermore, our results suggest that parasites themselves schedule rhythms in their replication to coordinate with rhythms in glucose in the host’s blood, rather than have rhythms imposed upon them by, for example, host immune responses. Our findings reveal a novel relationship between hosts and parasites that if disrupted, could reduce both the severity and transmission of malaria infection.
| The discovery of daily rhythms in parasites dates back to the Hippocratic era and a taxonomically diverse range of parasites (including fungi, helminths, Coccidia, nematodes, trypanosomes, and malaria parasites [1–6]) display rhythms in development and several behaviours. Yet, how rhythms in many parasite traits are established and maintained remains mysterious, despite their significance, as these traits underpin the replication and transmission of parasites [7]. For example, metabolic rhythms of Trypanosoma brucei have recently been demonstrated to be under the control of an oscillator belonging to the parasite, but the constituents of this oscillator are unknown [8]. In most organisms, endogenous circadian oscillators (“clocks”) involve transcription-translation feedback loops whose timing is synchronised to external cues, such as light-dark and feeding-fasting cycles [9,10] but there is generally little homology across taxa in the genes underpinning oscillators. Multiple, convergent, evolutionary origins for circadian oscillators is thought to be explained by the fitness advantages of being able to anticipate and exploit predictable daily changes in the external environment, as well as keeping internal processes optimally timed [11,12]. Indeed, the 2017 Nobel Prize in Physiology/Medicine recognises the importance of circadian oscillators [13,14].
The environment that an endoparasite experiences inside its host is generated by many rhythmic processes, including daily fluctuations in the availability of resources, and the nature and strength of immune responses [15,16]. Coordinating development and behaviour with rhythms in the host (or vector) matters for parasite fitness [17]. For example, disrupting synchrony between rhythms in the host and rhythms in the development of malaria parasites during asexual replication reduces parasite proliferation and transmission potential [18,19]. Malaria parasites develop synchronously during cycles of asexual replication in the host’s blood and each developmental stage occurs at a particular time-of-day. The synchronous bursting of parasites at the end of their asexual cycle, when they release their progeny to infect new red blood cells, causes fever with sufficient regularity (24, 48, or 72 hourly, depending on the species) to have been used as a diagnostic tool. Malaria parasites are assumed to be intrinsically arrhythmic and mathematical modelling suggests that rhythms in host immune effectors, particularly inflammatory responses, could generate rhythms in the development of malaria parasites via time-of-day-specific killing of different parasite developmental stages [20,21]. However, the relevant processes operating within real infections remain unknown [22].
Our main aim is to use the rodent malaria parasite Plasmodium chabaudi to ask which circadian rhythms of the host are involved in scheduling rhythms in parasite development. In the blood, P. chabaudi develops synchronously and asexual cycles last 24 hours, bursting to release progeny (schizogony) in the middle of the night when mice are awake and active. We perturbed host feeding time (timing of food intake), which is known to desynchronise the phase of rhythms from the host’s central and peripheral oscillators, and we then examined the consequences for parasite rhythms. In mammals, the central oscillator in the brain (suprachiasmatic nuclei of the hypothalamus, SCN), is entrained by light [10,23]. The SCN is thought to shape rhythms in physiology and behaviour (peripheral rhythms) by entraining peripheral oscillators via hormones such as glucocorticoids [24]. However, oscillators in peripheral tissues are self-sustained and can also be entrained by several non-photic cues, such as the time-of-day at which feeding occurs [25,26]. Thus, eating at the wrong time-of-day (e.g. diurnal feeding in nocturnal mice) leads to altered timing of oscillators, and their associated rhythms in peripheral tissues. This phase-shift is particularly apparent in the liver where an inversion in the peak phase of expression of the circadian oscillator genes Per1 and Per2 occurs [26]. Importantly, eating at the wrong time-of-day does not alter rhythmic outputs from the central oscillator [25].
In murine hosts with an altered (diurnal) feeding schedule, the development rhythms of parasites remained synchronous but became inverted relative to the rhythms of parasites in hosts fed at night. Thus, feeding-related outputs from the hosts peripheral timing system, not the SCN, are responsible for the timing (phase) of parasite rhythms. We also reveal that the inversion of parasite rhythms corresponds to a phase-shift in blood glucose rhythms. That parasites remain synchronous during the rescheduling of their rhythm coupled with evidence that immune responses do not set the timing of parasite rhythms, suggests parasites are responsible for scheduling their developmental rhythm, and may express their own circadian rhythms and/or oscillators. Furthermore, our perturbed feeding regimes are comparable to shift work in humans. This lifestyle is well-known for increasing the risk of non-communicable diseases (cancer, type 2 diabetes etc. [27]) but our data suggest the severity of malaria infection (weight loss, anaemia) is not exacerbated by short-term desynchronisation of the central and peripheral oscillators.
First, we examined the effects of changing the time of food intake on the phasing of circadian rhythms in host body temperature and locomotor activity (Fig 1). Body temperature is a commonly used phase marker of circadian timing because core body temperature increases during activity and decreases during sleep [28,29]. Mice were given access to food for 12 hours in each circadian cycle, either in the day (LF, light fed) or night (DF, dark fed). All food was available ad libitum and available from ZT 0–12 (ZT refers to ‘Zeitgeber Time’; ZT 0 is the time in hours since lights on) for LF mice, and from ZT 12–24 for DF mice. All experimental mice were entrained to the same reversed photoperiod, lights on: 7pm (ZT 0/24), lights off: 7am (ZT 12), for 2 weeks prior to starting the experiment (Fig 1).
We found a significant interaction between feeding treatment (LF or DF) and the time-of-day (day (ZT 0–12) or night (ZT 12–24)) that mice experience elevated body temperatures (χ2(5,6) = 75.89, p < 0.0001) and increase their locomotor activity (χ2(5,6) = 39.57, p < 0.0001; S1 Table). Specifically, DF mice have elevated body temperature and are mostly active during the night (as expected) whereas LF mice show no such day-night difference in body temperature and locomotor activity, due to a lack of night time elevation in both measures where food and light associated activity are desynchronised (Fig 2). We also find the centres of gravity (CoG; a general phase marker of circadian rhythms, estimated with CircWave), are slightly but significantly earlier in LF mice for both body temperature (approximately 2 hours advanced: χ2(3,4) = 28.17, p < 0.0001) and locomotor activity (approximately 4 hours advanced: χ2(3,4) = 27.32, p < 0.0001) (S1 Table). Therefore, the LF mice experienced a significant change in the daily profile of activity, which is reflected in some phase advance (but not inversion) relative to DF mice, and significant disruption to their body temperature and locomotor activity rhythms, particularly during the night. Because an altered feeding schedule does not affect the phase of the SCN [25], our data suggest that rhythms in body temperature and locomotor activity in LF mice are shaped by both rhythms in feeding and the light-dark cycle [30]. Finally, the body weight of LF and DF mice did not differ significantly after 4 weeks (χ2(3,4) = 0.02, p = 0.9) and both groups equally gained weight during the experiment (S1 Fig), corroborating that LF mice were not calorie restricted.
Having generated hosts in which the phase relationship between the light-entrained SCN and food-entrained rhythms are altered (LF mice) or not (DF mice), we then infected all mice with the rodent malaria parasite Plasmodium chabaudi adami genotype DK (Fig 1) from donor mice experiencing a light-dark cycle 12 hours out of phase with the experimental host mice. After allowing the parasite’s developmental rhythms to become established (see Materials and Methods) we compared the rhythms of parasites in LF and DF mice. We hypothesised that if parasite rhythms are solely determined by rhythms driven by the host’s SCN (which are inverted in the host mice compared to the donor mice), parasite rhythms would equally shift and match in LF and DF mice because both groups of hosts were entrained to the same light-dark conditions. Yet, if rhythms in body temperature or locomotor activity directly or indirectly (via entraining other oscillators) contribute to parasite rhythms, we expected that parasite rhythms would differ between LF and DF hosts. Further, if feeding directly or indirectly (via food-entrained oscillators) drives parasite rhythms, we predicted that parasite rhythms would become inverted (Fig 1).
In the blood, P. chabaudi parasites transition through five developmental stages during each (~24hr) cycle of asexual replication (Fig 3A) [6,31]. We find that four of the five developmental stages (rings, and early-, mid-, and late-trophozoites) display 24hr rhythms in both LF and DF mice (Fig 3B, S2 Table, S2 Fig). The fifth stage—schizonts—appear arrhythmic but this stage sequesters in the host’s tissues [32,33] and so, are rarely collected in venous blood samples. Given that all other stages are rhythmic, and that rhythms in ring stages likely require their parental schizonts to have been rhythmic, we expect schizonts are rhythmic but that sequestration prevents a reliable assessment of their rhythms.
The CoG estimates for ring, and early-, mid-, and late-trophozoite stages are approximately 10–12 hours out-of-phase between the LF and DF mice (Fig 3B and 3C, S2 Table). For example, rings peak at approximately ZT 10 in LF mice and peak close to ZT 23 in DF mice. The other stages peak in sequence. Schizogony (when parasites burst to release their progeny) occurs immediately prior to reinvasion, therefore we expect it occurs during the day for the LF mice and night for DF mice [7]. The almost complete inversion in parasite rhythms between LF and DF mice demonstrates that feeding-related rhythms are responsible for the phase of parasite rhythms, with little to no apparent contribution from the SCN and/or the light: dark cycle.
Changing the feeding time of nocturnal mice to the day time has similarities with shift work in diurnal humans [34]. This lifestyle is associated with an increased risk of acquiring non-communicable diseases (e.g. cancer, diabetes) [35] and has been recapitulated in mouse models [e.g. 36,37,38]. In contrast, in response to perturbation of their feeding rhythm, infections are not more severe in hosts whose circadian rhythms are desynchronised (i.e. LF hosts). Specifically, all mice survived infection and virulence (measured as host anaemia; reduction in red blood cells) of LF and DF infections is not significantly different (comparing minimum red blood cell density, χ2(3,4) = 0.11, p = 0.74; S3A Fig). As described above, changes in body mass were not significantly different between treatments (S1 Fig). Using a longer-term model for shift work may reveal differences in infection severity, especially when combined with the development of non-communicable disease.
There are no significant differences between parasite densities in LF and DF hosts during infections (LF versus DF on day 6 post infection, χ2(3,5) = 0.66, p = 0.42, S3B Fig). This can be explained by both groups being mismatched to the SCN of the host, which we have previously demonstrated to have negative consequences for P. chabaudi [18]. Our previous work was carried out using P. chabaudi genotype AJ so is not directly comparable to our results presented here, because DK is a less virulent genotype [39]. Instead, a comparison of our results to data collected previously for genotype DK, in an experiment where SCN rhythms of donor and host mice were matched (see Materials and Methods; infections were initiated with the same strain, sex, and age of mice, the same dose at ring stage) reveals a cost of mismatch of donor and host entrainment. Specifically, parasite density on day 6 (when infections have established but before parasites start being cleared by host immunity) is significantly lower in infections mismatched to the SCN (LF and DF) compared to infections matched to the SCN (χ2(3,5) = 16.71, p = 0.0002, mean difference = 2.21e+10 parasites per ml blood) (see S4A Fig). In keeping with a difference in parasite replication, hosts with matched infections reach lower red blood cell densities (χ2(3,5) = 18.87, p < 0.0001, mean difference = 5.29e+08 red blood cells per ml blood).
The mismatched and matched infections compared above also differ in whether hosts had food available throughout the 24-hour cycle or for 12 hours only (LF and DF). Restricting food to 12 hours per day does not affect host weight (S1 Fig) and mice still undergo their main activity bout at lights off even when food is available all the time. Therefore, we propose that rather than feeding duration, mismatch to the host SCN for as few as 5 cycles is costly to parasite replication and reduces infection severity. Because peripheral and SCN driven rhythms are usually in synchrony, we suggest parasites use information from food-entrained oscillators, or metabolic processes, to ensure their development is timed to match the host’s SCN rhythms.
Instead of organising their own rhythms (i.e. using an “oscillator” whose time is set by a “Zeitgeber” or by responding directly to time-of-day cues), parasites may allow outputs of food-entrained host oscillators to enforce developmental rhythms. Previous studies have focused on rhythmic immune responses as the key mechanism that schedules parasite rhythms (via developmental-stage and time-of-day specific killing [20,21]). Evidence that immune responses are rhythmic in naïve as well as infected hosts is increasing [15,16], but the extent to which peripheral/food-entrained oscillators and the SCN drive immune rhythms is unclear. Nonetheless, we argue that rhythms in host immune responses do not play a significant role in scheduling parasites for the following reasons: First, mismatch to the host’s peripheral rhythms (which occurs in DF mice but not LF mice as a feature of our experimental design) does not cause a significant reduction in parasite number (S3B Fig), demonstrating that stage-specific killing cannot cause the differently phased parasite rhythms in LF and DF mice. Second, while changing feeding time appears to disrupt some rodent immune responses [40,41], effectors important in malaria infection, including leukocytes in the blood, do not entrain to feeding rhythms [42,43]. Third, inflammatory responses important for killing malaria parasites are upregulated within hours of blood stage infection [44] so their footprint on parasite rhythms should be apparent from the first cycles of replication [19]. In contrast, rhythms of parasites in LF and DF mice do not significantly diverge until 5–6 days post infection, after 5 replication cycles (S3 Table, Fig 4). Fourth, an additional experiment (see Materials and Methods) reveals that rhythms in the major inflammatory cytokines that mediate malaria infection (e.g. IFN-gamma and TNF-alpha: [45,46,47,48]) follow the phase of parasite rhythms (Fig 5), with other cytokines/chemokines also experiencing this phenomenon (S5 Fig). Specifically, mice infected with P. chabaudi genotype AS undergoing schizogony at around midnight (ZT17), produce peaks in the cytokines IFN-gamma and TNF-alpha at ZT21 and ZT19 respectively (following a significantly 24h pattern: IFN-gamma p = 0.0055, TNF-alpha p = 0.0015). Whereas mice infected with mismatched parasites undergoing schizogony around ZT23 (6 hours later), experience 3–6 hour delays in the peaks of IFN-gamma and TNF-alpha (IFN-gamma: ZT0, TNF-alpha: ZT1; following a significantly 24h pattern: IFN-gamma p = 0.0172, TNF-alpha p = 0.0041). Thus, even if parasites at different development stages differ in their sensitivity to these cytokines, these immune rhythms could only serve to increase synchrony in the parasite rhythm but not change its timing.
More in-depth analysis of LF and DF infections provides further support that parasites actively organise their developmental rhythms. We examined whether parasites in DF mice maintain synchrony and duration of different developmental stages during rescheduling to the host’s SCN rhythms. Desynchronisation of oscillators manifests as a reduction in amplitude in rhythms that are driven by more than one oscillator (e.g. parasite and host oscillator). No loss in amplitude suggests that parasites shift their timing as a cohort without losing synchrony. Parasite rhythms in LF and DF mice did not differ significantly in amplitude (χ2(6,7) = 1.53, p = 0.22, S4A Table) and CoGs for sequential stages are equally spaced (χ2(10,18) = 11.75, p = 0.16, S2 Table) demonstrating that parasite stages develop at similar rates in both groups. The rhythms of parasites in LF and DF mice were not intensively sampled until days 6–8 PI, raising the possibility that parasites lost and regained synchrony before this. Previously collected data for P. chabaudi genotype AS infections mismatched to the host SCN by 12 hours that have achieved a 6-hour shift by day 4 PI also exhibit synchronous development (S4B Table and S6 Fig), suggesting that parasites reschedule in synch.
That parasite rhythms do not differ significantly between LF and DF mice until day 5–6 post infection (Fig 4) could be explained by the parasites experiencing a phenomenon akin to jet lag. Jet lag results from the fundamental, tissue-specific robustness of circadian oscillators to perturbation, which slows down the phase shift of individual oscillators to match a change in ‘time-zone’ [10]. We propose that the most likely explanation for the data gathered from our main experiment for genotype DK, and that collected previously for AJ and AS, is that parasites possess intrinsic oscillators that shift collectively, in a synchronous manner, by a few hours each day, until they re-entrain to the new ‘time-zone’. Because there is no loss of amplitude of parasite rhythms, it is less likely that individual parasites possess intrinsic oscillators that re-entrain at different rates to the new ‘time-zone’. The recently demonstrated ability of parasites to communicate decisions about asexual to sexual developmental switches [49] could also be involved in organising asexual development.
If parasites have evolved a mechanism to keep time and schedule their rhythms, what external information might they synchronise to? Despite melatonin peaks in lab mice being brief and of low concentration [50,51], the host’s pineal melatonin rhythms have been suggested as a parasite time cue [52]. However, we can likely rule pineal melatonin, and other glucocorticoids, out because they are largely driven by rhythms of the SCN, which follow the light-dark cycle and have not been shown to phase shift by 12 hours as a result of perturbing feeding timing [25]; some glucocorticoid rhythms appear resistant to changing feeding time [53]. Whether extra-pineal melatonin, produced by the gut for example [54], could influence the rhythms of parasites residing in the blood merits further investigation. Body temperature rhythms have recently been demonstrated as a Zeitgeber for an endogenous oscillator in trypanosomes [8]. Malaria parasites are able to detect and respond to changes in environmental temperature to make developmental transitions in the mosquito phase of their lifecycle [55,56], and may deploy the same mechanisms to organise developmental transitions in the host. Body temperature rhythms did not fully invert in LF mice but they did exhibit unusually low (i.e. day time) temperatures at night. Thus, for body temperature to be a time-of-day cue or Zeitgeber it requires that parasites at early developmental stages (e.g. rings or early trophozoites) are responsible for time-keeping because they normally experience low temperatures during the day when the host is resting. The same logic applies to rhythms in locomotor activity because it is very tightly correlated to body temperature (Pearson’s correlation R = 0.85, 95% CI: 0.82–0.88). Locomotor activity affects other rhythms, such as physiological oxygen levels (daily rhythms in blood and tissue oxygen levels), which can reset circadian oscillators [57] and have been suggested as a time cue for filarial nematodes [4].
Feeding rhythms were inverted in LF and DF mice and so, the most parsimonious explanation is that parasites are sensitive to rhythms related to host metabolism and/or food-entrained oscillators. Malaria parasites have the capacity to actively alter their replication rate in response to changes in host nutritional status [58]. Thus, we propose that parasites also possess a mechanism to coordinate their development with rhythms in the availability of nutritional resources in the blood. Further work could explore whether parasites use information via the kinase ‘KIN’ to regulate their timing [58]. KIN shares homology with AMP-activated kinases (AMPK), mammalian metabolic sensors implicated in both circadian timing and metabolic regulation [59]. Glucose, and other sugars that require metabolising, suppresses the activation of AMPK and its subsequent nutrient-sensing signalling cascade, with KIN proposed to act as a nutrient sensor to reduce parasite replication rate in response to calorie restriction during malaria infection [58]. Rhythms in blood glucose are a well-documented consequence of rhythms in feeding timing [60] and glucose is an important resource for parasites [61]. We performed an additional experiment to quantify blood glucose rhythms in (uninfected) LF and DF mice (Fig 6A and 6B). Despite the homeostatic regulation of blood glucose, we find its concentration varies across the circadian cycle, and is borderline significantly rhythmic in DF mice (p = 0.07, peak time = ZT17.84, estimated with CircWave) and follows a significantly 24-hour pattern in LF mice (p < 0.0001, peak time = ZT8.78). Glucose rhythms/patterns are shaped by feeding regime (time-of-day: feeding treatment χ2(18,32) = 45.49, p < 0.0001). Specifically, during the night, DF mice have significantly higher blood glucose than LF mice (t = 3.41, p = 0.01, mean difference 20.6mg/dl±7.32) and there is a trend for LF mice to have higher blood glucose than DF mice during the day (t = -0.94, p = 0.78, mean difference 7.9mg/dl±9.86).
Titrating whether glucose availability is high or low would only provide parasites with information on whether it is likely to be day or night, and a 12-hour window in which to make developmental transitions should erode synchrony, especially as glucose rhythms are weak in DF mice. Instead, parasites may use the sharp rise in blood glucose that occurs in both LF and DF mice after their main bout of feeding as a cue for dusk (S5 Table; regions with solid lines connecting before and after feeding in Fig 6), using KIN as a sensor [58]. In line with the effects of feeding timing we observe in mice, a recent study of humans reveals that changing feeding time can induce a phase-shift in glucose rhythms, but not insulin rhythms [43]. Alternatively, parasites may be sensitive to fluctuations in other factors due to rhythms in food intake, such as amino acids [62] or other rhythmic metabolites that appear briefly in the blood after feeding, changes in oxygen consumption, blood pressure or blood pH [63,64].
In summary, we show that peripheral, food-entrained host rhythms, but not central, light-entrained host rhythms are responsible for the timing of developmental transitions during the asexual replication cycles of malaria parasites. Taken together, our observations suggest that parasites have evolved a time-keeping mechanism that uses daily fluctuations in resource availability (e.g. glucose) as a time-of-day cue or Zeitgeber to match the phase of asexual development to the host’s SCN rhythms. Why coordination with the SCN is important remains mysterious. Uncovering how parasites tell the time could enable an intervention (ecological trap) to “trick” parasites into adopting suboptimal rhythms for their fitness.
We conducted an experiment to investigate whether host peripheral rhythms or those driven by the SCN affect rhythms in the asexual development of malaria parasites. Our findings stimulated the analysis of four further data sets stemming from three independent experiments. Data available from the Dryad Digital Repository: https://doi.org/10.5061/dryad.jt224 [65]. Here, we detail the approach used for our main experiment “Effect of feeding time on parasite rhythms” before briefly outlining the approaches used in the analyses of additional data “Costs of mismatch to host SCN rhythms”, “Rhythms in cytokines during malaria infection”, “Synchrony during rescheduling” and “Effect of feeding time on blood glucose rhythms”.
We compared the performance of parasites in our main experiment (in which infections were initiated with parasites from donor mice that were mismatched to the host’s SCN rhythms by 12 hours), to the severity of infections when infections are initiated with parasites from donor mice that are matched to the host’s SCN rhythms. Twelve infections were established in the manner used in our main experiment (eight-week-old male mice, strain MF1, intravenously infected with 1 x 107 P. chabaudi DK parasitised RBC), except that donor SCN rhythms were matched to the experimental host’s SCN rhythm and hosts had access to food day and night. Densities of parasites were quantified from blood smears and RBC density by flow cytometry on day 6 and 9 PI, respectively. We chose to compare parasite density in matched infections to LF and DF infections on day 6 PI because parasites are approaching peak numbers in the blood (before host immunity starts to clear infections) and their high density facilitates accurate quantification when using microscopy.
This experiment probes whether host immune responses mounted during the early phase of malaria infection could impose development rhythms upon parasites. We entrained N = 86 eight-week-old female mice, strain MF1, to either a reverse lighting schedule (lights on 7pm, lights off 7am, N = 43) or a standard lighting schedule (lights on 7am, lights off 7pm, N = 43). Donor mice, infected with P. chabaudi genotype AS, were entrained to a standard lighting schedule to generate infections matched and 12 hours mismatched relative to the SCN in the experimental mice. Mice were intravenously injected with 1 x 107 parasitised RBC at ring stage. Genotype AS has intermediate virulence [39] and was used to ensure immune responses were elicited by day 4 PI. We terminally sampled 4 mice every 3 hours over 30 hours starting on day 4 PI, taking blood smears, red blood cell counts and collecting plasma for Luminex cytokine assays.
Cytokines were assayed by the Human Immune Monitoring Centre at Stanford University using mouse 38-plex kits (eBiosciences/Affymetrix) and used according to the manufacturer’s recommendations with modifications as described below. Briefly, beads were added to a 96-well plate and washed in a Biotek ELx405 washer. 60uL of plasma per sample was submitted for processing. Samples were added to the plate containing the mixed antibody-linked beads and incubated at room temperature for one hour followed by overnight incubation at 4°C with shaking. Cold and room temperature incubation steps were performed on an orbital shaker at 500–600 rpm. Following the overnight incubation, plates were washed as above and then a biotinylated detection antibody was added for 75 minutes at room temperature with shaking. Plates were washed as above and streptavidin-PE was added. After incubation for 30 minutes at room temperature a wash was performed as above and reading buffer was added to the wells. Each sample was measured as singletons. Plates were read using a Luminex 200 instrument with a lower bound of 50 beads per sample per cytokine. Custom assay control beads by Radix Biosolutions were added to each well.
We staged the parasites from the blood smears collected from the infections used to assay cytokines (above) to investigate their synchrony during rescheduling. The infections from mismatched donor mice began 12 hours out of phase with the host SCN rhythms and the CoG for ring stage parasites reveals they had become rescheduled by 6 hours on day 4 PI. We focus on the ring stage as a phase marker–for the analysis of synchrony in these data and the divergence between LF and DF parasites–because rings are the most morphologically distinct, and so, accurately quantified, stage.
In a third additional experiment, we entrained 10 eight-week-old male mice, strain MF1, to a standard lighting schedule for 2 weeks before randomly allocating them to one of two feeding treatments. One group (N = 5) were allowed access to food between ZT 0 and ZT 12 (equivalent to the LF group in the main experiment) and the other group (N = 5) allowed access to food between ZT 12 and ZT 0 (equivalent to the DF group). After 10 days of food restriction we recorded blood glucose concentration every 2 hours for 30 hours, using an Accu-Chek Performa glucometer.
We used CircWave (version 1.4, developed by R.A. Hut; available from https://www.euclock.org) to characterise host and parasite rhythms, and R v. 3.1.3 (The R Foundation for Statistical Computing, Vienna, Austria) for analysis of summary metrics and non-circadian dynamics of infection. Specifically, testing for rhythmicity, estimating CoG (a reference point to compare circadian rhythms) for host (body temperature, locomotor activity, blood glucose concentration) and parasite rhythms, and amplitude for parasite stage proportions, was carried out with CircWave for each individual infection. However, the cytokine data display high variation between mice (due to a single sample from each mouse) so we calculated a more robust estimate of phase than CoG by fitting a sine curve with a 24h period (using CircWave) and finding the maxima. Linear regression models and simultaneous inference of group means (using the multcomp R package) were run with R to compare summary measures that characterise rhythms, parasite performance, glucose concentration and disease severity. R was also used to construct and compared linear mixed effects models using which included mouse ID as a random effect (to account for repeated measures from each infection) to compare dynamics of parasite and RBC density throughout infections, and glucose concentration throughout the day.
All procedures were carried out in accordance with the UK Home Office regulations (Animals Scientific Procedures Act 1986; project licence number 70/8546) and approved by the University of Edinburgh. Euthanasia was performed using anaesthesia (combination of Medetomidine and Ketamine) followed by cervical dislocation and rigor mortis as confirmation of death.
|
10.1371/journal.pcbi.1004325 | Alternatively Spliced Homologous Exons Have Ancient Origins and Are Highly Expressed at the Protein Level | Alternative splicing of messenger RNA can generate a wide variety of mature RNA transcripts, and these transcripts may produce protein isoforms with diverse cellular functions. While there is much supporting evidence for the expression of alternative transcripts, the same is not true for the alternatively spliced protein products. Large-scale mass spectroscopy experiments have identified evidence of alternative splicing at the protein level, but with conflicting results. Here we carried out a rigorous analysis of the peptide evidence from eight large-scale proteomics experiments to assess the scale of alternative splicing that is detectable by high-resolution mass spectroscopy. We find fewer splice events than would be expected: we identified peptides for almost 64% of human protein coding genes, but detected just 282 splice events. This data suggests that most genes have a single dominant isoform at the protein level. Many of the alternative isoforms that we could identify were only subtly different from the main splice isoform. Very few of the splice events identified at the protein level disrupted functional domains, in stark contrast to the two thirds of splice events annotated in the human genome that would lead to the loss or damage of functional domains. The most striking result was that more than 20% of the splice isoforms we identified were generated by substituting one homologous exon for another. This is significantly more than would be expected from the frequency of these events in the genome. These homologous exon substitution events were remarkably conserved—all the homologous exons we identified evolved over 460 million years ago—and eight of the fourteen tissue-specific splice isoforms we identified were generated from homologous exons. The combination of proteomics evidence, ancient origin and tissue-specific splicing indicates that isoforms generated from homologous exons may have important cellular roles.
| Alternative splicing is thought to be one means for generating the protein diversity necessary for the whole range of cellular functions. While the presence of alternatively spliced transcripts in the cell has been amply demonstrated, the same cannot be said for alternatively spliced proteins. The quest for alternative protein isoforms has focused primarily on the analysis of peptides from large-scale mass spectroscopy experiments, but evidence for alternative isoforms has been patchy and contradictory. A careful analysis of the peptide evidence is needed to fully understand the scale of alternative splicing detectable at the protein level. Here we analysed peptides from eight large-scale data sets, identifying just 282 splice events among 12,716 genes. This suggests that most genes have a single dominant isoform. Many of the alternative isoforms that we identified were only subtly different from the main splice variant, and one in five was generated by substitution of homologous exons by swapping one related exon for another. Remarkably, the alternative isoforms generated from homologous exons were highly conserved, first appearing 460 million years ago, and several appear to have tissue-specific roles in the brain and heart. Our results suggest that these particular isoforms are likely to have important cellular roles.
| Studies have estimated that alternative splicing can produce differently spliced messenger RNA (mRNA) transcripts for practically all multi-exon human genes [1,2]. These mRNA variants have the potential to expand the cellular protein repertoire far beyond the one gene–one protein model that formed part of the central dogma for many years [3,4].
The number of alternatively spliced transcripts annotated in reference human gene sets has grown steadily in recent years and manual genome annotation projects such as GENCODE [5] are identifying ever more alternative variants. The current version of the GENCODE gene set annotates more than 93,000 protein-coding variants, a number that has increased by 10,000 since 2009.
Theoretically all these transcripts could be translated into functional protein isoforms and could greatly diversify the cellular functional repertoire. However, although we have a limited understanding of the function of a small number of these alternative isoforms, there is a general lack of knowledge about the functional roles of the vast majority of annotated splice isoforms in the cell. All we can say is that most of the annotated splice variants in the human genome will produce isoforms with substantially altered 3D structure and consequent drastic change of biological function, if translated to protein [6,7].
There is considerable supporting evidence for the generation of multiple alternative mRNA transcripts from the same gene. EST and cDNA sequence evidence [8], microarray data [9] and RNAseq data [10] strongly support alternative splicing at the mRNA transcript level.
In spite of the overwhelming evidence of alternative splicing at the transcript level, there is limited support for the translation of these alternative transcripts into protein isoforms. Individual experiments can provide evidence for the expression of isoforms for single genes [11]. At the genome level large-scale antibody tagging [10] holds promise for the detection of alternative isoforms, but the broad specificity of most antibodies makes their discrimination almost impossible at present. For antibody tagging to be of use in distinguishing alternative isoforms, they should be designed from the beginning with this purpose in mind, and each antibody should be only capable of detecting one splice event in one protein. Ribosome profiling experiments [12] have been used in recent years as a proxy for protein coding potential [13,14], but ribosome-profiling data should be used with caution [15] not least because ribosome-profiling methods require transcript reconstruction algorithms to predict splicing variants. The reliability of these transcript reconstruction algorithms has recently been thrown into doubt [16–18]. These factors may have lead research groups to reach two entirely different and opposing conclusions from the same ribosome profiling data [19,20].
High-throughput tandem mass spectrometry (MS)-based proteomics [21] is the main source of peptide evidence. Reliable proteomics data can confirm transcript coding potential even where there is little other supporting evidence. MS-based proteomics has become an increasingly important tool in genome annotation thanks to advances over the last two decades and a number of groups have demonstrated how proteomics data might be properly used to validate translation of protein coding genes [22–24]. On a larger scale, the Human Proteome Project [11] is attempting to identify at least one protein product for each human gene.
Several groups have now identified small numbers of alternative protein isoforms in species ranging from human [22] to mouse [23], Drosophila [25], Arabidopsis [26] and Aspergillus flavus [27]. Recently two large-scale analyses produced similar results. Our group [24] detected the expression of multiple splice isoforms for 150 of 7,597 human genes from an analysis of spectra from the GPM [28] and PeptideAtlas [29] databases, while Low et al. [30] identify 83 alternative splicing events and 13,088 genes in rat.
By way of contrast, a number of recent proteomics studies claim to have found substantially more cases of alternative splicing at the protein level. Menon et al [31] identified 420 alternative isoforms from 1,278 mouse genes, but at the time the mouse genome was not well annotated and it is not clear whether this study required peptides to identify both constitutive and alternative splice isoforms. Recently the numbers of identified splice isoforms have escalated substantially. In two papers published in the same issue of Nature, Kuster and co-workers [32] identified 1,279 alternative proteins for more than 18,097 human genes, while Pandey and colleagues found “isoform-specific peptides” for 2,861 protein isoforms from more than 17,294 genes [33]. As we have shown [34], the main problem with these studies is that they dramatically overestimate the number of reliable peptide identifications. At the extreme end of the scale Ly et al claim to have found evidence for 33,575 separate protein isoforms from just 12,000 human genes [35], suggesting that they identified more than 21,000 alternative isoforms, an order of magnitude greater than any previous study. Here the authors did not use discriminating peptides, but instead chose to infer the expression of different isoforms based on peptide abundances in an analogous way to the protocols used for transcript level estimation in RNAseq studies [16,17]. This form of identifying alternative protein isoforms is wholly inappropriate in proteomics studies because of the low peptide coverage typical of these experiments and because of the non-uniform distribution of the peptides detected.
Given the wide variety in the numbers of splice isoforms reported in what are essentially similar, large-scale proteomics experiments, we felt that it was important to carry out a rigorous study of alternative splicing at the protein level. To accomplish this we produced as reliable a set of peptides as possible from eight high-throughput MS analyses. These analyses were carried out on a wide range of cell types.
We identify alternative splice isoforms for 246 genes from the reliable peptide evidence from the eight data sets. We demonstrate that this is far below what would be expected if the main and alternative splice isoforms were produced in comparable quantities in the cell, suggesting that most genes have a single main protein isoform. We found that homologous exons substitutions, consecutive exons that are homologous and are spliced in a mutually exclusive manner, were highly enriched among the splicing events that we did identify and we show that remarkably few of the events we identified affected the composition of functional domains.
Peptides for the human dataset were collected from eight distinct large-scale proteomics analyses. Two came from proteomics databases, the high quality peptide identifications from the PeptideAtlas database [21] and NIST (http://peptide.nist.gov/), a compendium of peptides identified in multiple proteomics experiments by multiple search engines. The other six correspond to recently published large-scale experiments [32,33,36–39]. These eight datasets cover a large range of tissues and cell types; the peptides from the PeptideAtlas database cover 51 different tissue types and developmental stages. In total the eight datasets contained peptides from over 100 distinct tissues and cell lines. All references to the separate peptide data sets or experiments are by first author name in the text and further details for each individual set after filtering can be found in S1 Table.
The accuracy of peptide identification is a complex issue. Here we decided to work with a conservative set of high quality peptides. In order to generate as reliable a set of peptides as possible we filtered the peptides from each experiment by excluding all non-tryptic and semi-tryptic peptides and all peptides containing missed cleavages that were not supported by at least one of the fully cleaved tryptic peptides. Where possible we included only peptides that were identified by at least two search engines (see Materials and Methods section for details of the individual experiments). The full details of all the filters are explained in the Materials and Methods section.
After filtering we obtained a total of 277,244 unique, high-quality peptides. The number of peptide identifications from each analysis can be seen in S1 Table. We mapped the peptides from the eight analyses to the GENCODE 20 annotation of the human genome (GENCODE 20 corresponds to Ensembl 76 [40] and is the first annotation from build 38 of the human genome). We detected at least two peptides for 12,716 genes, a total of 63.9% of the annotated protein coding genes.
It should be noted that 2,807 of these peptides (1%) did not match GENCODE 20 genes. Almost 75% of this subpopulation of peptides were only detected in one of the eight analyses (compared to 46% of those peptides that do map to annotated coding genes), so a significant proportion are likely to be false positive matches. Peptides detected in two or more experiments have been sent to the GENCODE annotators for further study.
We only considered those peptides that mapped unequivocally to a single gene (discriminating peptides) and only used peptides that were detected in two or more of the eight large-scale analyses in order to reduce the number of false positive identifications. In total 149,612 gene discriminating peptides were detected in multiple analyses; of these peptides 111,382 (74.3%) were able to discriminate between at least two isoforms. To detect peptide evidence of splice isoforms, we required peptides to map to regions that distinguished both sides of a splicing event, whether the splicing event was an indel (Fig 1A) or a substitution event (Fig 1B). To identify deletions we required the peptide to cross the exon boundary. For insertions or substitutions, peptides were permitted to map to any part of the insertion/substitution.
Over the 8 experiments we identified splicing events from 246 genes. This is 60% more than our previous study, in which we reported peptide evidence for 150 genes [24]. We identified splicing events for 77 of these 150 genes in this analysis.
Twenty genes had evidence for more than one alternative splice isoform, and three genes (PLEC, TPM1 and UGT1A) had evidence for five or more different splice isoforms. Here the UGT1A gene cluster is defined as a single gene, even though it is annotated as a cluster of nine independent genes in the GENCODE gene set. These “genes” differ in a set of mutually exclusive 5’ exons that are joined to a common set of 3’ exons by alternative splicing, so we have treated them, and four similar GENCODE 20 gene clusters, as splice variants of a single gene (see Materials and Methods). There was peptide evidence for 8 different protein isoforms from the UGT1A gene (S1 Fig).
To take into account multiple splice isoforms from the same gene, we carried out our analysis on alternative splicing events rather than genes. Alternative splicing events are those that differentiate the alternative isoform from the main isoform (the isoform for which we identify most discriminating peptides). Alternative splicing events identified in the analysis are referred to as identified splicing events (ISE). We found peptide evidence for 282 different ISEs from the 246 genes (S2 Table).
The ISE were classified by their effect on the protein since this is more relevant for a study at the protein level. Events were classified as (i) indels–insertion or deletion of more than 4 amino acid residues (S2 Fig); (ii) NAGNAG splicing [41], defined as the insertion or deletion of up to four amino acid residues (Fig 1); (iii) homologous substitutions (Figs 1 and S3): (iv) other C-terminal substitutions (S4 Fig); (v) other N-terminal substitutions (S5 Fig); (vi) internal non-homologous substitutions (S6 Fig); and (vii) generating two non-homologous proteins (S7 Fig).
The majority of the ISE were indels (109 ISE); this is to be expected since most annotated alternative splicing events are indels [42, 43]. The second most common alternative splicing events were homologous substitutions (60 ISE), followed by non-homologous C-terminal substitutions (43 ISE). There were also numerous alternative splice events involving GYNGYN donors or NAGNAG acceptors (39 ISE). These result in small insertions and are referred to in the paper as NAGNAG splicing events. There were fewer non-homologous N-terminal substitution events (24 ISE), while internal non-homologous substitutions (2 ISE) and events that generated distinct proteins (5 ISE) were less frequent (Fig 2).
We carried out a similar analysis with mouse. We would expect to identify fewer alternative splice variants because the mouse genome is annotated with fewer alternative isoforms (see the Materials and Methods section) and because we only interrogated three analyses, the equivalent peptides from the NIST and PeptideAtlas databases, and an in-house analysis of the mouse spectra in the PeptideAtlas and GPM databases.
As with the human analysis we required all peptides to have been identified in at least two analyses (a more stringent requirement because in this case there were only three proteomics analyses). We identified splice isoforms for 56 genes and 68 splicing events. We detected the identical splice event in the orthologous human gene for 35 of the 68 mouse ISEs. We also detected another 29 mouse events that were equivalent to human ISEs but that were only supported by peptides from a single analysis.
Twenty-one of the 68 mouse ISE (30.1%) we detected were generated from homologous exons and all but one of these 21 splicing events also had peptide evidence in the orthologous human genes. Hence, homologous substitutions are particularly highly represented among those splice events detected in both human and mouse (Fig 2), and make up almost 60% of the orthologous splicing events we detected in both human and mouse experiments.
We previously found that substitution by homologous exons was among the least common annotated splicing events [24], so at first glance the number of homologous exon events detected in the human and mouse analyses is surprisingly high. In order to test whether there were significantly more homologous exon substitution (HES) events than expected in the human analysis, we hand-curated the results from BLAST [44] searches against the GENCODE 20 gene set (see Materials and Methods section) to generate a set of 157 genes with HES events that was independent of the events we detected in the proteomics experiment. We found peptide evidence for 33 of these 157 genes (21%) in our proteomics analysis. We carried out a Fisher test to determine whether the 21% detection rate for the HES genes was significantly different from the 0.01% AS event detection rate for the remaining 19,850 annotated genes: the p-value was close to zero (< 2.2e-16). The set of identified AS events is significantly enriched in homologous exon events.
In pure theoretical terms, the greater the differences between the two isoforms, the easier it should be to identify peptides for both sides of the event and the fewer peptides we should need to detect that event. The easiest AS events to detect should be the longest substitutions (of any type), the largest indels and those cases where there are two completely different proteins. By way of contrast the hardest events to find ought to be substitutions that are almost identical, such as highly similar homologous exon events (there are events that change just a single residue) short indels, or NAGNAG splicing events (for NAGNAG-type splicing events, that generally result in a single residue indel, only a single peptide can identify each side of the splice event). Of course the detection of peptides is not totally random and there are other factors that influence, such as the number of lysines and arginines around the spliced region and how easy it is to detect the individual discriminating peptides by mass spectrometer. However, it is important to bear in mind that all these factors are rendered irrelevant if the protein isoform containing the peptide is not expressed in sufficient quantities to be detected in proteomics experiments.
It ought to be easier to find splicing events for those proteins that are more abundant. In order to identify a pair of splice isoforms for a gene, we need to detect at least two peptides for that gene. We binned genes by protein abundance (here protein abundance is measured as the number of peptides identified for each gene) and plotted the distribution of the AS genes against the background (the remaining annotated genes). As expected, there is a relation between peptide abundance in proteomics experiments and AS detection (see S8 Fig).
If the detection of different types of splicing events is determined by purely theoretical considerations, we should need fewer peptides per gene to identify those splice events that are easier to detect (two distinct proteins) and require more peptides per gene to identify those that are harder to detect (NAGNAG splicing events). However, it turns out that there are few significant differences in the numbers of peptides identified per gene for each of the different splice types (Fig 3A). Wilcoxon rank tests between each group show that the only significant differences were between the “two protein” type and the other types (fewer peptides were needed to detect two distinct proteins as ought to be expected), and between the C-terminal substitution events (fewer peptides) and indel events. We detected substantially fewer peptides in genes in which we identified NAGNAG-type splicing events than in genes where we identified indels when would expect to need to identify more peptides per gene to identify NAGNAG events. Even though NAGNAG splicing events are theoretically the hardest splicing events to confirm, 13.5% of the alternative splicing events we detected were generated from NAGNAG splicing.
We can infer from these results that there is not a strong bias in protein abundance towards any one of different types of splicing event, which implies that the abundance of HES within the set of detected AS events (ISE) is not due simply to their being easier to detect.
We next wanted to determine whether the enrichment of homologous exon events was simply because they are mostly found in highly expressed genes. In principle the results from our study suggest that this bias is unlikely; if we were detecting HES events only because they were in highly expressed genes, we would also expect to detect evidence for other events in the same genes (HES genes are all annotated with other types of splice event in addition to the HES event). In fact we only detected non-HES splice events in one HES gene, LDB3.
To test this we used the 157 HES genes identified with BLAST. We compared the numbers of peptides detected for these genes to a background set of the 13,157 genes that are annotated with multiple splice isoforms. We used these counts to bin the genes according to the number of peptides that were detected and compared the proportions for each of the two populations of genes (Fig 3B).
The two distributions (HES genes and background) are very similar, though there are a slightly higher proportion of the most highly expressed genes in the HES gene population. However, a t-test (two tailed, unequal variance, p = 0.191) shows that this is not significant. Hence, the enrichment of HES events among the AS events we identified is clearly not because HES genes are all highly expressed.
Nor is the enrichment in HES events a consequence of combining the results from the eight analyses—the proportions of each type of splicing event are similar in all eight individual analyses as can be seen in S3 Table and S9 Fig. However, we do find that the HES events that we identified are generally more ubiquitous than the other major splice event types; 36 of the 114 events identified within at least four of the individual analyses (31.6%) were HES events, while 34 were indels and just 12 were C-terminal substitutions (S10 Fig).
These results suggest that the abundance of homologous exons in the alternative splice isoforms identified at the protein level is a real biological phenomenon, and not an artefact of the methods employed in the analysis.
We detected peptides that mapped to paralogous splice events in six human protein families. These splice events clearly predated gene duplication in these families and have remained conserved in at least some of the paralogous genes that make up the family members. We identified homologous events for members of the enigma (LDB3/PDLIM3), alpha-actinin (ACTN1/ACTN4), dynamin (DNM1/DNM2), plasma membrane calcium-transporting ATPase (ATP2B1/ATP2B4), reticulon (RTN3/RTN4), and tropomyosin (TPM1/TPM2/TPM3/TPM4) families. According to EnsemblCompara 78 [45], the gene duplication events date back to vertebrates in the case of the tropomyosins, reticulons and alpha-actinins and jawed vertebrates in the case of the dynamins and the plasma membrane calcium-transporting ATPases. In the case of the enigma family the duplication dates back to chordata phylum. The origin of the splice events we identified must have pre-dated these duplication events. This strongly suggests functional relevance for the splicing events in these genes. The alternative isoforms we detected in five of the six families (enigma, alpha-actinin, dynamin, the plasma membrane calcium-transporting ATPases and tropomyosin) were generated from mutually exclusive homologous exons.
Since five of the six paralogous splice events were generated from HES events, we looked further into the evolutionary conservation of mutually exclusive homologous exons. We scanned the genomes of five distantly related aquatic vertebrates to identify long-standing conservation. We chose lamprey, spotted gar, fugu, zebrafish, and coelacanth as target species to date the origin of each splicing event. We found that every one of the 60 homologous exon splicing events that we identified in the proteomics analysis were present in at least one of these species, implying that they evolved in the ancestor jawed vertebrates or earlier, at least 460 million years ago [46].
As a comparison we calculated the proportion of alternative exons annotated in GENCODE 20 that were conserved between human and mouse. We found that just 19.3% of these alternative human exons were conserved in mouse. Human and mouse diverged close to 90 million years ago, so there is a notable difference in conservation between the HES events we identified and those annotated in the genome.
In a previous paper we showed that the older and more conserved a gene, the more likely we were to identify it in proteomics experiments [39]. The same seems to be true here—the splice isoforms that we identify in proteomics experiments, in this case the isoforms generated from homologous exon events, are the oldest, most conserved splice isoforms.
The analysis of peptide data from the Kim [33] experiments allowed us to investigate tissue-specific alternative splicing. Tissue or cell specific alternative splicing is difficult to detect. Many splice events can only be identified with a single peptide and if that peptide is only detected sporadically, its absence from a tissue does not necessarily imply that it is not expressed in that tissue. Ideally each experiment should have a number of replicates to increase the probability of catching a hard-to-detect peptide. In this regard the Kim study was particularly useful, since most tissues had a small number of replicates.
The evidence for tissue-specific expression of splice isoforms principally came from four tissues: foetal and adult heart, adult cortex and foetal brain. This is not because these tissues had the most peptides (while foetal heart found the most peptides, the other tissues found the 8th most, the 22nd most and the 24th most peptides). We found five tissue-specific isoforms in brain. FYN is a tyrosine protein kinase that has two isoforms generated by homologous exons. Isoform 1 (FYN-B) and 2 (FYN-T) are supposed to be highly expressed in the brain and in hematopoietic cells, respectively [47]. The Kim experiment identified tissue-specific splicing of FYN [33] and we confirmed that FYN-B was indeed present in adult and foetal brain samples, while there was evidence for FYN-T in blood cells. FYN regulates cytoskeletal remodelling and cell survival by phosphorylating a number of proteins. Curiously we identified brain-specific splice isoforms for two of these: MAPT (Tau protein) and AGAP2. Isoform 1 of AGAP2 (PIKE-L) is known to be brain specific [48] and we found peptides for this isoform in adult and foetal brain in the Kim experiment. Isoform 2 (PIKE-A) is described as ubiquitous: we detected peptides in blood cells. MAPT isoform PNS-tau is expressed in the peripheral nervous system while other isoforms are expressed in the central nervous system [49]. We found peptides for PNS-tau in adult heart, while peptides mapping to other MAPT isoforms were found in both adult and foetal brain. The splicing events that distinguish the MAPT and AGAP2 isoforms are indels. We also found a brain specific alternative isoform for GLS (gelsolin) with different N-terminals, and heart and brain-specific isoforms for VDAC3 (Voltage-dependent anion-selective channel protein 3), a mitochondrial membrane protein with a NAGNAG splicing event (Fig 1).
The other hotspot for tissue-specific splicing was the heart. We found heart-specific splice isoforms for a further nine genes and most of these genes locate to the thin filaments or Z-discs (see Fig 4). Peptides from the Kim analysis identify cardiac specific splice isoforms for three members of the ALP/enigma family [50]. All seven family members are cardiac expressed, but specific functions have only been found for PDLIM3 (ALP), PDLIM5 (ENH) and LDB3 (ZASP). All three are present in the Z-discs and interact with ACTN2, a gene that is estimated to make up 20% of the Z-disc mass [51]. LDB3 and PDLIM5 have both been implicated in dilated cardiomyopathy DCM [50,52]. The data from the Kim analysis suggested that ZASP1 is the major LDB3 isoform expressed isoform in foetal heart, while ZASP2 and ZASP6 seem more highly expressed in adult heart. Isoform ZASP1 differs from isoforms ZASP2 and ZASP6 by the substitution of a remotely homologous exon. The two isoforms of PDLIM3 we identified in the analysis (ALP-SK and ALP-H) are formed from a splicing event paralogous to the one in LDB3. The peptides from the Kim analysis locate ALP-H to adult and foetal heart and ALP-SK to oesophagus. The third gene, PDLIM5, has a heart-specific isoform (ENH1e) that differs from the non-cardiac isoform by a large insertion.
NEBL is one of the very few genes that generate splice isoforms with different Pfam domain compositions. The main isoform, nebulette, contains 14 nebulin repeats and a C-terminal SH3 domain. The shorter isoform LIM-nebulette swaps 12 of the N-terminal nebulin repeats for a LIM domain. Nebulette is a cardiac-specific isoform that is known to bind actin at the Z-discs [53]. At least four nebulette-specific amino acid variants [54] have been implicated in DCM. The data from the Kim analysis locates nebulette uniquely to adult and foetal heart tissue, while LIM-nebulette was found in frontal cortex, spinal cord, lung, kidney and prostate.
TPM1 (tropomyosin) is one of just nine human genes that is annotated with multiple HES events. It is closely involved with nebulette in heart muscle and their interaction is important in the maintenance and stability of the thin filaments [55]. Again TPM1 is known to be important in DCM; a number of TPM1 single nucleotide variants that are likely to be pathological for DCM have been described in the literature [56]. We mapped these variants to the TPM1 splice isoforms and found that all twelve likely pathogenic variants mapped to isoform TPM1-002 (TPM1alpha) suggesting that this was the most important isoform in heart tissue. Data from the Kim analysis confirmed that TPM1alpha is preferentially expressed in heart tissues [57]. We identified more peptides for TPM1alpha than any other TPM1 isoform in every single one of the nine heart tissue experiments and fewer peptides for TPM1alpha in all other tissues.
CAPZB, the beta subunit of F-actin-capping protein, generates two proteins via alternative 3’ homologous exons that localize to different regions in the heart [58]. Splice isoform CAPZB1 is located at the Z-discs where it caps the thin filaments (see Fig 4). We found more evidence for CAPZB1 in heart tissues in the Kim analysis, while the CAPZB2 isoform was identified in liver, kidney, foetal brain and all red blood cells. We also identified a known cardiac-specific isoform for TTN (cardiac novex-3) and possible heart-specific isoforms for TPM2 and ITGA7, both of which are known to be present in the heart but which are not annotated with cardiac-specific isoforms. The isoforms for all three of these genes were generated by HES events.
In total we found evidence for tissue-specific alternative splicing for 14 genes. Many of these genes are known to interact with each other. FYN interacts with MAPT and AKAP2 in the brain, while seven of nine heart-specific isoforms are located to the Z-discs and at least five are known to be important in DCM (Fig 4). Though this is only a small sample, it is interesting to note that there seem to be tissue-specific protein-protein interaction networks involving AS isoforms. More than half of these 14 genes (and 7 of the 9 genes with cardiac specific isoforms) generate their tissue specific splice isoforms via alternative splicing of homologous exons.
Very few structures of alternative isoforms have been crystallised in the PDB structural database [59]. Hegyi and colleagues [60] found just 15 cases with resolved 3D structures of splice isoforms in the PDB. It has been suggested that many splicing events (in particular non-conserved splicing events) will result in unstable conformations [61]. This may explain the lack of crystallised structures: proteins that do not fold in stable conformations will not form proper crystals and their structures will not be resolved.
We looked for evidence of protein 3D structures for all 282 splice events detected in the analysis in the PDB, allowing the splice isoforms to come from any related species. We found ten pairs of alternative isoforms in which each isoform had a resolved 3D structure. Nine of the ten pairs were generated from homologous exons (Table 1), suggesting that structures generated from homologous exons might be easier to resolve. Among the 157 HES genes identified by BLAST searches, we found another 9 pairs of structures for alternative isoforms generated from homologous exons making a total of 18 pairs of structures for isoforms generated from homologous exons deposited in the PDB.
The structures of the 9 pairs of alternative isoforms identified in our analysis are shown in Fig 5 and S11, S12, S13, S14 and S15 Figs. The HES events have a range of effects on the structures. The HES region of the gene ACP1 is confined to one surface of the protein, while the homologous substitution in gene H2AFY will clearly affect the binding to the nucleotide substrate. In gene PFN2 the homologous region (from the orthologous mouse protein, but 100% sequence identical) may affect binding to the proline rich peptide substrate. In gene MASP1 the whole trypsin domain is generated from homologous exons that have less than 33% identity, but the two proteins maintain the same fold.
The two structures resolved for the MASP1 gene illustrate an important principle regarding the effect of alternative splicing on protein structures. It is clear from Fig 5D that the overall structure of the alternative MASP1 trypsin domain is not affected even though the identity between the two alternatively spliced trypsin domains is only 33%. In fact this is to be expected since protein structures are very stable in the face of evolutionary change. Two proteins as little as 10% identity can have the same overall structure, as long as they are homologous [62]. By way of contrast indels and non-homologous substitutions that fall in regions with globular may produce completely different folds (or unfolded proteins) even though the two proteins have a relatively high identity over their whole structure.
The final pair of resolved alternative isoforms comes from the well-studied CDKN2A gene where translation from two different (conserved) frames results in two completely different proteins (S16 Fig). There are many structures for isoform p19INK, but just one for p16ARF and this structure is from mouse.
We looked at the effect of splicing events on the domains annotated in the Pfam functional domain database [63] for the 282 splicing events we identified. Pfam A domains were mapped to all isoforms with the program Pfamscan [63]. We counted a Pfam domain as broken if the splice event would cause the domain to lose or gain five or more residues.
The alternative splicing events identified in this study tended to not to have an effect on the composition of Pfam functional domains. Just 19 of the 282 splice events identified in the proteomics analyses (6.7%) would break Pfam domains, while 20 more (7.1%) would lead to the loss of one or more Pfam functional domains (while leaving the remainder intact). Five ISE would result in a swap of one set of Pfam domains for a different set. The remaining 84.4% of identified splice events would leave Pfam domains untouched. In the case of the homologous exons where the HES event coincided with a Pfam domain, we considered the Pfam domain unbroken, except for the plasma membrane calcium-transporting ATPases (ATP2B1/ATP2B4), where the substitution event clearly broke a poorly defined Pfam domain (see S17 Fig).
We calculated the effect of splicing events on Pfam domains over the whole GENCODE 20 gene set and against 4 other subsets of genes as a comparison. We took the principal isoform for each gene from the APPRIS database [64]. The APPRIS principal isoforms have been shown to be a reliable means of predicting dominant isoforms at the protein level [18]. We then generated pairwise alignments between these principal isoforms and all alternative isoforms. We counted all unique splice events from the pairwise alignments between the principal isoform and each of the alternative isoforms for all genes. We mapped the splicing events to the Pfam-A annotations from Pfamscan. The four subsets of genes compared were: all 12,716 genes we detected peptides for, all genes that we detected at least 20 peptides for (2,271 genes, the median number of peptides for the HES genes), all genes that we detected at least 50 peptides for (385 genes), and all the 246 genes for which we detected splice isoforms (Fig 6).
The GENCODE 20 gene set annotates 45,346 unique alternative splicing events; 37.3% of the splice events (16,937) occur inside Pfam functional domains and would generate alternative isoforms with damaged functional domains. Another 32% (14,766) of the splicing events would lead to the loss of one or more whole Pfam domains (Fig 6). The figures were similar for the other four gene subsets (Fig 6), even for the subset of 246 genes for which we detected splice isoforms. The 246 genes that we identified splice isoforms for are annotated with 1,650 splice events, 524 (31.8%) would split Pfam domains, while 477 (28.9%) would lead to the loss of one of more whole Pfam domains (Fig 6). In clear contrast, the splice events that we detected in the proteomics experiments tended not to lead to the loss or damage of Pfam functional domains. These results clearly show that most alternative isoforms with damaged or lost Pfam domains are not produced in quantities detectable in standard proteomics experiments. This strongly suggests that many variants that affect Pfam composition are produced in very low quantities or that there is some form of control at the level of translation, or post-translation, that protects the cell against protein isoforms with damaged domains.
While we find peptides for 12,716 genes, we identify a mere 282 splice events (0.62% of the annotated events). Over the 8 experiments we identified just 0.02 alternative splice events per identified gene while the numbers of annotated splice events per gene is 2.28. This seems to be a very small proportion.
Not only do we identify very few splice isoforms, we also detect very few alternative peptides. If we were to use trypsin to digest the human proteome (in this case the GENCODE 20 annotation of the human coding genes that we used to map the peptides from the eight analyses), we would generate 693,324 unique tryptic peptides of at least seven residues. A total of 11.21% of these 693,324 unique tryptic peptides would come exclusively from isoforms tagged as “alternative” by APPRIS (those isoforms that were not principal isoforms, see Materials and Methods section). In contrast in our analysis just 0.38% of the 149,612 discriminating peptides that we detected mapped to alternative isoforms. The equivalent numbers for the alternative peptides, splice events and genes identified in the individual experiments can be found in the supplementary S1 Table.
We attempted to estimate the number of AS genes we would have expected to identify from our data using a range of simulated null models. For the first model we generated tryptic peptides from the GENCODE 20 gene set by in silico lysis. Each peptide was represented just once in the database of peptides. We used just tryptic peptides for this simulation because all possible missed cleavage combinations would make the search space too large. For this first model we did not eliminate any redundancy from the GENCODE 20 annotation, so as to approximate a model in which all annotated transcripts were expressed equally. If we had only used tryptic peptides in our experiments we would have found evidence of alternative splicing for 226 genes (20 of the AS genes were identified via missed cleavages), while there would have been evidence of three or more isoforms for 14 genes.
To simulate the expected numbers of AS genes we drew peptides at random from this database. In order to approximate the expression levels from the experiments, the number of tryptic peptides drawn for each gene was the same the number we identified in our analysis. We repeated the simulation 100 times and took the mean value to be the estimation of the number of AS genes we would have expected to find if genes were expressed at experimental levels and all isoforms of each gene were expressed equally.
The numbers of AS genes from this in silico analysis were substantially larger than those in our experiment; in our model we found evidence of alternative splicing for 3,508 genes (15 times greater than the experiments), and evidence of at least three isoforms for 937 genes (67 times greater than the experiments).
We repeated the simulation, but this time carrying out an in silico analysis that produced 50-times more peptides for the main isoform of each gene than for all the other isoforms. For this model we used the principal isoforms from the APPRIS database as a stand-in for dominant splice isoforms [64]. Again for each gene we randomly selected exactly the same number of tryptic peptides as we had identified in our experiment. We repeated the simulation 100 times. This simulation approximates a model in which genes have expression levels similar to the experimental expression and all peptides are equally detectable, but in which one isoform is produced 50 times more than the other isoforms. The simulation showed that we would have expected to detect 1,289 genes with evidence of AS at the protein level and another 152 genes with at least three distinct splice isoforms in this model. Again this is in clear contrast with the observed results and suggests that the majority of alternative isoforms may be produced at levels that are considerably lower than 2% of those of the main isoform. This evidence clearly supports a model in which most genes have one dominant isoform at the protein level [18].
We generated a highly reliable set of peptides from eight large-scale proteomics analyses by applying rigorous filters. The rigorous quality controls on the peptide data allowed us to be confident that the isoforms we identified were expressed and present in high enough quantities to be detected in proteomics analyses. With these peptides we detected the expression of alternative isoforms for 282 distinct splice events from 246 distinct human genes. While the filters undoubtedly limited the number of splice events we detected, they do mean that this set of alternative splice isoforms can be regarded as a gold standard for what can be detected in large-scale proteomics experiments.
Even with peptide data from eight large-scale analyses that cover a wide range of tissues, cell lines and developmental stages, we still detect many fewer alternative isoforms than would be expected from transcript data. We found peptides for almost 64% of annotated protein coding genes, but identified less than 0.6% of the annotated alternative splice events. In part this may be due to proteomics technology. Standard MS experiments generate relatively low coverage of the proteome and cannot detect peptides expressed at very low levels. This is a technical problem that is unlikely to be resolved in the short term.
We found unexpectedly high numbers of isoforms generated by alternative splicing of homologous exons; more than 20% of the splice events we detected in the human proteomics experiments came from mutually exclusively spliced homologous exons and these homologous substitutions made up 60% of the orthologous splicing events detected in both mouse and human experiments.
A surprisingly high proportion of isoforms from homologous exon substitutions had resolved 3D structures. The explanation for this may be simply that structures generated from homologous exons are easier to crystallize [61]. Homologous structures will maintain their 3D fold, while non-homologous exons that fall in structured regions may cause the 3D structures to become partially unfolded and therefore difficult to resolve.
The recent publication of a large-scale tissue-based proteomics analysis with replicates [33] allowed us to carry out a study of alternative splicing at the level of tissues. We found evidence for tissue-specific expression of fourteen pairs of alternative isoforms. Curiously those genes for which we detected tissue-specific splicing isoforms had at least one isoform that was specifically expressed in either heart or brain and many of them are known to interact. The heart-specific isoforms were particularly interesting because the majority proteins coded by these variants are known to locate at the Z-discs and to be involved in dilated cardiomyopathy. For many of the remaining isoforms the data was inconclusive. Seven of the nine heart-specific isoforms we identified were generated from homologous exon splicing events.
In 2001, Kondrashov and Koonin [65] found evidence for 50 genes with homologous exon substitutions across a range of species. They reported that half of the 29 HES for which they identified an ancestor arose in the mammalian lineages. With the data now available, we find that all the HES detected in our experiments (and 27 of the 29 HES identified by Kondrashov and Koonin), had their origins in the ancestor of jawed vertebrates or earlier, more than 460 million years ago. This is a remarkable level of conservation for alternatively spliced isoforms.
In contrast Modrek and Lee [66] found that only 25% of what they termed “minor” alternative exons were conserved between human and mouse. We carried out our own analysis of alternative exons (see Materials and Methods section) and found a similar result, just 19.3% of the 3,626 alternative exons we analysed (excluding homologous exons) were conserved between human and mouse. The homologous exon substitution events we identified are clearly much more conserved than this.
The ancient conservation of isoforms generated from homologous exon substitution events, taken together with the abundance of peptide evidence for these isoforms, their tissue-specific expression, and the fact that these events have a demonstrably subtle, non-disruptive, effect on protein structure, strongly suggests that alternative splice isoforms generated from mutually exclusively spliced homologous exons are likely to have important cellular roles that merit further investigation.
Most of the splice events we identify in this analysis would have relatively modest effects on protein structure and function; many alternative isoforms were generated from homologous exons and even most indels were either short or fell in regions that are likely to be unstructured. In fact very few of the splice events we detected would damage or cause the loss of conserved Pfam functional domains. This is in sharp contrast to the splice variants annotated in the GENCODE gene set, where the majority of the splice events would be expected to have an effect on Pfam domains.
The preference for splice events that do not disrupt Pfam functional domains and the analysis of evolutionary conservation strongly suggest that not all annotated alternative transcripts will be converted into stable proteins. One possible explanation for this finding is that alternative protein isoforms with damaged or lost functional domains are more likely to have a disruptive effect on cellular processes and their production may be subject to regulation by one of the many cellular quality control pathways [67–70], to ensure that isoforms with damaged domains are not present in the call in large quantities.
At the moment, it is still not clear how much of the alternative splicing observed in the transcriptome is functionally relevant. Our results suggest that, at the protein level at least, the diversity generated by alternative splicing may be smaller than most previous estimates. If true these findings will have important practical implications for variant-calling analyses that include potentially non-relevant transcripts [71] and will affect our understanding of how organisms and complexity evolve.
The peptides for the human analysis came from eight proteomics datasets. Six of the peptide datasets came from large-scale experiments [32,33, 36–39]. The remaining sets of peptides came from two spectra databases, NIST (http://peptide.nist.gov/) and PeptideAtlas [21]. The peptides for the mouse analysis came from the same two mass spectrometry databases (NIST and PeptideAtlas). In addition we generated a set of peptides in house from an X!Tandem [72] search against spectra from mouse mass spectrometry experiments deposited in the GPM [28] and PeptideAtlas databases, following the protocol set out in Ezkurdia et al. [24].
Note that the nature of the PeptideAtlas and NIST databases means that a number of spectra from the experiments are likely to have been interrogated multiple times by different search engines and with different post-processing filters. The peptides that map to these spectra will be in our study when the peptide-spectrum mappings agree (mostly for reliably mapped peptides), but will not appear in our study when the peptide-spectrum mappings do not agree (which will often be the case for false positive mappings).
We wanted to make sure that the alternative isoforms that we detected were not identified from incorrectly mapped peptides, so we used a series of filters to remove as many false positive peptides as possible from each analysis. The peptides from the individual analyses were filtered as follows.
The peptides from the Geiger [36] and Nagaraj [37] experiments were treated in identical fashion. The peptides in these studies had a peptide false discovery rate (FDR) of 1%, but for the purposes of this experiment we required the peptides identified in the two datasets also to have an Andromeda [73] score of 100 or more. It has been shown that using multiple search engines increases performance [74,75] and since peptides identified by Andromeda with scores of 100 or greater are almost always in agreement with those identified by Mascot [76] for the same spectra [73], concentrating on the peptides with PSM above this score decreases the false positive rate.
The Kim experiment [33] used the Mascot and Sequest [77] search engines and the peptides identified in their analysis came from the union of these two search engines. The Kim experiment described the peptide-spectrum match (PSM) FDR as being 1%. We have shown that this is likely to be an underestimate [34]. In order to be more rigorous for the purposes of our experiment, we only included the peptides from the intersection of the two search engines used in the Kim analysis, that is, those peptides that were identified by both the Mascot and Sequest search engines.
The Wilhelm analysis [32] also used two search engines, this time Mascot and Andromeda. Again the peptides came from the union of the two search engines. The Wilhelm experiment had a 1% PSM FDR and a 5% peptide FDR, but again we found that this was likely to be an underestimate [34]. Upon analysis of the scores from the two search engines used we found a high number of dubious spectra in which Andromeda and Mascot agreed on a peptide match, but both search engines had very low scores. For the purposes of our experiment we treated the Wilhelm analysis in the same way as the Geiger and Nagaraj analyses; we only took those peptides identified with an Andromeda PSM score of 100 or more.
The NIST database uses five different search engines (Sequest, Andromeda, Mascot, X!Tandem and OMMSA [78]) to identify peptides from human spectra, and three search engines for spectra from mouse experiments. While the NIST database includes many peptides, the FDR is quite high. For the purposes of our experiment we filtered out those NIST peptide-spectrum matches identified by just one search engine.
The peptides from the Ezkurdia analysis [39] were identified with X!Tandem and had a peptide FDR of 0.1%, while the peptides from the Munoz [42] analysis were identified using Mascot and had a peptide FDR of 1%. PeptideAtlas peptides have a PSM FDR of 0.0002%.
All peptide data sets were then subject to the following filters: we filtered out non-tryptic and semi-tryptic peptides and only allowed peptides with missed cleavages that were supported by at least one of the fully cleaved sub-peptides. We applied the equivalent rule to the peptides from the Wilhelm analyses for peptides cleaved with LysC and chymotrypsin, and to the peptides detected in the Nagaraj analysis that were cleaved by GluC or LysC enzymes. In the case of the Ezkurdia, PeptideAtlas and NIST analysis we did not know the digesting enzyme a priori, so we chose the conservative option of assuming that all peptides were cleaved by trypsin. Search engines do not easily distinguish leucine from isoleucine due to their identical mass, so leucine and isoleucine residues were allowed to map to either leucine or isoleucine in the GENCODE20 gene set. All peptides that mapped to more than one gene were disregarded in the analysis. The numbers of peptides and genes identified in the individual experiments can be seen in S1 Table.
We only mapped peptides that were identified by two or more different experiments or databases in our analysis of alternative isoforms. We excluded peptides identified in just one of the eight peptide data sets, since, even after the filtering carried out in this analysis, a proportion of these peptides are likely to be false positives.
For the human analysis the peptides were mapped to the protein isoforms annotated in the GENCODE 20 human gene set. The gene set was first filtered for pseudoautosomal genes and for read-through transcripts. Read-through transcripts are flagged in the GENCODE 20 annotation and are transcripts that are (generally) formed by skipping the last exon of a gene and reading through to the neighbouring gene or non-coding gene, or pseudogene. All these transcripts are highly unlikely to be translated into proteins and (most importantly) overlap with known coding transcripts. Read-through transcripts that overlap with coding transcripts would render these coding transcripts indistinguishable. The remaining GENCODE 20 gene set was annotated with 19,906 protein-coding genes and 92,341 protein isoforms; the manual GENCODE annotations are highly enriched in alternative isoforms [49]. 15,548 genes were annotated with more than one distinct splice isoform.
For the mouse analysis we mapped peptides to the isoforms annotated in the GENCODE mouse M2 gene set (equivalent to Ensembl74). The M2 gene set had 22,645 protein-coding genes and 51,610 transcripts. In the mouse gene set just 10,607 genes were annotated with protein sequence distinct variants.
We mapped isoform-discriminating peptides to the splice isoforms annotated in the GENCODE 20 gene set looking for peptides that mapped unambiguously to distinct splice isoforms. The initial set of splice isoforms was checked by hand by mapping the isoform-discriminating peptides to multiple alignments. Only those splice events for which we identified peptides that mapped to both sides of the events were included in the final set.
There was peptide evidence for the expression of 7 different isoforms from the UGT1A gene cluster. Although UGT1A transcripts are annotated as independent genes in GENCODE 20, we have treated them as splice variants of a single gene in this analysis. In the UGT1A gene cluster the individual genes/transcripts share a set of common 3’ exons and each has a unique (but homologous) 5’ exon (S1 Fig). As with the Drosophila gene Lola (where the variable exons are at the 3’ end rather than the 5’ end [79]), the different protein products are formed by joining one of a set of variable exons to the common exons,
The selection of the 5’ exons is thought to be determined via alternative promoters. The use of alternative promoters is not a standard alternative splicing mechanism, but the end result is the alternative splicing of exons, just as it is in Lola in Drosophila, another non-standard alternative splicing mechanism, where variable and constant exons are joined by trans- rather than cis-splicing. There are three other similar gene clusters in the GENCODE 20 gene set (UGT2A, PCDH-gamma and PCDH-alpha) and we regarded all four gene clusters as single genes with alternatively spliced protein isoforms.
Based on Ensembl version 78 of December 2014 [40] we compared human transcripts to automatically identify a set of genes with mutually exclusive homologous exons. We defined the transcript with the longest amino acid sequence as the reference against which compare other transcripts and looked for pairs or sets of exons that are mutually exclusive, i.e. that do not co-occur in the same transcript. For exons that were more than 30 bps long, we obtained their amino acid sequences and compared them with BLAST v2.2.25 [44], setting an e-value threshold of 0.005. To validate each of the resulting potential homologous exons we assessed whether the exons occupied equivalent positions within the corresponding alternative transcripts. In addition, we discarded those cases in which one of the alternative exons belonged to a paralogous neighbour gene or pseudogene. Finally, we included additional cases of homologous exons that were identified by BLASTP searches [39], but missed in this automatic analysis. We visually inspected the alignments of all the potential cases. We ended with a set of 157 genes with (mutually exclusive) homologous exon substitutions. Although the proteomics analysis of alternative splicing revealed other additional cases of homologous exons beyond this set, we did not include them in this test set to avoid potential biases.
To identify whether the homologous exons substitutions we detected were of ancient origin, here defined as those that originated more than 400 million years ago, we scanned the genomes of five distantly related vertebrates using TBLASTN [44] with the exons as amino acid query sequences, turning off low complexity filtering and setting an e-value threshold of 0.1. These five taxa included lamprey, spotted gar, zebrafish, fugu, and coelacanth, all of which were retrieved from Ensembl v75. We used bedtools v2.17.0 [80] to group sequence similarity hits based on the 95 percentile of gene lengths of each target species, then assigning these hits to annotated or new genes in each target species. For every target gene we determined whether it was orthologous or paralogous to the query human gene using EnsemblCompara phylogenetic trees [45]. Results were carefully revised to determine which genes from each target species had specific hits to each query exon, i.e. whether the human homologous exons were already present in that species. The origin of the homologous exons was inferred under the assumption that homologous exons have not been acquired independently in different species, i.e. we relied on Dollo parsimony [81].
We determined the constitutive exons to be those that were annotated as principal isoforms in APPRIS [64]. We defined as alternative exons all those protein-coding exons that did not overlap with the constitutive exons. We found 13,079 of these exons in the GENCODE 20 gene set. We improved the reliability of the annotation by filtering out those genes where the principal isoforms was not determined by the core modules or by unique CCDS identifier [82]. We further filtered these exons as follows.
We removed exons that were too short to identify homology with mouse in the TBLASTN searches. These were defined as those with a BLAST e-value higher than 0.001 when compared against the whole human proteome, which includes the query exons (these are exons for which we may expect a significant sequence similarity hit in the mouse genome if the exon is conserved). We also removed exons that were similar to exons in the APPRIS principal isoforms (e-value threshold of 0.1). This avoids complicating the interpretation of potential conservation and also excludes exons that can be defined as homologous. If exons overlapped with each other we took the longer of the exons as the representative. We also filtered out 15 genes presenting a large number of alternative exons caused by a gene model that was clearly not finished or had errors that were influencing the selection of the principal isoform by APPRIS (for example, FRAS1, where the gene model was unfinished in GENCODE 20 or FIP1L1, which had an unannotated read-through transcript that APPRIS selected as the main isoform).
After all the filtering steps, we ended up with a set of 3,626 alternative human exons. Using these exons as amino acid query sequences we searched the mouse genome with TBLASTN, turning off low complexity filtering and setting an e-value threshold of 0.1. We defined the exon as conserved when a significant similarity was found within a mouse gene (or close to the gene, we set a distance threshold based on the 95 percentile of gene lengths) that is present in the same EnsemblCompara phylogenetic tree than the corresponding human gene.
APPRIS [64] derives a principal splice isoform for each gene based on the presence of protein features, such as protein structure, functional domains and cross-species conservation. These features can discriminate between splice isoforms because they have generally been conserved over large evolutionary timeframes. Isoforms that have “lost” these features are unlikely to be the principal isoform. APPRIS maps protein structural information from the structural homologs in the PDB [60] to individual isoforms, annotates functional information from the Pfam domain database [63] and from the functionally important amino acid residues from firestar [83] and evaluates the cross-species conservation of every isoform. The isoform with the most conserved protein features is chosen as the principal splice isoform. Where the APPRIS annotations are not sufficient to distinguish a single principal isoform, such as genes TMPO [84] or MARVELD3 [85], APPRIS uses external annotations, such as presence in the CCDS database [82] in order to make a decision.
The variant chosen from the core APPRIS modules have been shown to be in almost complete agreement with the main proteomics isoform [18].
Here we selected a single isoform as the main isoform (using APPRIS principal isoforms) and defined an alternative splice event as being an event that changed the protein sequence between the main isoform and alternative isoforms. Alternative proteins that were simply truncated were not used to count splice events in order to avoid including annotated fragments of transcripts from unfinished gene models. We used PfamScan to annotate Pfam domains onto the transcripts and counted all those cases where the splice event would cause the Pfam domain to lose five or more amino acid residues (a damaged Pfam domain), or the whole Pfam domain (a lost Pfam domain).
|
10.1371/journal.pgen.1002333 | A Genome-Wide Screen for Interactions Reveals a New Locus on 4p15 Modifying the Effect of Waist-to-Hip Ratio on Total Cholesterol | Recent genome-wide association (GWA) studies described 95 loci controlling serum lipid levels. These common variants explain ∼25% of the heritability of the phenotypes. To date, no unbiased screen for gene–environment interactions for circulating lipids has been reported. We screened for variants that modify the relationship between known epidemiological risk factors and circulating lipid levels in a meta-analysis of genome-wide association (GWA) data from 18 population-based cohorts with European ancestry (maximum N = 32,225). We collected 8 further cohorts (N = 17,102) for replication, and rs6448771 on 4p15 demonstrated genome-wide significant interaction with waist-to-hip-ratio (WHR) on total cholesterol (TC) with a combined P-value of 4.79×10−9. There were two potential candidate genes in the region, PCDH7 and CCKAR, with differential expression levels for rs6448771 genotypes in adipose tissue. The effect of WHR on TC was strongest for individuals carrying two copies of G allele, for whom a one standard deviation (sd) difference in WHR corresponds to 0.19 sd difference in TC concentration, while for A allele homozygous the difference was 0.12 sd. Our findings may open up possibilities for targeted intervention strategies for people characterized by specific genomic profiles. However, more refined measures of both body-fat distribution and metabolic measures are needed to understand how their joint dynamics are modified by the newly found locus.
| Circulating serum lipids contribute greatly to the global health by affecting the risk for cardiovascular diseases. Serum lipid levels are partly inherited, and already 95 loci affecting high- and low-density lipoprotein cholesterol, total cholesterol, and triglycerides have been found. Serum lipids are also known to be affected by multiple epidemiological risk factors like body composition, lifestyle, and sex. It has been hypothesized that there are loci modifying the effects between risk factors and serum lipids, but to date only candidate gene studies for interactions have been reported. We conducted a genome-wide screen with meta-analysis approach to identify loci having interactions with epidemiological risk factors on serum lipids with over 30,000 population-based samples. When combining results from our initial datasets and 8 additional replication cohorts (maximum N = 17,102), we found a genome-wide significant locus in chromosome 4p15 with a joint P-value of 4.79×10−9 modifying the effect of waist-to-hip ratio on total cholesterol. In the area surrounding this genetic variant, there were two genes having association between the genotypes and the gene expression in adipose tissue, and we also found enrichment of association in genes belonging to lipid metabolism related functions.
| Serum lipids are important determinants of cardiovascular disease and related morbidity [1]. The heritability of circulating lipid levels is estimated to be 40%–60% and recent genome-wide association (GWA) studies implicated a total of 95 loci associated with serum high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), total cholesterol (TC), and triglyceride (TG) levels [2]. Currently identified common variants explain 10%–12% of the total variation in lipid levels, corresponding to ∼25% of the trait heritability [2].
Epidemiological risk factors, such as alcohol consumption, smoking, physical activity, diet and body composition are known to affect lipid levels [3]–[5]. These risk factors also show moderate to high heritabilities, and over 120 loci with genome-wide significant association have been identified (http://www.genome.gov/26525384). To better understand the biological processes modifying lipid levels, several twin studies [6]–[8] and candidate gene studies [9]–[14] have tested for interactions between genes and epidemiological risk factors.
Interactions between genes and modifiable risk factors might help us develop new lifestyle interventions targeted to susceptible individuals based on their genetic information. The effects of genetic loci and risk factors have been studied widely separately, but to date no GWA studies for interactions on lipids have been reported.
We conducted a genome-wide screen for interactions between 2.5 million genetic markers and sex, lifestyle factors (smoking and alcohol consumption), and body composition (BMI and WHR) in association to serum lipid levels (TC, TG, HDL-C, and LDL-C) in 18 population-based cohorts (max N = 32,225; Table S1A, Text S1). We defined interaction as a departure from a linear statistical model allowing for the additive main effects of both the SNP and the epidemiological risk factor.
18 SNPs with suggestive interactions for at least one of the trait – epidemiological factor combinations (P-value for the interaction <10−6) in stage 1 analyses were taken forward to stage 2 analysis in eight additional cohorts (max N = 14,889; Table S1B, Text S1). In inverse variance meta-analyses combining the results from stage 1 and stage 2 (Table S2), the interaction between rs6448771 in chromosome 4p15 and WHR on TC (Figure 1) was statistically genome-wide significant (stage 1 and 2 combined P = 9.08×10−9). This interaction was tested in stage 3 in two further cohorts (N = 7,813; Table S1C, Text S1), which showed an effect to the same direction. After combining results from all three stages (total N = 43,903), the P-value for interaction was 4.79×10−9. The association between WHR and TC was strongest for individuals carrying two G alleles of rs6448771, for whom a one standard deviation (sd) difference in WHR corresponds to 0.19 sd difference (confidence interval 0.13–0.25) in TC concentration, while for individuals homozygous for the A allele the difference was 0.12 sd (confidence interval 0.09–0.16) (Table S3A, Figure S1). The effect corresponds to 0.5% and 0.2% of the total variance explained in a cohort of young individuals (YFS, mean age = 37.6) and an old cohort (HBCS, mean age = 61.49), respectively. Additionally, when looking at the effect of the SNP on TC in WHR tertiles, the estimates differed in a way that the estimated SNP effect is higher for the individuals with higher WHR (Table S3B). The SNP did not have a direct effect on either TC or WHR (P = 0.46 and P = 0.51, respectively, Figure 1). The SNP rs6448771 is located 249 kb downstream of the protocadherin 7 (PCDH7) gene.
Since the polymorphisms associated with complex phenotypes often influence gene expression, we examined whether individuals carrying different genotypes of rs6448771 have variation in their transcript profiles. As WHR reflects adipose tissue function, we selected 54 individuals from Finnish dyslipidemic families with available fat biopsies and GWA data. We used linear regression to find genes that were differentially expressed in adipose tissue depending on the rs6448771 genotype. We found two potential candidate genes with nominally significant cis-eQTL effects, PCDH7 (P = 0.027, distance from the rs6448771 250 kb) and CCKAR (P = 0.017, distance from the SNP 4.9 Mb). The region with CCKAR has previously been linked with obesity [15]. Additionally, using Ingenuity software (IPA), we conducted a pathway analysis for genes with eQTL P-value<0.01 (both trans- and cis-eQTLs). Among other diverse IPA-defined biological functions, there was an eQTL association enrichment among genes belonging to the ‘degradation of phosphatidylcholine’ (3 genes out of 6, P = 6.64×10−5, Benjamini-Hochberg corrected P = 0.0138) and ‘degradation of phosphatidic acid’ (4 genes out of 8, P = 4.71×10−4, B-H corrected P = 0.0349) functions, which are members of broader defined IPA categories “Lipid Metabolism” and “Carbohydrate Metabolism”. These pathways were up-regulated in individuals carrying the G allele of rs6448771, possibly indicating a role for rs6448771 in lipid and carbohydrate metabolism.
The associated SNP also shows evidence for interactions with WHR on LDL-C (effect estimate for the interaction = 0.03, P = 0.0016) and HDL-C (effect estimate = 0.02, P = 0.029) in our stage 1 meta-analysis and after adjusting for TC no residual interaction effect on LDL-C and a little on HDL-C remains (P = 0.834 and P = 0.131 respectively) when testing in data subset. Therefore we tested the SNP – WHR interaction also on a range of lipoprotein subclasses measured using NMR metabonomics platform [16] available in two cohorts (NFBC1966, N = 4624 mean age = 31.0; YFS, N = 1889, mean age = 37.6). The results show that the SNP has a positive interaction effect on large HDL particle concentration (combined effect for the interaction = 0.538, P = 0.0186) and a negative effect on large very-low-density lipoprotein (VLDL) particles (combined effect = −0.466, P = 0.0291) and total triglycerides (combined effect = −0.454, P = 0.0343) (Figure 2).
Our genome-wide scan for interactions between SNP markers and traditional epidemiological risk factors in population-based random samples found a genome-wide significant locus, rs6448771, modifying the relationship between WHR and TC. The effect of WHR is estimated to be 64% stronger for individuals carrying two copies of the G allele than for individuals carrying two A alleles. The interaction explains around half a percent of the TC variance that is in par with the main effects of the strongest previously identified TC SNPs individually. This SNP also shows similar interaction effects on a cascade of more detailed lipid fractions suggesting broad involvement in lipid metabolism, which was also suggested by our eQTL association enrichment analysis with adipose tissue expression data.
The eQTL analysis pointed towards two potential candidate genes in the region. The first one of these was protocadherin 7 (PCDH7) gene, which produces a protein that is thought to function in cell-cell recognition and adhesion. The other candidate gene, cholecystokinin A receptor (CCKAR) regulates satiety and release of beta-endorphin and dopamine in the central and peripheral nervous system. It has been previously shown that rats with no expressed CCKARs developed obesity, hyperglycemia and type 2 diabetes [17]. To test whether our eQTL finding was adipose tissue specific, we ran the eQTL analysis for PCDH7 and CCKAR in another dataset with genome wide expression data from blood leukocytes (N = 518) available. CCKAR could not be tested due to its negligible expression in blood leukocytes, and no association was found for the PCDH7 (P-value = 0.284) gene most likely indicating an adipose tissue specific eQTL for PCDH7 as a function of rs6448771.
One interesting aspect of this study, given our large sample size, is that only one signal achieved genome-wide significance, where previously published lipid GWA studies have found close to a hundred. Although power to detect interaction is typically lower than for main effects, especially for rare exposures and SNPs, several of the exposures considered here (such as WHR, BMI, and gender) were common and available for a large proportion of the study sample. This suggests that the contribution of two-way G×E interactions to lipid levels, at least for the risk factors we examined, is rather small, or that our current measures of risk factors may not be robust enough for identifying interactions. More specific measures of both phenotypes and interacting risk factors would give better statistical power in future screens of G×E interactions.
Our findings allow us to draw several conclusions. First, to our knowledge, this is the first time an interaction between a genetic loci and a risk factor has been identified in a genome-wide scan using a stringent statistical threshold for genome-wide significance. Second, in our samples, rs6448771 modified the relationship between WHR and TC, but was not associated with either WHR or TC alone. This observation suggests that genome-wide screens for interactions may be complementary to the current large-scale GWAS efforts for finding main effects. Third, in addition to careful harmonization of both risk factor data and phenotypes, large sample sizes are needed to identify interactions. In our study, 43,903 samples were combined to robustly identify the interaction. Our data, however, suggest that the contribution of G×E interaction using current phenotypes appears limited. Finally, from clinical point of view, the interaction may open up possibilities for targeted intervention strategies for people characterized by specific genomic profiles but more refined measures of both body-fat distribution and metabolic measures are needed to understand how their joint dynamics are modified by the newly found locus.
18 studies, with a combined sample size of over 30,000 individuals, participated in the discovery phase of this analysis; 8 studies were available for replication with over 14,000 individuals. In the discovery stage, only population-based cohorts not ascertained on the basis of phenotype, with a wide variety of well-defined epidemiological measures available, were included. In the replication datasets, the NTR cohort was selected on the basis of low risk for depression and the Genmets samples were selected for metabolic syndrome. In further replication of rs6448771, the EPIC cases were ascertained by BMI. Descriptive statistics for these populations are detailed in Table S1A (discovery), S1B (replication) and S1C (further replication). Brief descriptions of the cohorts are provided in the Text S1 section “Short descriptions of the cohorts”.
Individuals were excluded from analysis if they were not of European descent or were receiving lipid-lowering medication at the time of sampling. TC, HDL-C, and TG concentrations were measured from serum or plasma extracted from whole blood, typically using standard enzymatic methods. LDL-C was either directly measured or estimated using the Friedewald Equation (LDL-C = TC – HDL-C – 0.45×TG for individuals with TG≤4.52 mmol/l, samples with TG level higher than 4.52 were discarded in the calculation of LDL-C) [18].
Covariates and epidemiological risk factors were ascertained at the same time that blood was drawn for lipid measurements. BMI was defined as weight in kilograms divided by the square of height in meters. Waist circumference was measured at the mid-point between the lower border of the ribs and the iliac crest; hip circumference was measured at the widest point over the buttocks. Waist-to-hip ratio was defined as the ratio of waist and hip circumferences. Alcohol consumption and smoking habits were determined via interviews and/or questionnaires. Both behaviors were coded as dichotomous (abbreviations: ALC for drinker/abstainer and SMO for current smoker/current non-smoker) and semi-quantitative traits. Semi-quantitative alcohol usage (ALCq) was based on daily consumption in grams (0: 0 g/day; 1: >0 and ≤10 g/day; 2: >10 and ≤20 g/day; 3: >20 and ≤40 g/day; 4: >40 g/day). Semi-quantitative smoking (SMOq) was assessed based on the number of cigarettes per day (0: 0 cigarettes/day; 1: >0 and ≤10 cigarettes/day; 2: >10 and ≤20 cigarettes/day; 3: >20 and ≤30 cigarettes/day; 4: >30 cigarettes/day).
Affymetrix, Illumina or Perlegen arrays were used for genotyping in the discovery cohorts. Each study filtered both individuals and SNPs to ensure robustness for genetic analysis. After quality control, these data were used to impute genotypes for approximately 2.5 million autosomal SNPs based on the LD patterns observed in the HapMap 2 CEU samples. Imputed genotypes were coded as dosages, fractional values between 0 and 2 reflecting the estimated number of copies of a given allele for a given SNP for each individual. Cohort specific details concerning quality control filters, imputation reference sets and imputation software are described in Table S4.
Replication cohorts utilized genome-wide imputed data, as described above, where available. Details on the genotyping methods implemented in the replication samples are described in Table S4.
Proton NMR spectroscopy was used to measure lipid, lipoprotein subclass and particle concentrations in native serum samples. NMR methods have been previously described in detail [16], [19]. Serum concentrations of total triglycerides (TG), total cholesterol (TC) together with LDL-C and HDL-C were determined. In addition, total lipid and particle concentrations in 14 lipoprotein subclasses were measured. The measurements of these subclasses have been validated against high-performance liquid chromatography [20]. The subclasses were as follows: chylomicrons and largest VLDL particles (particle diameters from approx 75 nm upwards), five different VLDL subclasses: very large VLDL (average particle diameter 64.0 nm), large VLDL (53.6 nm), medium-size VLDL (44.5 nm), small VLDL (36.8 nm), and very small VLDL (31.3 nm); intermediate-density lipoprotein (IDL) (28.6 nm); three LDL subclasses: large LDL (25.5 nm), medium-size LDL (23.0 nm), and small LDL (18.7 nm); and four HDL subclasses: very large HDL (14.3 nm), large HDL (12.1 nm), medium size HDL (10.9 nm), and small HDL (8.7 nm).
Triglyceride concentrations were natural log transformed prior to analysis. BMI and WHR were transformed to normality using inverse-normal transformation of ranks. For analyses where sex was the epidemiological variable of interest, the phenotypes were defined as the rank-inverse normal transformed residuals resulting from the regression of the lipid measurement on age and age2. For the other analyses, the phenotypes were defined as the inverse normal transformed residuals resulting from the regression of the lipid measurement on age, age2, and sex.
Associations between the transformed residuals and epidemiological risk factors/SNPs were tested using linear regression models under the assumption of an additive (allelic trend) model of genotypic effect. The models regressed phenotypes on epidemiological factor, SNP, and epidemiological factor×SNP termsand tested if the effect for E×SNP was 0 using 1 df Wald tests. In family-based cohorts, linear mixed modeling was implemented to control for relatedness among samples [21]. Analysis software used by the individual cohorts is described in Table S1A and S1B.
The interaction terms from the regression analyses were meta-analyzed using inverse variance weighted fixed-effects models [22]. Prior to meta-analysis, genomic control correction factors (λGC) [23], calculated from all imputed SNPs, were applied on a per-study basis to correct for residual bias possibly caused by population sub-structure. Meta-analyses were performed by two independent analysts using METAL (http://www.sph.umich.edu/csg/abecasis/Metal/index.html) and the R [24] package MetABEL (part of the GenABEL suite, http://www.genabel.org/). All results were concordant, reflecting a robust analysis. Results were selected for in silico replication if the meta-analysis P-value was less than 10−6. Results passing the threshold of suggestive genome-wide association (P-value ≤5×10−7) were selected for further replication by direct genotyping.
The commonly accepted genome wide level of significance (5×10−8) reflects the estimated testing burden of one million independent SNPs in samples of European ancestry [25]. To address the multiple testing arising from testing interactions with multiple risk factors, we set the genome wide significance threshold to 5×10−8/3 = 1.67×10−8 corresponding to three principal components explaining 97.8% of the total variation of the risk factors (Table S5).
|
10.1371/journal.ppat.1000219 | Small RNAs with 5′-Polyphosphate Termini Associate with a Piwi-Related Protein and Regulate Gene Expression in the Single-Celled Eukaryote Entamoeba histolytica | Small interfering RNAs regulate gene expression in diverse biological processes, including heterochromatin formation and DNA elimination, developmental regulation, and cell differentiation. In the single-celled eukaryote Entamoeba histolytica, we have identified a population of small RNAs of 27 nt size that (i) have 5′-polyphosphate termini, (ii) map antisense to genes, and (iii) associate with an E. histolytica Piwi-related protein. Whole genome microarray expression analysis revealed that essentially all genes to which antisense small RNAs map were not expressed under trophozoite conditions, the parasite stage from which the small RNAs were cloned. However, a number of these genes were expressed in other E. histolytica strains with an inverse correlation between small RNA and gene expression level, suggesting that these small RNAs mediate silencing of the cognate gene. Overall, our results demonstrate that E. histolytica has an abundant 27 nt small RNA population, with features similar to secondary siRNAs from C. elegans, and which appear to regulate gene expression. These data indicate that a silencing pathway mediated by 5′-polyphosphate siRNAs extends to single-celled eukaryotic organisms.
| Regulation of gene expression can occur via multiple conserved pathways. One such mechanism is mediated by RNA molecules of about 21–24 nucleotides (called small RNAs), which can affect rates of RNA degradation or protein production. These small RNA molecules regulate diverse biological processes in a broad range of systems. The vast majority of the published literature about these molecules is from multi-cellular organisms. We have made a number of novel observations with respect to small RNA size, structure, and function in Entamoeba histolytica, a single-celled parasite and an important human pathogen. Our work has identified that E. histolytica has an abundant population of 27 nucleotide small RNAs, which have an unusual structure, indicating that they are generated by a relatively atypical mechanism. A substantial portion of these small RNAs are antisense to target genes and appear to silence them. These data establish a new paradigm for how gene expression is regulated in this organism. Furthermore, the identification of small RNAs with these structural characteristics dramatically broadens the evolutionary spectrum in which this phenomenon has been identified and indicates significant diversity and complexity of small RNAs and their functions in single-celled eukaryotes.
| Small RNAs mediate post-transcriptional gene silencing in a multitude of organisms and in diverse biological processes [1],[2],[3],[4],[5]. Two proteins central to the small RNA mediated gene silencing pathways are Dicer, an RNaseIII enzyme, which generates small RNAs and Argonaute, which associates with the small RNAs and target genes to mediate gene silencing. Multiple classes of small RNAs have recently been described including small interfering RNAs (siRNAs), microRNAs (miRNAs), trans-acting siRNAs (tasiRNAs), tiny noncoding RNAs (tncRNAs), small scan RNA (scRNA), repeat-associated small interfering RNA (rasiRNA), piwi-interacting RNA (piRNA), and secondary siRNAs [6],[7],[8],[9]. Some organisms have multiple populations of small RNAs associated with different mechanisms of gene regulation. Notably siRNAs, miRNAs, tasiRNAs, tncRNA, and scnRNA are all products of Dicer cleavage. In contrast, rasiRNA, piRNA, and secondary siRNA appear to be formed independent of Dicer processing [6],[7],[8],[10].
Primary siRNAs are produced from long double stranded RNAs and can be endogenously derived from repetitive genomic regions, transposon elements, or regions with active antisense transcripts. Primary siRNAs are generated by Dicer processing, which generates a 5′-monophosphate (5′-monoP) and 3′-hydroxyl (3′-OH) structure. Primary siRNAs are subsequently loaded into Argonaute to mediate gene silencing but can also serve as the “trigger” to initiate RNA-dependent RNA polymerase (RdRP) generation of secondary siRNAs. In plants, secondary siRNAs, although generated by RdRP, are eventually processed by Dicer and thus the majority have the classic 5′-monoP termini [9]. In C. elegans, secondary siRNAs have a 5′-polyphosphate (5′-polyP) structure, a feature not identified in any other siRNAs to date. C. elegans secondary siRNAs largely map antisense to genes, are biased towards the 5′ side of primary trigger RNAs, and amplify gene silencing by their association with CSR-1, an Argonaute protein [7],[8],[10],[11],[12]. Because of their 5′-polyP structure, C. elegans secondary siRNAs are most efficiently cloned in a 5′-phosphate independent manner [7],[8].
Entamoeba histolytica, a single celled eukaryote, is an important human pathogen and a leading parasitic cause of death worldwide [13]. The parasite has two stages in its life cycle: an invasive trophozoite form, which causes disease and a dormant cyst form, which transmits disease [14]. The genome of E. histolytica is predicted to encode a number of genes conserved in the RNAi pathway including three genes with Piwi and PAZ domains (EHI_186850, EHI_125650, and EHI_177170), and two genes with RdRP domains (EHI_139420 and EHI_179800) [15]. However, no obvious homologue of Dicer was identified, although RNaseIII activity was detected in E. histolytica trophozoites and a protein with an RNaseIII domain (EHI_068740) has been identified in the genome sequence [15],[16]. Recently it has been shown that dsRNA, siRNA, and short-hairpin RNAs are effective in achieving gene silencing in E. histolytica suggesting that the machinery for small RNA mediated silencing is functional in this parasite [17],[18],[19]. Additionally, putative microRNAs were identified in E. histolytica using a bioinformatics-based approach, however, none of those predictions were confirmed using high resolution Northern blot analysis [20]. Thus, although there are hints that a functional RNAi pathway exits in E. histolytica, endogenous small RNAs have not previously been identified in this eukaryotic pathogen.
In order to identify endogenous small RNAs in E. histolytica, we used a 5′-phosphate independent cloning method to clone small RNAs from E. histolytica trophozoites. Our analysis identified an abundant population of small RNAs (∼27nt) with features highly reminiscent of secondary siRNAs from C. elegans. E. histolytica 27 nt small RNAs have 5′-polyphosphate termini, largely map antisense to genes with bias towards the 5′ ends of genes, are associated with a Piwi-related protein, and appear to regulate strain-specific gene expression. Thus, Entamoeba histolytica, an organism typically considered a simple eukaryote, appears to use complex regulatory mechanisms to mediate gene silencing. Even though there are some important differences between these small RNAs in E. histolytica and C. elegans (including the size of the 5′-polyP small RNAs), our data indicate that the secondary siRNA mechanism of gene silencing extends deep into the evolutionary spectrum.
In order to visualize small RNAs in E. histolytica trophozoites, we fractionated total RNA with a YM100 column and visualized the samples on 12% denaturing polyacrylamide gel stained with SYBR gold. Three distinct populations of small RNAs were visualized: ∼27 nt, ∼22 nt, and ∼16 nt with the 27 nt population the most abundant (Figure 1A). A similar pattern was seen in trophozoites of the non-invasive human parasite Entamoeba dispar (ED), and the reptilian parasite Entamoeba invadens (EI) (Figure S1). Thus, it appears that the overall pattern of three distinct small RNA populations is conserved in Entamoeba species.
Since the 27 nt small RNA population was the most abundant we subsequently focused our efforts on this population. We first attempted to define the features of the 5′ and 3′ termini of this small RNA population. We identified that the E. histolytica 27 nt fraction is likely to have a 3′-OH, as it can be efficiently labeled using RNA ligase and cytidine 3′,5′-bisphosphate [5′-32P] (abbreviated 32pCp hereafter) (Figure 1B). By contrast, the 5′ end is not likely to be a 5′-monoP, as it could not be efficiently labeled using polynucleotide kinase (PNK), unless first treated with calf intestinal phosphatase (CIP) (Figure 1B). To further confirm that the 5′ termini was not a 5′-monoP, we took advantage of the specificity of Terminator exonuclease for 5′-monoP RNA species: 5′-monoP RNAs are efficiently degraded by treatment with Terminator (a 5′-3′ exonuclease), whereas RNAs with a 5′-cap, 5′-polyP, or 5′-OH will be resistant. A substantial portion of the 27 nt small RNA population was resistant to treatment with Terminator enzyme, whereas the control (PNK labeled RNA ladder with 5′-monoP termini) was largely degraded by Terminator treatment (Figure 1C). Finally, treatment of the 27 nt fraction with a capping enzyme (which only caps RNAs with 5′ termini containing di- or tri-phosphate structures) increases the signal and the size of the 27 nt fraction (Figure 1D). Collectively, these data indicate that a substantial portion of the E. histolytica trophozoite 27 nt small RNAs have 5′-polyP termini. Some of the 27 nt population may be 5′-monoP species, as some RNA was labeled with PNK and there was a slight reduction in RNA abundance after treatment with Terminator enzyme.
The standard method for small RNA cloning depends on the presence of a 5′-monoP on the small RNA of interest [10]. However, this 5′-phosphate dependent method of cloning will not efficiently capture small RNAs with a 5′-polyP structure. In order to clone 27 nt small RNAs with 5′-polyP termini from E. histolytica trophozoites, we utilized a 5′-phosphate independent method of cloning and performed limited sequencing [8]. A total of 289 small RNA sequences were obtained, 243 of which were unique, and 196 of which mapped to the E. histolytica genome sequence (Table S1 and Figure S2). Overall, 27% of small RNAs mapped to ribosomal RNAs, 27% mapped to tRNAs, 20% mapped antisense to ORFs, 13% mapped to intergenic regions, 5% mapped to multiple genomic loci, 3% mapped sense to ORFs, 5% mapped to repetitive regions, and 0.5% mapped to retrotransposon elements. The size distribution of the cloned small RNAs peaked at 27–28 nt, matching the size of the population from which they were cloned (Figure 2A).
We also cloned small RNAs from a 15–30 nt size selected fraction from E. histolytica trophozoites using a 5′-phosphate dependent method; 802 small RNA sequences were obtained, 544 of which were unique, and 342 of which mapped to the E. histolytica genome sequence (Table S1 and Figure S2). The majority of the cloned RNAs were smaller in size with a peak size at ∼16 nt. However, high resolution Northern blot analysis demonstrated that small RNAs cloned in a 5′-phosphate dependent manner all mapped at ∼27–32 nt, regardless of the size at which they were cloned, indicating that we had cloned partial degradation products of the 27 nt small RNAs (Figure S3). In contrast, small RNAs cloned by the 5′-phosphate independent manner were detected by Northern blot analysis at sizes matching the sizes of the cloned products (Figure 2B). This confirms that the full length E. histolytica 27 nt small RNAs with 5′-polyP termini are not efficiently cloned using a 5′-phosphate dependent approach, while full length small RNAs can be cloned using a 5′-phosphate independent method. To confirm that we had not cloned degradation fragments of larger transcripts, we used Northern blot analysis on total RNA using probes for a number of cloned small RNAs and did not detect any signal >200 nt (data not shown). Thus, the majority of small RNAs that map antisense to genes do not appear to be degradation products of larger antisense transcripts.
In order to confirm the 5′ and 3′ structure of the cloned 27 nt small RNAs, we tested individual small RNAs that had been identified in our library using biochemical approaches. The E. histolytica 27 nt cloned small RNAs are likely to have a free 3′-OH, as they are susceptible to loss of a single base and generation of a smaller RNA species following a β-elimination reaction, as indicated by faster migration following denaturing PAGE (Figure 3A). A 32P-PNK labeled synthetic RNA 18mer (with a 3′-OH terminus) also loses a base after a β-elimination reaction, as indicated by faster migration. In agreement with previous analyses of the whole 27 nt population, individual 27 nt small RNAs are initially resistant to Terminator treatment (Figure 3B), but become sensitive to Terminator exonuclease activity after CIP treatment followed by addition of a 5′-monoP by PNK (Figure 3C). The control synthetic 18mer RNA species, with a 5′-monoP is degraded by Terminator treatment (Figure 3B). We tested three different 27 nt small RNAs and found the same end structures for all three (some data not shown). These results confirmed that small RNAs cloned from the 27 nt population have a 5′-polyP and 3′-OH structure. Thus, the 27 nt small RNAs in E. histolytica are likely not Dicer products (based on 5′-polyP termini). Instead, the structure of these RNAs matches that of secondary siRNAs in C. elegans, the only siRNA species identified to date with 5′-polyP termini, and which are known to be products of RdRP processing [7],[8]. One major difference between the E. histolytica and C. elegans 5′-polyP small RNAs is the size of the populations: the E. histolytica population is ∼27 nt, whereas the C. elegans 5′-polyP small RNAs are ∼22 nt. The molecular mechanism that determines the sizes of these small RNAs in either system is not currently known.
In order to determine if the 27 nt small RNAs are involved in a silencing complex, we tested to see if they associate with E. histolytica Piwi-related protein (EhPiwi-rp) (EHI_125650). Of the three genes in the E. histolytica genome that are predicted to contain PIWI domains, this is the only one that is highly expressed in trophozoites [21]. A construct with N-terminal Myc tagged EhPiwi-rp was generated and transgenic parasites selected. Western blot analysis demonstrated that anti-Myc antibody detected a protein of the appropriate molecular mass specifically in the Myc-EhPiwi-rp transgenic strains (Figure 4A). Immunoprecipitation (IP) experiments with α-Myc antibody were performed and small RNAs of ∼27 nt were specifically observed in the α-Myc IP of the Myc-EhPiwi-rp parasite cell line (Figure 4B). The 27 nt small RNA population was significantly enriched in the α-Myc EhPiwi-rp immunoprecipitated sample compared to the starting sample (data not shown). No small RNAs immunoprecipitated with a number of controls including untransfected parasites or transgenic E. histolytica strains expressing one of a number of Myc-tagged proteins (Green Fluorescent Protein, EhKinase, or EhRNaseIII) (Figure 4B). These data indicate that 27 nt small RNAs are in a complex specifically with EhPiwi-rp.
In order to determine the 5′ and 3′ structure of the small RNAs that immunoprecipitated with Myc-EhPiwi-rp, we performed IP with α-Myc and used the biochemical approaches outlined earlier. In agreement with our analysis of the total 27 nt population, the small RNAs in the IP sample could be successfully labeled at the 3′ end using RNA ligase and 32pCp, consistent with a free 3′-OH. As before, the small RNAs in the IP sample were resistant to labeling at the 5′ end using PNK, unless first dephosphorylated (CIP+PNK), and were able to be efficiently labeled by capping enzyme and 32pGTP (Figure 4C). These data indicate, that as with the bulk of the 27 nt population, the majority of the 27 nt small RNAs that associate with EhPiwi-rp have 5′-polyP termini.
In order to determine the composition of the small RNAs that IP with EhPiwi-rp, we generated a small RNA library from the EhPiwi-rp immunoprecipitated sample using a 5′P-independent method. A total of 309 small RNAs that mapped to the E. histolytica genome were cloned (Table S1 and Figure S2). As expected the amount of rRNA or tRNA contamination was minimal with only 6% of small RNAs mapping to these elements. A total of 36% of small RNAs mapped antisense to genes, 25% mapped to intergenic regions, 10% mapped as mixed hits, 18% mapped sense to genes, 5% mapped to repetitive regions, and 0.32% mapped to retrotransposons elements. The overall pattern, with the greatest percentage of small RNAs mapping antisense to ORFs, was the same as seen in the other E. histolytica small RNA libraries (Figure S2). There was also significant overlap in the genes targeted by the antisense small RNAs cloned from the non-IP and the IP libraries (Figure 5). Of the 83 genes with antisense small RNAs from the non-IP libraries, 36 genes also had antisense small RNAs cloned from the EhPiwi-rp IP library (p-value = 1.7e−61). Additionally, three small RNAs cloned from the IP library were highly similar (identical except for a few nucleotides at the 5′ or 3′ end) to small RNAs cloned from total RNA. Furthermore, one small RNA that was cloned from total RNA and that was previously shown to be detectable by Northern blot analysis (small RNA EHS-ID-27-1-30) (Figure 2B) was also cloned from the EhPiwi-rp IP library (EHS-IP-2-54 and EHS-IP-3-253). In C. elegans 5′-polyP small RNAs are in complex with an Argonaute protein, CSR-1 or with the Piwi protein, NRDE-3 for gene silencing [11],[22]. The E. histolytica Piwi-related protein has two conserved domains (PAZ-piwi like and Piwi), and the Piwi domain appears to contain the conserved residues necessary for slicer activity [6]. Thus, the fact that E. histolytica 27 nt 5′-polyP small RNAs associate with EhPiwi-rp strongly suggests that these small RNAs could be part of a silencing complex in E. histolytica.
In order to determine the genomic regions to which cloned small RNAs mapped, we performed BLAST analysis of the small RNA sequences to the E. histolytica genome sequence. No significant differences were identified in the overall characteristics of the genomic regions to which the small RNAs mapped, regardless of whether they were cloned in a 5′-phosphate dependent or 5′-phosphate independent manner (Figure S2). Of the 847 unique small RNAs that were cloned, 38% were identical to ribosomal and tRNA sequences, 25% mapped antisense to predicted open reading frames (ORFs), 16% mapped to intergenic regions, 8% matched multiple genomic loci, 9% matched the sense strand of ORFs, 4% mapped to repetitive regions, and 0.5% mapped to retrotransposon elements. Sequences from ribosomal RNAs, tRNAs, and sense tags to highly expressed loci are often found in small RNA libraries. For these RNAs we have no means to distinguish between biological effects and degradation products; thus, they were not further considered in our analysis. The paucity of matches to retrotransposons, especially in light of the large number of retrotransposons in E. histolytica was in contrast to other parasitic systems such as G. intestinalis and T. brucei where the majority of small RNAs appear to be derived from retrotransposons [15],[23],[24],[25].
Since secondary siRNAs in C. elegans are largely antisense to coding regions, we focused our analysis on E. histolytica small RNAs that map antisense to predicted genes (Figure 6 and Table S2). Small RNAs cloned from either the 5′-phosphate dependent or 5′-phosphate independent manner and that mapped antisense to predicted ORFs were analyzed. A total of 214 unique small RNAs mapped antisense to predicted genes and a substantial portion (∼55%) mapped to the 5′ end of ORFs or very close upstream of the start codon, in the presumptive 5′-UTR (Figure 6A and 6C). Upon closer inspection, we identified that for several genes multiple, non-overlapping, antisense small RNAs were identified. The limited sequence data does not allow us to address the issue of “phasing” previously noticed for secondary siRNAs in C. elegans, however, the multiple non-overlapping small RNAs identified for a given gene suggests that this may also occur in E. histolytica [7],[8]. The small RNAs that map antisense to genes appear to be an abundant population since ∼4% of the small RNAs were cloned twice. An alternate explanation is that certain small RNAs were cloned multiple times due to biases with cloning or PCR amplification.
We identified genomic regions in which clusters of genes all had antisense small RNAs mapping to them (Figure 6A and 6C). Given the identification of genomic clustering without performing deep sequencing it would appear that this phenomenon is relatively common in E. histolytica. Genomic clusters of genes are coordinately regulated in response to stress and development in E. histolytica [21]. Whether the siRNA mechanism controls expression of adjacent genomic loci is not currently known. Of the 120 genes, which had small RNAs mapping antisense to them, 15% also had small RNAs, which mapped in the sense orientation (p = 4.8e−19) (Table S3). In all cases, the sense and antisense small RNAs were not complementary, however, as of yet we cannot make generalizations about this phenomenon. In C. elegans secondary siRNAs, both sense and antisense tags were seen for genes undergoing amplified silencing [8].
In C. elegans, secondary siRNAs are generated by RdRP processing of mature RNA transcripts; thus, secondary siRNAs that span exon-exon junctions have been identified [7],[8]. In our work, no antisense small RNAs were identified at exon-exon junctions, but our limited sequencing makes this negative result difficult to interpret. We did identify small RNAs that mapped to intergenic regions in a genomic locus that had antisense small RNAs to adjacent genes. Since the E. histolytica genome is very compact (∼9,900 predicted genes in a 24 Mb genome), one possibility is that the small RNAs that map to intergenic regions represent extension of amplified silencing that occurs downstream of a trigger, a phenomenon noted in C. elegans [7],[8],[15]. Alternatively, the intergenic small RNAs could map to unannotated genes.
Forty eight unique small RNAs mapped to multiple genomic regions, including antisense to predicted ORFs (Figure 6A). Small RNAs that mapped to multiple genomic loci are likely those that map to closely related gene families or repetitive regions. The overall features of these small RNAs: mapping to the 5′ ends of genes and association with low gene expression of the cognate gene (see below) were similar to the small RNAs that exclusively map antisense to genes.
The cumulative data on the E. histolytica 27 nt small RNAs (5′ polyphosphate termini, antisense orientation, and bias towards 5′ ends of genes) makes them sufficiently similar to secondary siRNAs involved in amplification of silencing in C. elegans [7],[8] so that we will subsequently refer to this fraction of small RNAs as siRNAs.
In order to determine if the 27 nt 5′-polyP small RNAs that map antisense to genes have any effect on gene expression of the ORF to which they map, data from whole genome expression profiling were analyzed. A custom short oligonucleotide microarray (Affymetrix) has probes for 9,435 amebic genes and was previously used to generate expression profiles of E. histolytica trophozoites from three strains (HM-1:IMSS, 200:NIH, and Rahman) [21]. The data demonstrate that the genes to which antisense small RNAs map had extremely low expression values in E. histolytica HM-1:IMSS trophozoites, the strain and stage of the parasite from which they were cloned (Figure 6B and Table S3). The median normalized expression value for all genes on the array was 0.65±0.16 (median±standard error). The expression of genes to which antisense small RNAs mapped was 0.04±0.09 (median±standard error) (p<0.0001 compared to expression data for all genes on the array). The expression of genes that had small RNAs from the “mixed” category mapping antisense to them was 0.04±0.02 (median±standard error) (p<0.0001 compared to expression data for all genes on the array). A total of 6 genes to which antisense small RNAs map were expressed, but 4 of these were represented by probe sets that completely cross-hybridize with other genes (as indicated by a probe set identifier that ends with _s_at), making it difficult to address the expression of a specific gene. Considering that >80% of genes were expressed under trophozoite conditions [21], the genes to which antisense small RNAs map have significantly disproportionately low expression profiles. A subset of genes have both antisense small RNAs and sense small RNAs that mapped to them. In these instances, the location of the sense small RNAs was biased towards the middle and 3′-end of the gene, in contrast to the 5′ bias of the antisense small RNAs (data not shown). In instances of C. elegans amplified silencing, sense small RNAs were also identified [8]. Consistent with that data, genes that have antisense and sense small RNAs have low gene expression level (Table S3).
We analyzed the expression profiles of genes to which antisense small RNAs map from trophozoites of other strains including E. histolytica 200:NIH and E. histolytica Rahman [21]. The data demonstrate that some genes to which antisense small RNAs map are expressed in trophozoites of other E. histolytica strains (Table S3). Of the 101 genes with antisense small RNAs, 73 were not expressed in any of the three E. histolytica strains, 11 were expressed in one strain, 11 were expressed in two strains, and 6 were expressed in all strains (4 of these were represented by cross-hybridizing probes).
In order to determine if there was a correlation between the presence of a small RNA and the corresponding expression level of its putative target gene (the ORF to which it maps), we performed Northern blot analysis of small RNAs where the associated gene had variable expression between amebic strains. We identified that for a given gene, small RNAs were detectable in E. histolytica strains with low gene expression but were not detectable in E. histolytica strains with high gene expression (Figure 7). The expression data for these genes were confirmed using semi-quantitative reverse transcriptase polymerase chain reaction (RT-PCR) and in all cases the results matched the array data. We noted that some small RNAs mapped at sizes larger than 30 nt. Further analysis of E. histolytica trophozoite small RNAs of >30 nt indicates that some ∼32 nt small RNAs have 5′-polyP structure (Figure S3 and data not shown). In selected cases, we sequenced the relevant genomic region in E. histolytica 200:NIH or E. histolytica Rahman strains and confirmed that the regions encompassing the small RNAs were identical among the three E. histolytica strains (data not shown). Thus, the lack of detection of a small RNA in a given amebic strain was not due to genomic sequence divergence in this strain. The inverse correlation between the E. histolytica 27 nt 5′-polyP small RNAs and gene expression level and the association of these small RNAs with EhPiwi-rp strongly suggests parallel functions of amebic and C. elegans siRNAs in mediating target silencing.
The work presented herein suggests that the process of gene silencing, mediated by small RNAs with 5′-polyphosphate termini, is evolutionarily conserved in the single celled parasitic eukaryote Entamoeba histolytica. We have identified an abundant repertoire of ∼27 nt small RNAs in E. histolytica with features reminiscent of secondary siRNAs in C. elegans, including 5′-polyP termini and antisense orientation to genes. The 27 nt small RNAs are in association with EhPiwi-related protein and an inverse correlation between small RNA and gene expression was noted suggesting that these small RNAs could mediate gene silencing. This identification of small RNAs with 5′-polyP termini in E. histolytica indicates that this mechanism of gene regulation is functional in single celled eukaryotes.
An increasing number of distinct small RNA species are being identified. The majority are processed by Dicer, an RNaseIII endonuclease, which generates 5′-monophosphate termini, a feature typical of siRNAs, miRNAs, tasiRNAs, tncRNA, and scnRNA [6]. However, recently small RNAs that could be generated independently of Dicer processing (rasiRNA, piRNA, and secondary siRNA) have been described [6],[7],[8]. To date, the only siRNA species proven to have 5′-polyphosphate termini are the siRNAs from C. elegans. These small RNAs are generated by RdRP amplification of an initial trigger siRNA [7],[8] and preferentially associate with an Argonaute protein, CSR-1, to mediate gene silencing via slicer activity [11]. Importantly, polyphosphate small RNAs are more effective than primary monophosphate small RNAs in inducing slicer activity of CSR-1 and are thus more robust at cleaving target mRNA [11].
In E. histolytica 27 nt RNAs with 5′-polyphosphate termini associate with EhPiwi-related protein and appear to silence genes in a strain-specific manner. We observed that genes to which these antisense small RNAs map are not expressed under trophozoite conditions, in the strain and stage of the parasite from which the small RNAs were cloned. However, a number of these genes are expressed in other parasite strains. An inverse correlation between small RNA and gene expression, and the association of the 27 nt 5′-polyP small RNAs with a Piwi-related protein, strongly suggests that the 27 nt small RNAs mediate gene silencing in E. histolytica. Direct proof of the roles of small RNAs in regulating amebic gene expression has been hampered due to significant issues with genetic manipulation of E. histolytica strains (unpublished data, H. Zhang, GM Ehrenkaufer, and U. Singh).
One difference between E. histolytica and C. elegans secondary siRNAs is their size: the C. elegans secondary siRNAs are ∼22 nt, whereas the E. histolytica counterparts are ∼27 nt. What determines the sizes of the secondary small RNAs in either species is not known; whether the size reflects an inherent feature of the RdRP or whether other endonucleases cleave secondary siRNAs needs further investigation. In E. histolytica, the siRNAs are a highly abundant endogenous population and were readily identified by limited sequencing. The molecular mechanisms that generate these abundant small RNAs is not known but some architectural features of the E. histolytica genome, including small intergenic regions and an AT rich genome with cryptic TATA-like promoter elements potentially mediating bi-directional transcription, may be contributing factors. One possibility is that in E. histolytica abundant secondary siRNAs are needed to deal with aberrant transcripts, which occur due to specific features of the ameba genome as outlined above.
Our preliminary analysis of the E. histolytica trophozoite 22 nt and 16 nt small RNA populations indicates that they likely do not have a 5′-polyP termini, however further investigations will be needed to more clearly define the features of these small RNA populations (unpublished data, H. Zhang and U. Singh). Thus far, we have not identified a small RNA species consistent with the trigger siRNAs in C. elegans (with a 5′-monoP) likely due to the rarity of these species and our lack of deep sequencing [7],[8]. Elucidation of the mechanism by which the 5′ polyP small RNAs are produced should shed some light on the nature and/or existence of such a trigger siRNA in E. histolytica. Based on functional similarity, we presume that similar trigger siRNAs exist in E. histolytica, however, whether other small RNAs or other structures function as triggers for the secondary siRNAs in E. histolytica is not currently known.
In summary, Entamoeba histolytica, a protozoan parasite, is the first single celled eukaryote in which gene silencing via siRNAs with 5′-polyphosphate termini has been described. Based on analogies with the amplified silencing pathway as described in C. elegans, our data suggest that this silencing pathway is broadly evolutionarily conserved.
Entamoeba histolytica trophozoites (HM-1:IMSS, 200:NIH, and Rahman strains), E. dispar SAW760 and E. invadens IP-1 were grown under standard conditions as previously published [21],[23],[26]. Amebae were harvested in mid log phase and small RNA extracted using a mirVana kit (Ambion). N-terminal Myc tagged constructs for EhPiwi-rp, EhRNaseIII, EhRdRP, and GFP were generated. Briefly, full-length coding regions were PCR amplified and cloned into a vector to express a N-terminal Myc tag [27]. E. histolytica HM-1:IMSS parasites were transfected using previously published protocols, stable transfectants selected with 12–24 µg/ml G418, and western blot analysis performed using standard protocols [23],[28]. Anti-Myc antibody (Cell Signaling) was used at 1∶1000 dilution; Anti-Lgl antibody (kind gift of William Petri) was used at 1∶50 dilution.
In order to visualize small RNAs in E. histolytica trophozoites, we fractionated 100 µg of total RNA with a YM100 column, resolved the sample on a 12% denaturing polyacrylamide gel, and stained the gel with SYBR gold. Small RNA cloning was based on two published protocols [8],[10]. For the 5′-P independent method, 100 µg of small RNA enriched sample was resolved on a denaturing 12% polyacrylamide gel, the 27 nt RNA fraction gel extracted, ligated to a 3′ adapter oligonucleotide (5′ rAppCTGTAGGCACCATCAAT/3ddC/ 3′) (3′-terminal dideoxy-C (ddC) base) (RT, 4 hours) and the product gel purified. The material was subjected to RT (Superscript II, 42°C, 30 min), treated with Exonuclease-I (37°C, 1 hour), a second 3′ ligation performed (5′ rAppCACTCGGGCACCAAGGA/3ddC/-3′) (RT, 4 hours), and the product gel purified. The material was PCR amplified using the adaptor primers, the final PCR products concatamerized and cloned into the pCR2.1-TOPO vector (Invitrogen) vector for sequencing. For the 5′-phosphate dependent method, 150 µg of small RNA enriched material was fractionated on a denaturing 12% polyacrylamide gel, RNA from 15–30 nt purified, and ligated to the 3′ adapter oligonucleotide (5′-rAppCTGTAGGCACCATCAAT/3ddC/-3′, 3ddc = 3′-terminal dideoxy-C base) and the 5′ adapter oligonucleotide (5′-TCGTAGGCACCTGaaa-3′; uppercase = DNA, lowercase = RNA). After reverse transcription, two rounds of PCR (32 cycles total) were performed, the PCR products concatamerized, cloned and sequenced as outlined above. The library generated from the EhPiwi-rp immunoprecipitated sample was made in a 5′-independent manner (personal communication, Sam Gu and Andy Fire). Briefly, RNA was extracted from immunoprecipitation, directly ligated to the 3′ adapter oligonucleotide, purified, treated with CIP and PNK, and ligated to the 5′ adapter oligonucleotide. Subsequent steps were as for the 5′-P dependent cloning method as outlined above.
The small RNA sequences were extracted and BLASTed against the E. histolytica HM-1:IMSS genome sequence (http://www.tigr.org/tdb/e2k1/eha1/) (http://pathema.jcvi.org/cgi-bin/Entamoeba/PathemaHomePage.cgi) using search settings for short and nearly exact matches (expect = 1,000, Word size = 7). Additionally, all E. histolytica sequences annotated as belonging to the SINE/LINE retrotransposons were downloaded and analyzed by BLAST analysis. Sequence tags with 100% match, 1 mismatch, or up to 2 mismatches (1 terminal and 1 internal mismatch) to the E. histolytica genome sequence were analyzed further [29]. Antisense small RNAs are categorized as follows: map to the first 25% of a coding region (map to the 5′ of a gene); map to the last 25% of a coding region (map to the 3′ of a gene); and map to the middle 50% of a coding region (map to the middle of a gene). Small RNAs were considered to have genomic clustering if ≥ 2 adjacent genes had antisense small RNAs. The untranslated region (UTR) of a gene was defined as the 50 bp region upstream or downstream of a predicted start or stop codon respectively.
High resolution Northern blot analysis was done using standard protocols [30]. Briefly, 10 µg–100 µg of RNA was separated on a denaturing 12% polyacrylamide gel, transferred to a membrane, probed with end-labeled 32P-labeled oligonucleotides in perfectHyb buffer (Sigma) at 37°C and washed using low (2X SSC, 0.1% SDS at RT for 15 min) and medium (1X SSC, 0.1% SDS at 37°C for 15 min) stringency conditions. Standard Northern blot analysis was done using 10 µg of total RNA resolved on a denaturing 1.2 % agarose gel, probed with end-labeled 32P-labeled oligonucleotides in perfectHyb buffer (Sigma) at 37°C and washed using low stringency conditions according to manufacturer's instructions.
Biochemical analysis of small RNAs were performed using standard methods [7],[8],[11]. Briefly, the structure of the 3′ termini was determined with an RNA ligation reaction using T4 RNA ligase (NEB) and α-32P pCp (RT; 2 hours). Additionally, potential modifications at the 3′-OH termini were identified in a β-elimination reaction by treating 10 µg of small RNA with sodium periodate (25 mM, RT, 10 min), followed by heating to 45°C for 90 min. The 5′-termini were analyzed by CIP, PNK, Capping enzyme, and Terminator treatment [7],[8],[11]. Calf intestinal alkaline phosphatase (NEB) was used to treat RNA (37°C; 1 hour), followed by phenol/chloroform extraction. T4 Polynucleotide Kinase (NEB) was used to treat RNA (37°C; 1 hour). Capping was assessed with Guanylyltransferase (Ambion), the vaccinia virus capping enzyme, which was used to add α-32P-GTP to RNA product (37°C; 1 hour). For Terminator susceptibility, the RNA sample was treated with Terminator enzyme (30°C; 1 hour). A control sample (synthetic RNA 5′-end labeled with 32P and a 3′-OH terminus) was added to the RNA material, the combined sample resolved on a 12% polyacrylamide gel, and probed with a radiolabeled probe to detect the small RNA of interest.
For immunoprecipitation experiments, α-Myc antibody was incubated with parasite lysate (2 hours, 4°C) washed x2 and pelleted. RNA was isolated (mirVANA kit), labeled with PNK, 32P-pCp, and Capping enzyme as outlined above and resolved on a 12% denaturing polyacrylamide gel.
A custom Affymetrix DNA microarray for E. histolytica with probe sets representing 9,435 ORFs has previously been used to determine the expression profile of E. histolytica trophozoites of HM-1:IMSS, 200:NIH, and Rahman strains [21]. For RT-PCR, total RNA was extracted from log phase E. histolytica parasites of the appropriate strain using a mirVana reagent kit (Ambion). cDNA was made using SuperScript II Reverse Transcriptase kit (Invitrogen), including a DNAse treatment before RT. PCR was performed with serial 10-fold dilutions of cDNA and a negative RT control was included with each reaction. Four genes, 152.t00021, 18.t00040, 77.t00031, and 73.t00015 were analyzed by RT-PCR. Primers used are listed in Table S4. The ssRNA gene was used as a loading control [21].
Statistical analyses were done using an unpaired Student's t-test (comparison of expression data for genes with antisense small RNAs compared to all genes in the genome) or a hypergeometric distribution in R downloaded from the BioConductor project (http://www.bioconductor.org) (comparison of sense small RNA distribution).
|
10.1371/journal.pntd.0006553 | Risk factors for Cryptosporidium infection in low and middle income countries: A systematic review and meta-analysis | Cryptosporidium infection causes gastrointestinal disease and has a worldwide distribution. The highest burden is in developing countries.
We sought to conduct a systematic review and meta-analysis to identify Cryptosporidium risk factors in Low and Middle Income countries (LMICs).
Medline Ovid and Scopus databases were searched with no restriction on year or language of publication. All references were screened independently in duplicate and were included if they presented data on at least 3 risk factors. Meta-analyses using random effects models were used to calculate overall estimates for each exposure.
The most frequently reported risk factors in the 15 included studies were overcrowding, household diarrhoea, poor quality drinking water, animal contact, open defecation/ lack of toilet and breastfeeding. The combined odds ratio for animal contact was 1.98 (95%CI: 1.11–3.54) based on 11 studies and for diarrhoea in the household 1.98 (95%CI: 1.13–3.49) based on 4 studies. Open defecation was associated with a pooled odds ratio of 1.82 (95%CI: 1.19–2.8) based on 5 studies. Poor drinking water quality was not associated with a significant Cryptosporidium risk, odds ratio 1.06 (95%CI: 0.77–1.47). Breastfeeding was protective with pooled odds ratio 0.4 (95%CI: 0.13–1.22), which was not statistically significant.
Based on the included studies, crowded living conditions, animal contact and open defecation are responsible for the majority of Cryptosporidium cases in LMICs. Future studies investigating Cryptosporidium risk factors should have a good study design and duration, include appropriate number of cases, select suitable controls, investigate multiple relevant risk factors, fully report data and perform multivariate analysis.
| Cryptosporidium is a parasite that causes diarrhoea and is transmitted through faecal contamination of water and food. Though it occurs in developed nations, it is much more prevalent in developing countries and is associated with high mortality in children under 2 years old. In this review, we looked at published studies on factors that increase the risk of contracting cryptosporidiosis in Low and Middle Income Countries. These factors could be targeted to limit the transmission of the disease. Based on the selected studies, the most important risk factors identified were contacts with animals and presence of infected people in the household. Open defecation was also contributing to the risk of infection by this parasite transmitted through the faecal oral route. Breastfeeding was protective from the infection. Poor drinking water was not responsible for causing the disease.
| Cryptosporidium is a protozoan parasite with a worldwide distribution infecting humans and animals. In high income countries, Cryptosporidium occasionally causes sizable outbreaks due to contaminated water supplies or food sources [1]. In Low and Middle Income Countries (LMICs), cryptosporidiosis is much more prevalent and is associated with a significant burden of gastrointestinal disease. Cryptosporidiosis is highly prevalent in early childhood, with 45% of children infected before the age of 2 [2]. More recent studies showed higher prevalence rates: 77% in slum dwelling Bangladeshi children [3] and 97% in children under 3 years from a birth cohort in Southern India [4]. A study in children under 2 years old estimated 2.9 million and 4.7 million Cryptosporidium infections in sub-Saharan Africa and South Asia, respectively [5]. Cryptosporidiosis is associated with watery diarrhoea persisting for over 2 weeks. This chronicity increases the vulnerability of children in LMICs and is the result of interplay of immune naivety, malnutrition and HIV infection [2]. Cryptosporidium infection in children in LMICs is associated with malnutrition, stunted growth, and cognitive impairment [6]. Cryptosporidiosis exacerbates malnutrition and is more severe in malnourished subjects [7]. Cryptosporidium is the second cause of severe diarrhoea in children under 5 years old in sub-Saharan Africa and south Asia and the leading cause of mortality in children aged 12–23 months [8]. In immunocompromised people such as HIV-positive and transplant patients, cryptosporidiosis is more severe and could result in high mortality rates [1]. The disease burden in both developed and developing countries is likely to be underestimated, due to a large number of asymptomatic or self-limiting diarrhoeal cases, lack of systematic diagnosis of etiologic agent of diarrhoeal disease and reliance on microscopy for routine clinical detection, which is associated with low specificity and sensitivity.
Due to the significant burden of cryptosporidiosis, several studies attempted to elucidate Cryptosporidium transmission pathways and risk factors [1, 7, 9–11]. The two Cryptosporidium species causing the majority of human infections are C. parvum and C. hominis. The former is transmitted mainly through a zoonotic cycle between humans and animals, while the latter is predominately anthroponotic. The main Cryptosporidium risk factors were summarised in three reviews [7, 10, 11] and are related to drinking contaminated water, contact with infected animals or humans (particularly children), consumption of contaminated food, recreational use of contaminated water and travel to disease endemic areas. However, these reviews focused on developed countries and one on the USA. To our knowledge, no such review of cryptosporidiosis risk factors in LMICs exists. Therefore, we attempted to address this gap and conducted a systematic review and meta-analysis of higher quality studies investigating risk factors for Cryptosporidium infection in LMICs.
The methodology and reporting were in accordance with the “Preferred Reporting Items for Systematic Reviews and Meta-Analyses” (PRISMA) (S1 File). Medline Ovid and Scopus databases were searched with no restriction on year or language of publication up to 12th January 2017. The search strategy was limited to title/ abstract/ keyword using the following MeSH terms/ keywords: (Cryptosporidium OR cryptosporidiosis) AND (risk factor OR case control OR cohort OR infection OR sporadic OR prevalence). Reference lists from relevant papers were screened for additional eligible articles.
All references were screened by title and abstract independently in duplicate by MB and EK. Studies investigating Cryptosporidium transmission and risk factors were considered for full text analysis and data extraction. Eligibility disagreements were resolved by discussion. Abstracts without full text or complete results section, such as conference proceedings, were excluded. Only studies from LMICs as defined by the Official Development Assistance (ODA) of the Organisation for Economic Co-operation and Development (OECD) were included.
In order to restrict the analysis to higher quality studies, additional inclusion criteria were applied at full text analysis stage: at least 20 Cryptosporidium infections were reported and the study assessed at least three relevant risk factors. Examples of irrelevant risk factors: age, gender, rural/urban living, stunting, malnutrition (arguably potential cause and consequence of cryptosporidiosis), household income and type of dwelling, as these could not be directly targeted by preventive public health strategies.
For each article, the following information was extracted: location of the study, duration, type of study, Cryptosporidium detection method, age range of participants, number of cases, number of controls (if applicable), selection criteria for cases and controls, risk factors (exposures) investigated and odds ratios (or relative risk or hazard ratio) as reported by the authors or calculated from data presented in the paper (when available).
The Newcastle-Ottawa scale (NOS) was used to provide a quality assessment score for all included studies [12]. Case control studies were judged across three domains: selection of cases and controls, comparability of cases and controls and ascertainment of exposure. Cohort studies were judged across three domains: selection of cohorts, comparability of cohorts and assessment of outcome. Cross-sectional studies were assessed as per case control studies. Details of subdomains assessed within each criterion are provided in S2 File. A study is awarded a maximum of one star for each subdomain. For this systematic review, only one star was given for the comparability domain as opposed to two possible stars for the traditional Newcastle Ottawa Scale. Therefore, a maximum of 4 stars for selection, 1 star for comparability and 3 stars for exposure/outcome could be awarded, totalling 8 stars if all factors included in the NOS were unlikely to introduce bias. The studies were considered of high quality if NOS score was 6–8 stars, moderate quality for a score of 3–5 stars and of poor quality if the NOS score was 0–2.
The risk factors identified in each study were pooled in a table and categorised. The number of risk factors and the proportion achieving statistical significance were noted. Both univariate and multivariate risk factors were extracted. When at least four papers reported on a particular risk factor, a meta-analysis was performed to calculate a combined random effect odds ratio for this exposure using Reference Manager software (RevMan) [13]. Any available (or calculated) risk factor was included in the meta-analysis regardless of significance. As the majority of studies reported on univariate estimates of risk factors, only these were considered when calculating pooled odds ratios (ORpooled). Funnel plots generated using RevMan were used to assess publication bias through visual assessment.
Population attributable fraction (PAF) was calculated using the formula PAF = Pepooled x [(ORpooled-1)/ ORpooled)] [14]. Pepooled = proportion of source population exposed to the risk factor, was calculated if the number of exposed cases and total number of cases was available from at least 50% of the studies used to calculate the pooled odds ratio. Pepooled calculations were performed in OpenMeta [Analyst] software [15], entering the data as untransformed proportions and performing a binary random effects meta-analysis.
The combined search retrieved 3830 studies, which was reduced to 3356 after duplicate removal (Fig 1). Based on title screening, 627 papers were retained for potential inclusion. An additional study was found by screening reference lists. Therefore, 628 studies were subjected to abstract screening, of which 105 were retained for full text analysis (Fig 1). 523 papers were excluded because of the following reasons: not a developing country, no Cryptosporidium specific risk factors, cryptosporidiosis outbreak/ case report, review, cryptosporidiosis in animals, treatment studies, seroprevalence surveys, detection in water/rivers, and cryptosporidiosis in immunocompromised patients.
A total of 15 papers were included for meta-analysis (Fig 1). Table 1 summarises the characteristics of these studies. Six case control studies, 4 cohort studies and 4 cross-sectional studies are included, together with one paper that used case control design for hospital study and cross-sectional design for community study. The quality assessment using NOS showed that 7/15 studies were of high quality, while 6 studies were of medium quality (Table 1 and S2 File). 2 studies were of low quality.
For each study, odds ratios (or equivalent) for each relevant risk factor were extracted and are detailed in S1 Table. The most frequently reported Cryptosporidium transmission pathways are: i) person to person (related to living in overcrowded accommodation, diarrhoea in the household or attending nursery), ii) water related (drinking poor quality water or contact with contaminated water bodies for hygiene or leisure purposes), iii) environmental transmission (animal contact, feces contaminated soil, farming or sewage proximity), iiii) inadequate sanitation (lack of toilet and/or open defecation) and iiiii) hygiene (hand, food or household). Details of transmission pathways are provided in S2 Table. Though not a transmission pathway, breastfeeding was investigated in 10/15 studies and as such was included in the analysis. When ≥ 4 studies reported on a risk factor, it was then included in the meta-analysis. This was the case for overcrowding, household diarrhoea, poor quality drinking water, animal contact, open defecation/ lack of toilet and breastfeeding (S2 Table).
Animal contact was investigated in all 15 included studies. The type of animal species (when available) is provided in S1 Table. Information needed for meta-analysis could be extracted from 11/15 studies. While the majority of studies found that contact with animals is associated with increased risk of cryptosporidiosis [16–22], a few reported a protective effect. The combined odds ratio was 1.98 (95%CI: 1.11–3.54) p = 0.02 (Fig 2). Heterogeneity was substantial with I2 score of 87%.
The impact of non-piped drinking water on cryptosporidiosis was investigated in all 15 studies. Meta-analysis was possible for 10 studies. The combined odds ratio was 1.06 (95%CI: 0.77–1.47), which was not significant (Fig 3). This was due to conflicting results between studies, with 6 studies reporting that non-piped water is a risk factor [17, 18, 20, 21, 23, 24], while 4 studies considered it to be protective [19, 22, 25, 26]. Heterogeneity was moderate (I2 = 33%).
This risk factor was investigated in 7 studies. The combined odds ratio from 5 studies was 1.82 (95%CI: 1.19–2.8) p = 0.006 (Fig 4). Heterogeneity was substantial (I2 = 81%). Despite the relatively moderate combined risk associated with open defecation, all studies consistently showed increased cryptosporidiosis risk.
Overcrowding was reported as a risk factor for Cryptosporidium infection in 7 studies. The combined odds ratio based on 5 studies was 1.37 (95%CI: 1.07–1.75) p = 0.01 (Fig 5). Heterogeneity was substantial (I2 = 72%). Only one study reported that overcrowding is protective [23].
Household diarrhoea was a cryptosporidiosis risk factor in 4 studies, all of which were included in the meta-analysis. The combined odds ratio was 1.98 (95%CI: 1.13–3.49), which was statistically significant (Fig 6). Heterogeneity was moderate (I2 = 38%). One study reported that diarrhoea in the household was protective [18].
Breastfeeding was investigated in 10 studies. Meta-analysis was restricted to 5 studies that provided the required information. The combined odds ratio was 0.4 (95%CI: 0.13–1.22) suggesting a protective overall effect (Fig 7). However, this was not statistically significant. Heterogeneity was substantial (I2 = 83%). Only one study reported that breastfeeding was conducive to acquiring cryptosporidiosis in infants [21].
For each exposure, publication bias was assessed using funnel plots. All funnel plots had a symmetrical shape suggesting minimal publication bias (S1 Fig). Calculation of population attributable fraction (PAF) was possible for 4/5 risk factors. Breastfeeding was protective and therefore no PAF was calculated. Crowding was responsible for 18% of cases (95%CI 4–29%) based on 3 studies (Table 2). Open defecation was attributable to 17% of cases (95%CI 6–25%) (based on 4 studies), while animal contact accounted for 25% of cases (95%CI 5–36%) (8 studies). Poor drinking water quality was responsible for 2% of cases (95%CI—10%, 11%) (based on 8 studies).
This systematic review aimed to identify the most frequently reported risk factors for Cryptosporidium infection from LMICs based on good quality studies. Although the search strategy retrieved > 3000 papers, only 15 studies were of acceptable quality warranting inclusion. The pitfall of this strategy is the exclusion of relevant risk factors and decreasing the power of meta-analysis by including fewer studies. Nevertheless, this review identified six risk factors that are likely to be the main drivers of Cryptosporidium infection in LMICs. Animal contact had the highest combined odds ratio 1.98 (95%CI: 1.11–3.54), which was statistically significant. This is in accordance with reviews from developed countries where animal contact/ farm visits/ petting zoo visits were significantly associated with acquiring cryptosporidiosis [7, 11]. Diarrhoea in the household was associated with a similar Cryptosporidium infection risk, pooled odds ratio 1.98 (95%CI: 1.13–3.49), which was also statistically significant. Case contact is understandably a risk factor for transmitting any infectious disease and this is relevant in both developed and developing nations. However, as the majority of Cryptosporidium infections are associated with mild self-limiting symptoms in healthy adults and could be asymptomatic in children, the number of diagnosed cases are substantially underestimated, contributing to further Cryptosporidium transmission unless proper hand hygiene and prevention measures are implemented. Similarly, overcrowded living conditions are associated with an increased risk of Cryptosporidium infection, pooled odds ratio 1.37 (95%CI: 1.07–1.75). Another person to person transmission pathway is children attending nursery. This risk factor was not assessed in many studies and therefore could not be included in the meta-analysis. This was also the case in developed countries, where only a few studies reported that nursery attendance and changing nappies are risk factors for Cryptosporidium infection [10].
Poor WASH (Water, Sanitation and Hygiene) conditions are paramount to the spread of Cryptosporidium and other gastrointestinal infections. The search strategy was not restricted to focus on WASH. Nevertheless, lack of appropriate sanitation/ open defecation was associated with a significant risk of acquiring cryptosporidiosis, pooled odds ratio 1.82 (95%CI: 1.19–2.8) based on 5 studies. This result is comparable to the systematic review by Speich and colleagues, who reported that lack of sanitation was associated with Cryptosporidium infection risk, pooled odds ratio 1.47 (95%CI: 0.37–5.88) based on 5 studies [27]. Interestingly, the 5 papers included in our systematic review and the one by Speich and colleagues were different, yet the associated risk was comparable. Our search strategy retrieved the papers included in Speich and colleagues, but these were not included as they considered less than 3 risk factors and/ or the data were missing from the full text. Indeed Speich and colleagues reported that they contacted some of the authors to obtain data that was collected but not analysed/ presented in the full text. This was an issue that we encountered while conducting this systematic review as several authors reported the investigation of several risk factors, which were omitted in the results section. Improving sanitation coverage is one of the aims of the Sustainable Development Goals. Though the number of people practising open defecation globally decreased from 38% to 25% between 1990 and 2015, there are currently 946 million people lacking sanitation worldwide (1 in 8) [28]. Open defecation is a clear indicator of extreme poverty and is associated with significant disease burden.
In this systematic review, poor drinking water was not associated with Cryptosporidium infection, however, this was not statistically significant. This was due to the contradicting findings of the included studies and wide confidence intervals. We considered the absence of piped water an indicator of poor drinking water quality. However, this is not necessarily true. The microbiological quality of spring and well water could be satisfactory for the majority of time unless contamination events occur. Indeed, many of the papers retrieved by our search strategy highlighted the increased risk of Cryptosporidium infection in the wet season and/or following extreme rain events. Regular consumption of contaminated drinking water, though not recommended from a public health view, could be associated with building protective immunity [29, 30]. Furthermore, drinking water could be a minor transmission pathway in endemic settings. Indeed, in a quasi-experimental study in India, drinking bottled water was not associated with reduced risk of cryptosporidiosis in children [31].
Breastfeeding (or lack of) was investigated in numerous studies that focused on childhood cryptosporidiosis. Breastfeeding was associated with a protective effect, however, this was not statistically significant. The protection potentially conferred by breastfeeding could be due to the passive immunity acquired through ingestion of Cryptosporidium specific antibodies in breast milk [32]. Additionally, bottle feeding was found to increase the risk of cryptosporidiosis [33], most likely due to one or a combination of the following factors: poor water quality, lack of sterilisation and substandard hand and household hygiene. Indeed, one study found that washing hands before infant feeding was associated with a significant cryptosporidiosis risk, multivariate adjusted odds ratio 5.02 (95%CI: 1.11–22.78) [34], which demonstrates the poor quality of water used for drinking and hand washing.
The main limitation of this systematic review is the small number of studies included. As strict inclusion criteria were applied, a large number of papers that could have added to the body of evidence were excluded. This was because they had a small number of Cryptosporidium cases, explored less than 3 risk factors (excluding age, gender, rural/ urban residence, malnutrition) or reported their results incompletely or inappropriately for inclusion in meta-analysis. This resulted in a small number of studies for each risk factor, which in turn reduced the power of meta-analyses performed. Additionally, this could have inevitably resulted in the exclusion of some other relevant risk factors for Cryptosporidium infection, that are investigated less frequently and/or not in conjunction with well-known transmission pathways. The main limitations of some of the included studies are the small number of cryptosporidiosis cases and poor quality in terms of study design and duration, number of exposures investigated and data reporting. Another shortcoming was that several papers presented risk factors for diarrhoeal diseases in general without categorisation/ sub group analysis for each etiologic agent. Some did not even seek to diagnose diarrhoeal pathogens. This limits the usefulness of such epidemiological studies and hinders the identification of relevant risk factors and the formulation of specific prevention measures. A heterogeneity between the included studies was noted. While, the majority of studies used diarrhoea free, Cryptosporidium negative control groups, some used diarrhoeal subjects that were Cryptosporidium negative. Both sets were undistinguishably included in the meta-analysis, however, combining them would introduce bias in the overall risk associated with each exposure.
In summary, this systematic review identified animal contact, diarrhoea in the household and open defecation as the most relevant risk factors associated with Cryptosporidium infection in LMICs. Improving sanitation coverage is one of the Sustainable Development Goals and progress is likely to happen despite the high number of people still practising open defecation globally. Animal contact and case contact/ household diarrhoea are relevant for both developed and developing countries and prevention measures should include awareness campaigns and better hand hygiene. Other relevant risk factors could have been omitted from the systematic review because of the paucity of data and poor quality of several studies. Considering the significant morbidity and mortality of cryptosporidiosis in sub-Saharan Africa and South Asia, especially for under 5 years (and HIV+), strategies to reduce the prevalence and burden of cryptosporidiosis and other gastrointestinal opportunistic diseases should be prioritised and offered adequate funding.
|
10.1371/journal.pgen.1004530 | Patterns of Admixture and Population Structure in Native Populations of Northwest North America | The initial contact of European populations with indigenous populations of the Americas produced diverse admixture processes across North, Central, and South America. Recent studies have examined the genetic structure of indigenous populations of Latin America and the Caribbean and their admixed descendants, reporting on the genomic impact of the history of admixture with colonizing populations of European and African ancestry. However, relatively little genomic research has been conducted on admixture in indigenous North American populations. In this study, we analyze genomic data at 475,109 single-nucleotide polymorphisms sampled in indigenous peoples of the Pacific Northwest in British Columbia and Southeast Alaska, populations with a well-documented history of contact with European and Asian traders, fishermen, and contract laborers. We find that the indigenous populations of the Pacific Northwest have higher gene diversity than Latin American indigenous populations. Among the Pacific Northwest populations, interior groups provide more evidence for East Asian admixture, whereas coastal groups have higher levels of European admixture. In contrast with many Latin American indigenous populations, the variance of admixture is high in each of the Pacific Northwest indigenous populations, as expected for recent and ongoing admixture processes. The results reveal some similarities but notable differences between admixture patterns in the Pacific Northwest and those in Latin America, contributing to a more detailed understanding of the genomic consequences of European colonization events throughout the Americas.
| We collaborated with six indigenous communities in British Columbia and Southeast Alaska to generate and analyze genome-wide data for over 100 individuals. We then combined this dataset with existing data from populations worldwide, performing an investigation of the genetic structure of indigenous populations of the Pacific Northwest both locally and in relation to continental and worldwide geographic scales. On a regional scale, we identified differences between coastal and interior populations that are likely due to differences both in pre- and post-European contact histories. On a continental scale, we identified differences in genetic structure between populations in the Pacific Northwest and Central and South America, reflecting both differences prior to European contact as well as different post-contact histories of admixture. This study is among the first to analyze genome-wide diversity among indigenous North American populations, and it provides a comparative framework for understanding the effects of European colonization on indigenous communities throughout the Americas.
| The population history of indigenous peoples of the Americas is of perennial interest to scholars studying human migrations. The Americas were the last continents historically peopled by modern humans, with recent evidence supporting an initial human entry via Beringia after the last glacial maximum [1]–[4]. Despite the absence of a deep written record, abundant archaeological sites and rich anthropometric, cultural, and linguistic variation in the Americas have long facilitated thriving programs of investigation of Native American population history and relationships [1], [5]–[8].
Population-genetic approaches applied to dense genome-wide datasets have recently expanded the forms of evidence available for studies of human migration [9]–[14]. In the Americas, genomic studies have been of particular value in understanding the diversity of admixture processes that indigenous communities have experienced with non-native populations following European contact [13], [15]–[21]. Studies have identified considerable variation in the level of admixture among populations, in the level of admixture among individuals within a population, in the contributions from different source populations, and in the magnitudes of the various ancestry contributions at different points in the genome [13], [15], [20]–[23].
Most of this genomic work has focused on populations in Latin America and the Caribbean, evaluating the demographic impact of colonizing individuals of European and African descent on local indigenous groups, and relatively few genome-wide investigations have been performed specifically on indigenous North American populations. Owing, in part, to differences in colonization practices between the British and French in North America and the Spanish and Portuguese in Central and South America, North America experienced a substantially different history of admixture [24]. In the Pacific Northwest, extensive contact with non-native populations began relatively recently, with the Russian expansion and the maritime fur trade in the 1700s. These events allowed indigenous communities to initially receive economic benefits from trade without the disruptive effects of colonization [25]. Multiple immigrant groups then entered the region in the 1800s, as Russian Alaska was transferred to the United States, and as borders between the United States and British-controlled Canada were settled. For example, Scandinavians migrating to Alaska were early contributors to the forestry and fishing industries [26]. In addition, the construction of the Canadian Pacific railway between 1881 and 1885 in British Columbia employed numerous Chinese and Japanese immigrants [27]. These immigrant groups had ample opportunity to intermarry with indigenous members of a variety of local communities.
To obtain a detailed picture of the genetic landscape of the Pacific Northwest of North America, we generated data on over 600,000 genome-wide single-nucleotide polymorphisms (SNPs) in 104 indigenous individuals from four coastal communities in Southeastern Alaska and British Columbia and two communities living in interior British Columbia. We combined these data with existing data from other geographic regions that together encompass 64 worldwide populations. This worldwide dataset allowed us to investigate the genetic structure of indigenous populations of the Pacific Northwest both locally and in relation to continental and worldwide geographic scales, and to further analyze the admixture landscape in the region. The results uncover both differences in the admixture patterns seen among indigenous Pacific Northwest populations as well as notable differences from comparable patterns observed in admixture studies of Latin America, illuminating differences in the histories of admixture experienced by Native American populations from across the American landmass.
We genotyped 104 individuals from six native Pacific Northwest populations and one native population of Mexico at 616,794 SNPs. To assess the new samples in relation to relevant populations from other geographic regions, we integrated our new data with previously published standard data sets on 64 worldwide populations (Figure 1; Tables 1 and S1). After quality control, exclusion of related individuals, and reduction to a set of SNPs overlapping with the earlier datasets, our final dataset included 82 Pacific Northwest individuals, three Mexican Seri individuals and 2,055 individuals from 64 additional populations from Africa, Eurasia, Oceania, and the Americas (including two admixed populations from the United States) at 475,109 SNPs. We performed analyses of heterozygosity, population structure, and admixture, in each analysis focusing on the placement of the populations of the Pacific Northwest in relation to the other populations, both at a worldwide and at a continental scale.
Previous investigations in multiple fields have proposed that populations originally from Asia migrated into the Americas via Beringia after the last glacial maximum and subsequently colonized the continent via north–south migration [1], [4], [13], [17], [45]–[48]. The origin and number of migration waves into the Americas, the pre-contact demography of the populations, and the post-contact recent history of admixture after European contact all represent topics of great interest for understanding the population history of these continents [4], [13], [17]. Despite this interest, however, relatively few genomic investigations of indigenous North American populations have been conducted [13], [48], and most have been centered on Central and South American groups [15], [18]–[21], [23].
To address this imbalance, with an interest in post-contact admixture, we investigated genome-wide SNP diversity in six Pacific Northwest populations, representing coastal and interior regions previously proposed to lie along separate migratory routes from Beringia [4]. The results provide insight into features of migration and admixture in the Pacific Northwest region, as well as differences in population-genetic history from the more frequently studied populations of Central and South America.
Various analyses of population structure placed the Pacific Northwest populations in relatively close genetic proximity, suggesting that these populations share an indigenous component of ancestry more recent than their divergence from other groups. Native American populations were distributed from north to south along a single branch of the neighbor-joining tree, as would be expected under a scenario with a common origin for all of the Native American groups followed by a north–south serial-founder model [16], [28]–[32], [34]. Hierarchical genetic structure among Native American populations detected using Admixture identified clusters specific to Northern, Central, and Southern indigenous American populations within a broader cluster comprising all Native Americans, as might be expected under the model.
In the Pacific Northwest, both our MDS analysis and an Admixture cluster at K = 10 revealed substantial genetic differentiation between coastal and interior populations. It is perhaps plausible that these population groups descend from different groups along separate migratory routes from Beringia into North America [4], [48]. However, because both population groups clustered together consistently in all other Admixture analyses, and they are placed nearby in MDS plots and in the neighbor-joining analysis, our results provide stronger support for a shared origin for the Pacific Northwest populations, and, after the initial peopling of the region, divergence due to isolation and drift. This scenario is consistent with paleoanthropometric studies that also proposed recent isolation, drift, and ecological differences to explain skeletal differences between coastal and interior individuals in British Columbia [49], [50].
Genetic diversities among Pacific Northwest populations were higher than expected under a serial-founder model, as the model predicts intermediate levels of diversity between Northeast Asians and Central and South Americans [31], [32]. Instead, however, heterozygosity levels among Pacific Northwest populations are substantially closer to those of Eurasian populations than to those of Central or South Americans. This result parallels the patterns observed in African Americans and Mexican Americans, two recently European-admixed populations in the Americas, who also showed inflated levels of genetic diversity compared to African and Native American source populations, respectively [9], [29], [34]. It is thus possible that admixture events following European contact might explain high genomic diversity in the Pacific Northwest populations in relation to Central and South Americans [16], [24], [25].
Our MDS and Admixture analyses produced high mean levels of European admixture in Pacific Northwest populations compared with Native American populations from Central and South America. Indeed, we observed high levels of European admixture in the Tlingit, Tsimshian and Haida populations comparable in magnitude to the recently admixed Mexican American population. This result contrasts with patterns in the Amazonian Karitiana and Surui populations, for which no admixture signals were evident, and with the low levels of European admixture observed in Colombians and Central American groups [13], [18]–[21], [23].
Our estimates of the most recent time of admixture support a longer history of European admixture among Central American admixed populations than among Pacific Northwest populations, with the within-population variance of individual admixture estimates across individuals higher in the Pacific Northwest. This result accords with the delayed post-European contact admixture processes in the Pacific Northwest relative to Central and South America [24], the later arrival of Russian and Northern European migrants in the Pacific Northwest fur trade toward the end of the 1700s, and the later colonization period centered on fishing and canning [25], relative to the Spanish and Portuguese colonial periods beginning after 1492.
We detected signals of East Asian admixture in several Pacific Northwest populations, particularly the interior Splatsin and Stswecem'c groups. Consistent with previous studies, we observed no signal of genome-wide East Asian admixture in our set of Central and South American populations [28], [29], [34], [51]. It is possible that the East Asian admixture signal in the Pacific Northwest could represent waves of ancient Asian migrations into the Americas prior to European contact, or an inability of Admixture to fully separate genetic signals from similar groups. However, two features of the pattern support the view that it represents recent East Asian admixture. First, high variance in East Asian admixture proportions across individuals within Pacific Northwest populations indicates a relatively short and recent history of East Asian admixture, a pattern uniquely observed in this region of the Americas. Second, the pattern differs noticeably between the coastal and interior groups, two sets of populations that are otherwise difficult to distinguish. Thus, we surmise that the evidence for East Asian admixture reflects the documented history of Chinese and Japanese immigrants to British Columbia working in the mining, railway and cannery industries in the second half of the 19th century [52], and that these groups had different influences on the coast and in the interior [53].
While our approach using two different methods [43] has provided simple strategies for estimating admixture times, the complexity of the admixture pattern in the Pacific Northwest, likely involving both European and East Asian sources and a different pattern in coastal and interior groups, suggests that simple models may be somewhat limited in applicability to the region. Future theoretical development of admixture models—that, for example, explicitly formulate pre- and post-contact admixture periods—together with approximate Bayesian computation or other techniques that can more fully incorporate admixture patterns into inference of the mechanistic admixture model, will help to enhance understanding of the variable histories of admixture experienced by indigenous American populations, both in the understudied Pacific Northwest and throughout the hemisphere.
We collected DNA samples for 101 individuals from six indigenous populations of British Columbia (Nisga'a n = 8; Splatsin n = 16, Stswecem'c n = 15; Tsimshian n = 32) and Southeastern Alaska (Haida n = 12; Tlingit n = 18), and for three Seri individuals from northwestern Mexico (Figure 1). Collection of the Haida and Tlingit samples was approved by the Institutional Review Board (IRB #10379) at Washington State University, as described by Villanea et al. [54]. Appropriate informed consent and sample collection protocols for the communities from British Columbia and Alaska was approved by Institutional Review Boards at the University of Illinois (IRB #10538). Each participant from British Columbia and Alaska provided familial anthropology information concerning geographic and tribal affiliation of their maternal and paternal lines. For all individuals and populations, knowledge of family histories, including possible recent admixture events, were obtained through classical familial anthropology interviews.
The presence of related individuals in a dataset can influence genetic diversity patterns [40], [55]. We therefore identified pairs of close relatives in the initial dataset using identity-by-state (IBS) allele-sharing and the likelihood approach of Relpair (v2.0.1) [56], [57]. Following Pemberton et al. [40], Relpair was applied to five non-overlapping sets of 9,999 SNPs (the maximum number of markers allowed by Relpair) in which all SNPs were separated by at least 100 kb. In these analyses, we considered only the 210,639 autosomal SNPs that were polymorphic in all seven indigenous populations, using genetic map positions obtained by interpolation on the Rutgers combined physical–linkage map [58], [59]. We set all putative pairwise relationships to “unrelated,” the genotyping error rate to 0.001 (a likely overestimate), and the critical value for likelihood ratio computation to 100. We only considered first- and second-degree relationship inferences, as cousin inferences are less reliable than inferences for closer relationships [40], [56], [57]. To exclude intra-population relative pairs, separately in each population, we applied Relpair using count estimates of allele frequencies in that population. To exclude inter-population relative pairs, we applied Relpair to the whole dataset using count estimates of allele frequencies in the dataset.
We identified 24 intra-population first- and second-degree relative pairs that involved 36 distinct individuals: 11 in the Tsimshian population (four parent–offspring, two full-sibling, and five avuncular, involving 15 individuals in total), six in the Splatsin population (one parent–offspring, two full-sibling, and three half-sibling, involving eight individuals), three in the Stswecem'c population (one parent–offspring, one full-sibling, and one avuncular, involving five individuals), and two each in the Haida (two parent–offspring, involving four individuals) and Tlingit (one parent–offspring and one full-sibling, involving four individuals) populations. No inter-population pairs of close relatives were identified.
To minimize the number of individuals excluded, we first removed from the dataset 11 individuals that appeared in more than one pair. Next, we removed five individuals appearing in only a single pair, selected on the basis of higher levels of missing data. Following the removal of the 16 related individuals from the preliminary dataset containing 565,635 autosomal SNPs, we repeated the population-genetic quality control procedure (Stage 3, Figure S5) and excluded 20,914 SNPs monomorphic in the sample of 85 individuals and 339 SNPs with at least 10% missing data. Thus, our final dataset contained 544,384 autosomal SNPs with genotypes in 85 unrelated individuals from seven populations (Table 1). A version of this dataset restricted to the 82 unrelated individuals from six British Columbian and Alaskan populations newly sampled and genotyped here can be requested from R.S.M. for population and evolutionary history studies in accord with the informed consent documents used for this study.
To investigate the indigenous populations in relation to genetic variation in other populations, we merged the indigenous Northwest dataset with similar publically available data for the 11 populations in release 3 of HapMap project [60] and the 53 populations represented in the HGDP-CEPH cell line panel. First, we separately prepared and merged the HapMap Phase III and HGDP-CEPH datasets using the pipeline of Pemberton et al. [61] and considering only autosomal SNPs; SNPs on the mitochondrion and on the X and Y chromosomes were excluded.
|
10.1371/journal.ppat.1003775 | SPOC1-Mediated Antiviral Host Cell Response Is Antagonized Early in Human Adenovirus Type 5 Infection | Little is known about immediate phases after viral infection and how an incoming viral genome complex counteracts host cell defenses, before the start of viral gene expression. Adenovirus (Ad) serves as an ideal model, since entry and onset of gene expression are rapid and highly efficient, and mechanisms used 24–48 hours post infection to counteract host antiviral and DNA repair factors (e.g. p53, Mre11, Daxx) are well studied. Here, we identify an even earlier host cell target for Ad, the chromatin-associated factor and epigenetic reader, SPOC1, recently found recruited to double strand breaks, and playing a role in DNA damage response. SPOC1 co-localized with viral replication centers in the host cell nucleus, interacted with Ad DNA, and repressed viral gene expression at the transcriptional level. We discovered that this SPOC1-mediated restriction imposed upon Ad growth is relieved by its functional association with the Ad major core protein pVII that enters with the viral genome, followed by E1B-55K/E4orf6-dependent proteasomal degradation of SPOC1. Mimicking removal of SPOC1 in the cell, knock down of this cellular restriction factor using RNAi techniques resulted in significantly increased Ad replication, including enhanced viral gene expression. However, depletion of SPOC1 also reduced the efficiency of E1B-55K transcriptional repression of cellular promoters, with possible implications for viral transformation. Intriguingly, not exclusive to Ad infection, other human pathogenic viruses (HSV-1, HSV-2, HIV-1, and HCV) also depleted SPOC1 in infected cells. Our findings provide a general model for how pathogenic human viruses antagonize intrinsic SPOC1-mediated antiviral responses in their host cells. A better understanding of viral entry and early restrictive functions in host cells should provide new perspectives for developing antiviral agents and therapies. Conversely, for Ad vectors used in gene therapy, counteracting mechanisms eradicating incoming viral DNA would increase Ad vector efficacy and safety for the patient.
| Viruses have acquired functions that target and modulate host cell signaling and diverse regulatory cascades, leading to efficient viral propagation. During the course of productive infection, Ad gene products manipulate destruction pathways to prevent viral clearance or cell death prior to viral genome amplification and release of progeny. Recently, we reported that chromatin formation and cellular SWI/SNF chromatin remodeling processes play a key role in Ad transcriptional regulation. Here, we observe for the first time that SPOC1, identified as a regulator of DNA damage response and chromatin structure, plays an essential role in restricting Ad gene expression and progeny production. This host cell antiviral mechanism is efficiently counteracted by tight association with the major core protein pVII bound to the incoming viral genome. Subsequently, SPOC1 undergoes proteasomal degradation via the Ad E1B-55K/E4orf6-dependent, Cullin-based E3 ubiquitin ligase complex. We also show that other viruses from RNA and DNA families also induce efficient degradation of SPOC1. These analyses of evasion strategies acquired by viruses and other human pathogens should provide important insights into factors manipulating the epigenetic environment to potentially inactivate, or amplify host cell immune responses, since detailed molecular mechanisms and the full repertoire of cellular targets still remain elusive.
| DNA viruses require nuclear transport of their genomes to productively infect the host cell and initiate efficient replication. Simultaneously, introduction of viral nucleic acids into the host cell nucleus triggers danger signals, and activates DDR (DNA damage response) prior to cell cycle arrest, and apoptosis. Many viruses counteract these regulatory measures in infected cells in order to ensure productive infection, which necessitates proper viral gene expression and adequate progeny synthesis [1].
In line with this, Ad (Adenoviruses) have gained functions to modulate DSBR (double-strand break repair), apoptosis, cellular gene expression, and host cell immune responses. The incoming viral genome is complexed with core factors and capsid protein VI after endosomal release. Our recent work showed that Ad DNA remains transcriptionally inactive until protein VI mediates activation of the viral E1A promoter by functionally inhibiting chromatin-associated transcription factor Daxx [2]. Ad E1A (early region 1A protein) is the first protein expressed after infection, playing an essential role in subsequent transcriptional activation, and induction of cell cycle progression [3].
Recently, E1A was seen to interact with the cellular PML-II isoform, and together this complex elevates transcription from Ad promoters [4]. E1B-55K (early region 1B 55 kDa protein) supports efficient viral replication by inhibiting anti-proliferative processes induced by the host cell [5]. However, additional functions of E1B-55K mainly require its interaction with E4orf6 (early region 4 open reading frame protein 6). So far, several reports have demonstrated that Ad E4orf6 connects Ad E1B-55K to an E3 ubiquitin ligase complex in the nucleus, containing cellular factors Rbx1/Roc1/Hrt1, Elongin B/C, and either Cullin 2 or Cullin 5 [6]. Recent work has shown that E1B-55K is the substrate recognition unit, while E4orf6 assembles the cellular components, and this functional complex sequesters cellular target proteins into a proteasomal degradation pathway C [6], [7], [8], [9], [10], [11], [12], [13].
Another hurdle to the viability and propagation of DNA viruses is imposed by host DDR and repair machinery. To counter this, several DNA viruses have acquired early viral genes that degrade or redistribute key cellular factors of the repair machinery to protect viral genome integrity [6]. Ad-mediated DDR inhibition by the Ad E1B-55K/E4orf6 E3 ubiquitin ligase complex and E4orf3 (early region 4 open reading frame protein 3)-dependent relocalization of the MRN complex into nuclear tracks and cytoplasmic inclusions together block concatemer formation and DNA damage signaling, therefore allowing productive infection and efficient virus growth [14], [15], [16], [17], [18].
The human SPOC1 (survival-time associated PHD protein in ovarian cancer 1/PHF13) protein has been identified as a novel regulator of DDR and chromatin structure [19], [20]. The spoc1 gene is located in chromosomal region 1p36.23, a region with frequent heterozygous deletions implicated in tumor development and progression [21], [22]. Consistent with this, elevated SPOC1 RNA levels in primary and recurrent epithelial ovarian cancers have been associated with decreased survival rates in patients [23]. Moreover, SPOC1 RNA can be detected in most human tissues, with the highest levels in the testis, where it has been exclusively detected in spermatogonia [23], [24].
SPOC1 is a nuclear protein with a PHD (plant homeodomain), predicted to bind H3K4me2/3 and to regulate chromatin-specific interactions [20], [25]. In line with this, Kinkley and co-workers observed that SPOC1 is dynamically associated with chromatin and induces chromosome condensation to regulate proper cell division [20]. Particularly, SPOC1 plays a role in radiosensitivity and DNA repair by selective modulation of, and functional cooperation with chromatin modifiers and DDR regulators [19].
There is also evidence that SPOC1 is recruited to DSBs and regulates the kinetics of DSB repair and cellular radiosensitivity. It is proposed that H3K4me2/3-containing chromatin can be converted into more compact chromatin by SPOC1-mediated increase of H3K9 KMTs and H3K9me3. Hence, loss of SPOC1 promotes chromatin decondensation, and is associated with increased levels of DDR transducers and efficient DNA repair. The correlation between SPOC1 protein levels and H3K9me3, as well as expression of several H3K9 KMTs, implicates SPOC1 functions in both chromatin condensation and DDR [19]. In sum, SPOC1 is a multifunctional protein with a additional role in stem cell differentiation, oncogenesis, chromatin structure, and DNA repair processes.
Here, we identified SPOC1 as a novel host restriction factor targeted during viral infection. SPOC1 protein levels decreased in Ad infected cells, which we could attribute to proteasomal degradation mediated by the E1B-55K/E4orf6 E3 ligase complex. Not only did SPOC1 interact with E1B-55K, but also Ad5 E2A-DBP, a marker for nuclear Ad replication sites, as seen by co-immunoprecipitation and immunofluorescence assays. When SPOC1 was overepressed, Ad virus yield, viral DNA synthesis, and viral protein synthesis decreased; reporter gene and chromatin immunoprecipitation assays showed that SPOC1 repressed gene expression at the level of transcription. Intriguingly, interaction with structural viral core protein pVII initially stabilized SPOC1 protein levels before expression of any viral early proteins.
SPOC1 induces chromosome condensation [20], stimulates cellular gene silencing and influences DDR, which potentially contributes to its transformation potential [19]. Since host DDR prevents viability and propagation of DNA viruses, Ad efficiently targets a multitude of host cell DSB repair regulatory factors in order to promote productive infection [9], [10], [17], [26], [27]. Based on these data, we examined SPOC1 protein levels in infected human cells, and observed SPOC1 levels reproducibly reduced after 24 hours post infection (Fig. 1A). This decrease was even more pronounced at a higher multiplicity of infection (data not shown), and was comparable to the reduction of already known targets of the E1B-55K/E4orf6 E3-ubiquitin ligase complex [6].
To discover whether SPOC1 is a new host cell substrate of the Ad5 E3 ubiquitin ligase complex, we determined SPOC1 protein concentrations in wild type (H5pg4100), and mutant-virus infected cells lacking either E1B-55K (H5pm4149) or E4orf6 (H5pm4154) (Fig. 1B). As anticipated, SPOC1 was dramatically reduced in cells infected with the wild type virus H5pg4100 (Fig. 1B, lane 2), whereas the cellular protein accumulated to levels comparable to non-treated cells in infected cells lacking E1B-55K (H5pm4149; Fig. 1B, lane 3). Similarly, SPOC1 was not reduced in cells infected with the E4orf6-minus virus mutant H5pm4154 (Fig. 1B, lane 4). We further examined SPOC1 protein levels in H1299 cells infected with an E4orf6 virus mutant (H5pm4139) carrying point mutations in the BC Box that abrogate the formation of the Ad5 E3 ubiquitin ligase complex, by inhibition of E4orf6 binding to Elongins B and C. Again activity of the E1B-55K/E4orf6 E3 ligase complex apparently played a role in reducing SPOC1 protein concentrations (Fig. 1B, lane 5). Consistent with previous publications, Mre11 was not degraded during lytic infection with the BC Box virus mutant H5pm4139 (Fig. 1B, lane 5). Our results confirm that the formation and ligase activity of the E1B-55K/E4orf6 ubiquitin complex are essential to reduce SPOC1 protein levels.
To confirm that we had identified SPOC1 as a novel target of virus-induced proteasomal degradation via E1B-55K/E4orf6-dependent E3 ligases, we treated infected cells with the proteasome inhibitor MG-132 (Fig. 2A). Functionally inhibiting host cell proteasomes abolished Mre11 and SPOC1 reduction seen in wild type infected cells, supporting the role of ubiquitin proteasome system in SPOC1 degradation.
We finally validated our findings by analyzing the levels of SPOC1 protein in transfected cells (Fig. 2B). Expression of E1B-55K alone (Fig. 2B, lane 2) or E4orf6 alone (Fig. 2B, lane 3) had no effect on steady-state concentrations of endogenous SPOC1 protein, while expression of both E1B-55K and E4orf6 (Fig. 2B, lane 4) diminished SPOC1 steady-state concentrations to levels similar to virus-infected cells (Fig. 1 and 2A).
Since E1B-55K is the substrate recognition unit of the SCF-like E3 ubiquitin ligase, we next tested whether E1B-55K interacts with the endogenous SPOC1 protein. As anticipated, in E1B-55K-transfected human H1299 cells E1B-55K co-immunoprecipitated with SPOC1-specific antibody, revealing an interaction between these factors (Fig. 3A, lane 2). No E1B-55K signal was observed in the corresponding negative controls (Fig. 3A, lane 1).
The next question was whether E1B-55K interferes with the intracellular localization of SPOC1. Consistent with previous findings, immunofluorescence analysis in E1B-transfected human cells revealed that wild type E1B-55K protein localizes in the cytoplasm, mostly concentrated in perinuclear bodies Fig. 3B, panel F, H; [28], [29], [30], [31], [32], [33], [34]. In contrast to the diffuse nuclear localization in non-treated cells (Fig. 3B, panel C, D), SPOC1 was completely sequestered into the E1B-containing aggregates in the presence of the viral protein (Fig. 3B, panel G, H).
To gain better understanding of SPOC1 role during Ad infection, we investigated the subcellular localization of SPOC1 in infected human DLD1 cells, which stably overexpress SPOC1 after doxycyclin treatment to overcome degradation of the cellular factor (Fig. S1). In mock-infected cells, SPOC1 is diffusely distributed inside the nucleus in approximately 90% of the cells investigated (n = 50; data not shown). When monitored together with Ad5 E2A-DBP, a marker for Ad replication sites in the nucleus, at 24 to 48 hours post infection SPOC1 colocalized with sites of viral replication in approximately 60% of cells (n = 50; Fig. 4A, panels D, H, L and P).
We confirmed our observation by monitoring SPOC1/E2A-DBP association using immunoprecipitation analysis after wild type and mutant virus infection in DLD1 and U2OS cells stably overexpressing SPOC1 after doxycyclin treatment (Fig. 4B). As expected, E2A-DBP co-immunoprecipitated with SPOC1-specific antibody, revealing an interaction in all infected SPOC1-overexpressing human cells (Fig. 4B, lanes 2, 3, 4, 6, 7, 8), whereas no signal was obtained in the corresponding negative controls (Fig. 4B, lanes 1 and 5). We also were able to confirm the SPOC1/E1B-55K interaction within the infected SPOC1-induced cells. To further test specificity of the SPOC1 antibody used in all of the coIPs, we tested SPOC1 binding after proteasomal degradation of SPOC1 (Fig. S2). Therefore, we infected H1299 cells with Ad5 wild type virus (H5pg4100) and virus mutants depleted for either E1B-55K (H5pm4149) or E4orf6 (H5pm4154) expression. After 24 and 48 hors post infection, total cell extracts were prepared. E1B-55K and E2A/DBP were immunoprecipitated using rabbit polyclonal SPOC1 antibody. Proteins were separated on 10% SDS-PAGE and visualized by immunoblotting. Input levels of total-cell lysates and co-precipitated proteins were detected using monoclonal antibody 2A6 (E1B-55K), B6-8 (E2A/DBP), SPOC-1-specific rat monoclonal antibody, and mouse monoclonal antibody AC-15 (β-actin) as a loading control. As expected, E2A-DBP co-immunoprecipitated with SPOC1-specific antibody (Fig. S2, lanes 3, 5–8), only revealing binding in infected cells, where SPOC1 could not be degraded due to the time point (Fig. S2, lane 3) or the absence of either E1B-55K (Fig. S2, lanes 5 and 6) or E4orf6 (Fig. S2, lanes 7 and 8). We also were able to confirm the specific SPOC1/E1B-55K interaction within the infected cells expressing SPOC1 and E1B-55K (Fig. S2, lanes 3, 7 and 8). Together these data improved our co-immunoprecpitation analysis and reassured specificity of the SPOC1 antibody usd in these experiments.
Together, our results show that the cellular factor SPOC1 associates with E2A-DBP within Ad5 replication compartments in virus-infected human cells. This novel observation prompted us to investigate whether SPOC1 is linked to Ad transcriptional regulation during productive infection. Such recruitment to nuclear sites associated with viral genome replication suggests that SPOC1 is involved in regulating Ad5 gene expression.
To better analyze the role of SPOC1 during Ad5 infection, we performed experiments in SPOC1-inducible human colon carcinoma (DLD1) and osteosarcoma (U2OS) cell lines, which express exogenous SPOC1 after doxycyclin (dox) treatment. Prior to our analysis, we investigated SPOC1 protein expression in the absence and presence of dox (Fig. 5A). Additionally, proliferation of the SPOC1 overexpressing cells was quantified, revealing no significant difference in comparison to the uninduced controls (Fig. S1).
To assess the effect of SPOC1 on overall virus growth, we determined total virus yield after SPOC1 induction in DLD1 cells (Fig. 5B). SPOC1 overexpression reduced progeny production seven- (48 h p.i.) to eight-fold (72 h p.i.) compared to non-treated DLD1 control cells (Fig. 5B). When we analyzed Ad5 progeny production in SPOC1 inducible U2OS cells we observed similar effects (Fig. 5B).
These results suggest that SPOC1 mediates repressive effects during Ad5 infectious cycle. To further validate this hypothesis, expression of viral early and late proteins was monitored at different time points after infection (Fig. 5C). Consistent with Ad5 progeny production, protein synthesis was inefficient in SPOC1 overexpressing DLD1 and U2OS cells (Fig. 5C). Particularly expression of E1A was affected, since it was substantially higher in non-treated DLD1 (Fig. 5C, upper panel lanes 1–6) and U2OS (Fig. 5C, lower panel lanes 1–6). Similar effects were observed when monitoring viral DNA synthesis in infected cells. As with Ad5 progeny production (Fig. 5D), DNA synthesis was less efficient in SPOC1 induced cells than in uninduced cells (Fig. 5D). Next we investigated whether Ad transcription is negatively regulated by SPOC1 expression. Therefore, we analyzed whether early and late viral mRNA levels are affected by SPOC1 overexpression in Ad wild type virus infected cells. We observed that viral early E1A and E1B mRNA production is lower in SPOC1-induced cells, compared to cells lacking treatment with doxycyclin (Fig. 5E). Similar results were obtained for hexon mRNA expression, suggesting either an impact of enhanced synthesis of early gene products or direct repression of the late promoter (Fig. 5E). Moreover, we tested cellular mRNA levels to exclude non-specific repression of gene expression by SPOC1 expression affecting cellular RNA Pol II transcription (Fig. 5E). Our results show similar GAPDH mRNA synthesis in the cells tested, indicating that SPOC1 expression does not reflect an overall reduction in cellular RNA pol II transcription.
To validate these observations and to exclude non-specific impact of the overexpression studies, we depleted SPOC1 expression by transient knock-down with siRNA (Fig. 6A) and analyzed Ad5 replication. Virus growth was enhanced three-fold (24 h p.i.) to five-fold (48 h p.i.) in the absence of SPOC1, compared to control cells (Fig. 6B). Consistent with data observed in DLD1 and U2OS cells, early and late viral protein synthesis was more efficient in SPOC1-depleted cells (Fig. 6C). DNA synthesis was monitored and showed two-fold more efficient synthesis of viral DNA in SPOC1-depleted cells compared to the control cells (Fig. 6D). To further investigate a role of SPOC1 in Ad transcription and viral mRNA synthesis, we analyzed whether early and late viral mRNA levels are affected by the absence of SPOC1 expression in infected cells (Fig. 6E). We observed that viral early E1A and E1B mRNA production is two-fold reduced in SPOC1 expressing cells, compared to cells lacking this cellular transcription factor. Results obtained for hexon mRNA expression showed three-fold difference in viral mRNA expression (Fig. 6E). As shown above, we did not observe altered GAPDH mRNA levels, indicating that SPOC1 depletion does not affect cellular transcription in general (Fig. 6E).
To exclude that SPOC1 expression impacts on virus entry, we analyzed whether capsid protein VI from incoming Ad capsids show altered accumulation in the host cell nuclear fraction. Using nucleo-cytoplasmic fractionation, we observed rapid protein VI accumulation in the nuclear fraction after infection of U2OS control cells and cells treated with doxycyclin to induce SPOC1 expression (Fig. S3).
Together, these results indicate that cellular factor SPOC1 is a potent repressor of Ad5 growth and thus represents a novel host cell restriction factor during Ad productive infection.
Due to the fact that SPOC1 is a chromatin bound protein, we next questioned whether SPOC1-dependent Ad restriction occurs on the transcriptional level. To assess this, we performed reporter gene assays with luciferase expression vectors under the control of specific Ad promoters in SPOC1-induced and non-induced DLD1 and U2OS cells. In the absence of other viral factors, SPOC1 was able to repress luciferase expression from all the viral promoters tested (Fig. 7A).
To exclude non-specific repression of gene expression, we investigated whether we can detect SPOC1-dependent effects on the cellular E2F-dependent H2A promoter (Fig. S4). In contrast to the viral reporter constructs, co-expression of SPOC1 did not affect transcription from the cellular promoter (Fig. S4, lanes 3 and 4). The inhibitory effect of SPOC1 expression is specific at least for the promoter sequences tested so far.
To further demonstrate direct association of SPOC1 with Ad promoter sequences, we performed chromatin immunoprecipitation assays from wild type virus infected cells, using SPOC1-specific monoclonal antibody, unrelated IgG control antibody and Ad promoter-specific oligonucleotides (Fig. 7B). The results show that SPOC1 is associated with E1A, E1B, E2early and MLP Ad promoters in infected cells (Fig. 7B). Conclusively, these results indicate that SPOC1 is a component of host cell antiviral mechanisms, playing an important role in the Ad gene expression program through transcriptional repression of viral promoters.
SPOC1 has been previously reported to bind histone H3 and to be capable of promoting repressive epigenetic transitions. Since Ad major core protein VII shares homology with the N-terminal regulatory tail of histone H3 [35], we investigated if SPOC1 may also associate pVII. Whole extracts from H1299 cells that had been transfected with plasmid DNA expressing pVII products (Fig. 8A, lanes 1–3) as well as Ad wild type infected lysates (Fig. 8A, lanes 4 and 5) were immunoprecipitated with SPOC1 specific ab and analyzed by Western blotting using anti-HA mab (Fig. 8A). We detected co-immunoprecipitation of SPOC1 with pVII, when both factors were present (Fig. 8A, lanes 2, 3 and 5).
As core protein pVII remains associated with the viral genome during entry [36], we performed further pVII transfection studies mimicking immediate early phase of infection (Fig. 8B). First, we observed a significant increase in SPOC1 protein levels (Fig. 8B, lane 2); and repressive H3K9me3 histone marks as expected due to other reports [19]; Fig. 8B, lane 2. Next, we carried out additional infection studies with H5pg4100 Ad wild type virus infected cells (Fig. 8C). Again, we monitored SPOC1 and H3K9me3 protein levels and detected loss of H3K9me3, most presumably due to SPOC1 reduction by proteasomal degradation via E1B-55K/E4orf6 E3 ubiquitin ligases (Fig. 8C, lane 2). As pVII was reported to occupy Ad DNA during the immediate early phase of infection [36], [37], [38], [39], our results indicate that during early Ad infection, pVII likely protects its genome from SPOC1 mediated repressive silencing, prior to the onset of transcription by pVII removal from the genome [40], [41] including loss of pVII-bound repressive factor SPOC1.
To determine whether SPOC1 is a key player in antiviral defense in general, and whether strategies to restrain SPOC1 are conserved among different human pathogenic viruses, we tested protein expression in permissive cell lines infected with HSV-1 (herpes simplex virus type 1; Fig. 9A), HSV-2 (herpes simplex virus type 2; Fig. 9A), HIV-1 (human immunodeficiency virus type 1; Fig. 9B) or HCV (human hepatitis C virus; Fig. 9C). To evaluate efficient infection, we monitored viral protein levels of HSV-1 nucleocapsid protein crossreacting with HSV-2 nuclear protein (Fig. 9A), HIV-1 p24 (Fig. 9B) or HCV NS5A (Fig. 9C). Intriguingly, we detected significantly reduced SPOC1 protein levels in all virus-infected cells (Fig. 9). Taken together, functional inhibition of the antiviral host factor SPOC1 by Ad early proteins E1B-55K and E4orf6 can also be achieved by other viral factors expressed in the course of HSV-1, HSV-2, HIV-1 and HCV infection.
Viruses exploit cellular pathways for their own benefit, often achieved by providing high affinity binding sites on viral factors that recruit key regulatory proteins from cellular pathways to outcompete their physiological binding partner. This strategy allows viruses to infect cells and establish efficient replication with nothing more than the incoming viral genome and viral capsids. Besides transport of the genome to appropriate replication sites, viruses also need to assure decondensation and transcriptional activation of the viral genome, which in most cases has been packed and stored in the most economical way. Once the viral genome becomes transcriptionally activated new viral proteins are synthesized providing the virus with the capacity to reprogram the cell for viral replication.
All of the very early steps during viral infection represent an essential moment in establishing productive infection of all human pathogenic viruses, but are as yet poorly characterized. In this report, we show that the cellular protein SPOC1 is tightly regulated in the course of productive Ad infection, identifying SPOC1 as a antiviral restriction factor in cellular defense. Ad possesses strategies to neutralize SPOC1 via an E1B-55K/E4orf6-dependent proteasomal degradation pathway. Moreover, functional inhibition of SPOC1 seems to be conserved among different human pathogenic viruses.
Despite the well-characterized functions of Ad early genes and core/capsid proteins, it is still unclear how Ad transcription is initiated in detail. To summarise our findings we have put together a scheme of how various factors may interact at early stages after Ad infection (Fig. 10). Initially, the genome enters the cell as a highly condensed, transcriptionally inactive nucleoprotein complex, assembled with capsid and core proteins pVI and pVII, identified to recruit cellular factors to the Ad genome Fig. 10; [2], [42]. Recently, we reported transactivating properties of protein VI involving a conserved PPxY motif required for binding to ubiquitin ligases of the Nedd4 family of E3 ubiquitin ligases, prior to Ad-dependent depletion of Daxx/ATRX dependent transcriptional restriction Fig. 10; [2].
The Ad major core protein VII remains bound to the Ad genome during the early phase of infection and is subsequently released due to transcription Fig. 10; [37]; however the duration and amount of pVII complexed with the viral genome is still unclear. Moreover, it also remains elusive whether complete disassociation of pVII from viral DNA is required for active transcription. Nevertheless, pVII is the most abundant structural component of the viral core, is strongly associated with viral DNA in a sequence-independent manner [43], and shares homology with the N-terminal regulatory tail of histone H3 [35]. When this viral factor is imported into the nucleus together with the viral genome, it apparently packages the incoming viral DNA into chromatin-like structures Fig. 10; [37], [44], [45], [46], [47].
SPOC1 is a nuclear PHD-protein, predicted to bind H3K4me2/3 and to regulate chromatin-specific interactions [20], [25]. Therefore, SPOC1 is dynamically associated with chromatin, and plays a major role in chromosome condensation to regulate proper cell division [20]. It is proposed that H3K4me2/3-containing chromatin is converted into more compact chromatin by SPOC1-mediated increase of H3K9 KMTs (lysine methyltransferases) and H3K9me3 [19].
We observed an association between SPOC1 and viral promoter regions; and with Ad core protein pVII in the absence of any other viral protein, and during Ad infection (Fig. 7B; Fig. 8; Fig. 10). We also observe SPOC1-mediated pVII stabilization, 48 hours post transfection (Fig. 8) and hypothesize that pVII cooperation with viral DNA protects the incoming viral genome from immediate early checkpoint signaling. In accordance with recent observations by Karen and co-workers, we believe that pVII prevents the onset of DNA repair signals prior to efficient viral gene expression [36].
These ideas are further supported by recent evidence that SPOC1 expression levels have a strong impact on DDR, DNA repair, and cellular radiosensitivity [19]. It was observed that SPOC1 and associated factors are recruited to DSBs to modulate chromatin structure as well as DDR. Due to SPOC1 binding and stabilization of repressive factors, i.e. H3K9 KMTs and H3K9me3, this antiviral protein helps conversion into a compact chromatin status (Fig. 10). Conversely, loss of SPOC1 releases several factors and repressive histone marks, promoting chromatin decondensation.
In the early phase of Ad productive infection SPOC1 is recruited to the viral genome and Ad5 replication centers (Fig. 4; Fig. 7B). The SPOC1-rich chromatin environment promotes the activation and recruitment kinetics of DDR transducers (γH2AX, pATM) [19]. However, the Ad core protein pVII masks the genome and antagonizes recognition by host cell repair factors (Fig. 10). Subsequently, core protein VII and pVII bound SPOC1 is released from viral DNA prior to synthesis of Ad replication proteins E2A-DBP, resulting in recognition of viral origins of DNA replication. The replication proteins presumably compete with components of DDR for binding to Ad sequences to form a functional preinitiation complex prior to DDR being inhibited by E1B-55K, E4orf6 and E4orf3.
Interestingly, the correlation between SPOC1 protein levels and H3K9me3 implies that Ad may regulate expression of several H3K9 KMTs. H3K9me3 is not only an epigenetic mark characteristic of heterochromatin, but is also the direct binding platform of cellular and/or viral factors that regulate heterochromatin compaction and spreading [48], [49]. In line with this, Gupta and coworkers recently showed that Ad-dependent E1B-55K/E4orf6 E3 ubiquitin ligase induces proteasomal degradation of the acetyltransferase TIP60 [10]. TIP60 chromodomain binding to H3K9me3 at DSBs is required for its activation [50], and is a prerequisite for efficient ATM activation by acetylation. E1B-55K/E4orf6 might interfere with TIP60 binding to H3K9me3 and its activation, possibly contributing to the observed reduction in activated ATM and delayed DDR.
Furthermore, Ad-mediated SPOC1 (Fig. 1 and 2) and TIP60 depletion would promote chromatin relaxation and decreases accessibility of H3K9me3 for TIP60, leading to reduction in ATM activation. So far, several histone H3K9 KMTs were shown to be modulated by SPOC1 (SETDB1, G9A, GLP) [19]. After the onset of viral transcription and efficient E1B-55K/E4orf6 expression, we observe efficient degradation of SPOC1 host factor (Fig. 1 and 2), promoting efficient reduction of H3K9me3 repressive histone marks (Fig. 8). Our findings provide hints that this complex might be altered in stability and activity due to Ad-dependent proteasomal degradation of SPOC1 (Fig. 1, 2 and 8), since these KMTs form a multi-subunit complex that is destabilized by depleting the individual components [51].
In this report, we show that SPOC1 is tightly regulated in the course of productive Ad infection to neutralize host cell defense processes. SPOC1 represents a novel antiviral restriction factor, being associated with Ad major core protein VII and therefore being removed from the viral genome during the very early phase of infection prior to the onset of gene expression during productive virus life cycle (Fig. 10). In other words SPOC1 apparently plays a biphasic role during Ad life cycle by regulating determinants of chromatin compaction and damage response immediately after viral genome entry into the host nucleus, prior to activation of Ad transcription leading to its proteasomal degradation by newly expressed viral proteins.
Besides Ad, different human pathogenic DNA and RNA viruses can impose reduction on SPOC1 protein levels. Our findings provide a model of how viruses might antagonize intrinsic SPOC1-mediated antiviral responses of their host cells, and create novel awareness for general antiviral restriction factors. These future insights into the immune evasion strategies acquired by viruses and other human pathogens mediated within the host will contribute to identifying new therapeutic strategies and targets to limit or prevent human pathogenic virus-mediated diseases and mortality of patients.
HEK293 [52], H1299 [53], SPOC1-inducible U2OS and DLD1 cells [19] were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum (FCS), 100 U of penicillin and 100 µg of streptomycin per ml in a 5% CO2 atmosphere at 37°C. For HepaRG cells, the media was supplemented with 5 µg/ml of bovine insulin and 0.5 µM of hydrocortisone.
Ad proteins were expressed from their respective complementary DNAs under the control of the CMV immediate-early promoter, derived from the pcDNA3 vector (Invitrogen) to express Ad wild type E1B-55K and E4orf6 [54], [55]. cDNAs encoding E1B-55K-products from each Ad type were cloned as described previously [56], [57], [58]. Transient transfections with luciferase reporter constructs were performed as described previously [2]. pRE-Luc and the RGC-firefly luciferase reporter plasmids pGalTK-Luc have been described previously [59]. SPOC1 wild type protein was expressed from pcDNA4TO-SPOC1 constructs.
H5pg4100 served as the wild type Ad5 parental virus in these studies [60]. The mutant viruses H5pm4149 and H5pm4154 were generated as described recently [61]. Both viruses carry stop codons and do not express the respective viral protein [62]. E4orf6 BC-box mutant virus H5pm4139 was generated, resulting in a drastically reduced ability to associate with the Ad5 E3 ubiquitin ligase complex compared to the E4orf6 protein from wild type virus [61]. Viruses were propagated, titrated and infected as described previously [63]. Virus yield was determined by quantitative E2A-72K immunofluorescence staining and viral DNA replication was monitored by quantitative PCR exactly as described previously [13].
ChIP analysis was performed as described previously [2]. The average Ct-value was determined from triplicate reactions and normalized with standard curves for each primer pair. The identities of the products obtained were confirmed by melting curve analysis.
For dual luciferase assays, subconfluent cells were transfected prior to preparation of total cell extracts (48 h) [13]. RGC firefly luciferase and pGL-H2A-promoter activity was assayed with lysed extract in an automated luminometer (Berthold Technologies). All samples were normalized for transfection efficiency by measuring Renilla-luciferase activity. All experiments shown were performed in triplicate and data are presented as mean values.
Subconfluent cells were infected with wild-type virus and harvested at 24 h p.i.. Total RNA was isolated with Trizol reagent (Invitrogen) as described by the manufacturer. The amount of total RNA was measured and one microgram of RNA was reverse transcribed using the Transcriptor High Fidelity cDNA Synthesis Sample Kit from Roche including anchored-oligo(dT)18 primer specific to the poly(A)+RNA. Quantitative real-time PCR was performed with a first strand method in a Rotor-Gene 6000 (Corbett Life Sciences, Sydney, Australia) in 0.5 ml reaction tubes containing a 1/100 dilution of the cDNA template, 10 pmol/µl of each synthetic oligonucleotide primer, 12.5 µl/sample Power SYBR Green PCR Master Mix (Applied Biosystems). The PCR conditions were as follows: 10 min at 95°C, 55 cycles of 30 s at 95°C, 30 s at 55 to 62°C (depending upon the primer set) and 30 s at 72°C. The average Ct value was determined from triplicate reactions and levels of viral mRNA relative to cellular 18S rRNA were calculated as described recently [13]. The identities of the products obtained were confirmed by melting curve analysis.
For protein analysis cells were resuspended in RIPA buffer as described previously [64]. After 1 h on ice, the lysates were sonicated and the insoluble debris was pelleted at 15,000×g/4°C. For immunoprecipitation and immunoblotting protein lysates were treated as described recently [2]. Primary Ab specific for Ad proteins used in this study included E1B-55K mab 2A6 [65], E2A-72K mouse mab B6-8 [66], E4orf6 mab RSA3 [67], rabbit polyclonal serum against protein VI [68] and anti-pVII rabbit polyclonal antibody (generously provided by Dan Engel, University of Virginia). To evaluate efficient infection with different RNA and DNA viruses primary antibodies specific for HSV-1 nucleocapsid protein (monoclonal mouse mab H1.4; Acris antibodies) crossreacting with HSV-2 nuclear protein, HIV-1 p24 hybridoma 183-H12-5C [69] and HCV NS5A (monoclonal mab 2F6/G11 from immunological and biochemical test systems) were used.
Primary antibodies specific for cellular proteins included SPOC1 rabbit polyclonal CR56 and rat mab [20], rabbit polyclonal ab specific for histone variant H3K9me3 (Upstate), Mre11 rabbit polyclonal antibody pNB 100–142 (Novus Biologicals, Inc.), p53 rabbit ab FL393 (Santa Cruz Biotechnology, Inc. [70]), polyclonal rabbit antibody raised against SAF-A protein [71] and ß-actin mouse mab AC-15 (Sigma-Aldrich, Inc.). HA-epitopes were detected with rat monoclonal 3F10 (Roche). Secondary Ab conjugated to horseradish peroxidase (HRP) to detect proteins by immunoblotting were anti-rabbit IgG, anti-rat IgG and anti-mouse IgG (Jackson/Dianova).
Cells were prepared and analyzed as described recently [57]. Images were cropped using Adobe Photoshop CS4 and assembled with Adobe Illustrator CS4.
|
10.1371/journal.pgen.1002427 | Autosomal Recessive Dilated Cardiomyopathy due to DOLK Mutations Results from Abnormal Dystroglycan O-Mannosylation | Genetic causes for autosomal recessive forms of dilated cardiomyopathy (DCM) are only rarely identified, although they are thought to contribute considerably to sudden cardiac death and heart failure, especially in young children. Here, we describe 11 young patients (5–13 years) with a predominant presentation of dilated cardiomyopathy (DCM). Metabolic investigations showed deficient protein N-glycosylation, leading to a diagnosis of Congenital Disorders of Glycosylation (CDG). Homozygosity mapping in the consanguineous families showed a locus with two known genes in the N-glycosylation pathway. In all individuals, pathogenic mutations were identified in DOLK, encoding the dolichol kinase responsible for formation of dolichol-phosphate. Enzyme analysis in patients' fibroblasts confirmed a dolichol kinase deficiency in all families. In comparison with the generally multisystem presentation in CDG, the nonsyndromic DCM in several individuals was remarkable. Investigation of other dolichol-phosphate dependent glycosylation pathways in biopsied heart tissue indicated reduced O-mannosylation of alpha-dystroglycan with concomitant functional loss of its laminin-binding capacity, which has been linked to DCM. We thus identified a combined deficiency of protein N-glycosylation and alpha-dystroglycan O-mannosylation in patients with nonsyndromic DCM due to autosomal recessive DOLK mutations.
| Idiopathic dilated cardiomyopathy (DCM) is estimated to be of genetic origin in 20%–48% of the patients. Almost all currently known genetic defects show dominant inheritance, although especially in younger children recessive causes have been proposed to contribute considerably to DCM. Knowledge of the genetic causes and pathophysiological mechanisms is essential for prognosis and treatment. Here, we studied several individual young patients (5–13 years old) with idiopathic and sometimes asymptomatic dilated cardiomyopathy. The key to identification of the gene was the finding of abnormal protein N-glycosylation. Via homozygosity mapping and functional knowledge of the N-glycosylation pathway, the causative gene could be identified as dolichol kinase (DOLK). Since DCM is very rare in N-glycosylation disorders (Congenital Disorders of Glycosylation, CDG) and most patients with CDG present with a multisystem involvement, we studied the underlying pathophysiological cause of this life-threatening disease. Biochemical experiments in affected heart tissue showed deficient O-mannosylation of alpha-dystroglycan, which could be correlated with the dilated cardiomyopathy. Our results thus highlight nonsyndromic DCM as a novel presentation of DOLK-CDG, via deficient O-mannosylation of alpha-dystroglycan.
| Dilated cardiomyopathy (DCM) is a life-threatening disease characterized by left ventricular enlargement and systolic dysfunction, which can lead to congestive heart failure and is a common cause of patients requiring heart transplantation. In view of the progressive disease course and the acuteness of presenting symptoms, early recognition and diagnosis of the underlying etiology is essential. Genetic causes for DCM are estimated to explain 20–48% of all idiopathic patients [1]–[3]. Until now, 33 nonsyndromic DCM genes have been identified, two on the X chromosome and 31 on the autosomes, of which only one shows recessive inheritance [4]. We expect more recessive genes, since recessive forms have been shown to explain up to 16% of familial DCM [5]. Especially in young children (<10 years old) these are expected to contribute considerably to disease [6].
Protein N-glycosylation is a very common co-translational modification of many proteins, following a sequential and highly ordered pathway in the cytoplasm, endoplasmic reticulum (ER) and Golgi apparatus. Genetic defects in this pathway generally lead to a multisystem disease. These inborn errors of metabolism form the group of Congenital Disorders of Glycosylation (CDG) for which currently more than 40 different genetic defects are known [7]. Defects in the ER during the assembly of the lipid-linked oligosaccharide [8], and glycan transfer to nascent protein chains affect all N-linked proteins and typically lead to a multisystem presentation in CDG-I patients. Clinically such patients are characterized by psychomotor and intellectual disability, muscle hypotonia, seizures, ophthalmologic anomalies, failure to thrive, endocrine and coagulation abnormalities and variable dysmorphic features. Dilated cardiomyopathy in CDG is very rare and has only been described in one of the two reported families with dolichol kinase deficiency (DOLK-CDG, MIM 610768) as part of a multisystem presentation with profound muscular hypotonia, ichthyosiform skin, nystagmus, epilepsy and pulmonary infections, leading to death within the first months of life [9], and in patients with liver involvement [10] or cognitive delay [11]. On the other hand, cardiomyopathy of the hypertrophic type is common in CDG type I [12]–[14]; it is one of the lethal comorbidity factors in CDG-Ia (PMM2-CDG, MIM 212065) patients in infancy.
In this paper, we present eleven young patients (age 5–13 years) with CDG and recessive mutations in DOLK with a predominantly nonsyndromic presentation of DCM. In addition, we show that the main presenting symptom of DOLK-CDG is caused by deficient O-mannosylation of sarcolemmal alpha-dystroglycan.
Dilated cardiomyopathy was diagnosed in several children (see pedigree; Figure 1), without significant muscular weakness or creatine kinase (CK) elevation. Central nervous system involvement, such as cerebellar ataxia, epilepsy or intellectual disability was not present in the patients, except for transient muscular hypotonia and mild developmental delay with a minor increase of CK in family IV. Decreased coagulation parameters were observed in all individuals.
Patient I/2, the second male child of healthy, consanguineous parents of Druze origin was referred to the pediatric metabolic unit for evaluation of mild failure to thrive and persistent elevated transaminases during infancy. Impaired glycosylation (CDG-I) was diagnosed at the age of 10 months [15]. At 6 years of age a mild asymptomatic dilatation of the left ventricle was shown on echocardiogram. He developed acute heart failure at age 11.
Patient I/3, the younger brother of patient I/2, is clinically asymptomatic. At age 4 years, following the diagnosis of his brother, mildly elevated transaminases were noticed. He underwent echocardiography, which revealed mild dilated cardiomyopathy.
Patient II/2, the second male child of healthy, consanguineous parents of Druze origin, was clinically healthy until the age of 9 years, when he was admitted with acute congestive heart failure and dilated cardiomyopathy of unknown etiology. He died suddenly following heart arrhythmia.
Patient II/5, sister of patient II/2, was diagnosed with a dilated cardiomyopathy at age 7. Biopsied ventricles of the explanted heart revealed myocyte hypertrophy and interstitial fibrosis (Figure 2), more pronounced in the left ventricle than in the right ventricle.
Patient II/6, a younger sister of patient II/5, had a history of mild hypotonia, failure to thrive, short stature and ichthyosiform dermatitis. She was diagnosed with mild dilated cardiomyopathy at the age of 6 years.
Patient III/2 was the second male child of healthy, consanguineous parents of Beduin origin. At age 9 years, he presented with progressive weakness over the last month. These symptoms led to the diagnosis of “viral myocarditis” resulting in an end-stage dilated cardiomyopathy and death after unsuccessful reanimation.
Patient III/1, a 13 years old sister of III/2, was found to have asymptomatic dilated cardiomyopathy following the diagnosis of her brother. Her younger brothers, patient III/3 of 11 years old, and III/4 of 9 years old showed asymptomatic minimal evidence of cardiomyopathy detected by repeated echocardiograms. On supportive treatment no further deterioration of the cardiac function was observed during the 3 years of follow up. Ichthyosiform dermatitis was noticed in siblings 2, 3 and 4.
An 11-year-old female of Indian origin with a background of learning difficulties, mild hypotonia and ichthyosis, presented with cardiac failure secondary to severe dilated cardiomyopathy. Prior to the diagnosis of CDG, her condition deteriorated; she required mechanical support and was listed for cardiac transplant. She died of thrombotic and septic complication whilst having Berlin heart as bridging procedure for transplant. Her younger brother was diagnosed with the same defect. He has mild developmental delay but cardiac function is normal.
Transferrin isoelectric focusing for analysis of N-glycosylation abnormalities was performed during metabolic screening. Convincingly abnormal profiles were found for all affected patients, showing an increase of asialo- and disialotransferrin and low or decreased tetrasialotransferrin (Figure 3). These results indicated a diagnosis of CDG type I with a genetic defect in the cytoplasm or endoplasmic reticulum. The most common subtype PMM2-CDG (CDG-Ia) was excluded by analysis of phosphomannomutase activity in patient fibroblasts. In view of the specific clinical symptoms and consanguinity in families I and II, we chose a direct homozygosity mapping approach instead of lipid-linked oligosaccharide analysis.
Homozygosity mapping was performed in two Israeli families (I and II, Figure 1) using the Affymetrix GeneChip Mapping 10 K 2.0 array (family I) and the Affymetrix GeneChip Human Mapping 250 k NspI Array (family II). The largest overlapping homozygous region in families I and II was found on chromosome 9. In family I, the 19.1 Mb region at 9q33.1–9q34.3 was delimited by SNP_A-1518745 and 9pter. By using the two siblings of family II, the overlapping region could be confined to 5.0 Mb at 9q33.3 and 9q34.11, delimited by SNP_A-4223282 and SNP_A-2111464. Short tandem repeat (STR) marker analysis confirmed homozygosity of this region and showed that the haplotypes of the two Israeli families were identical (Figure 1, I and II). The three affected siblings of family III with a similar phenotype were homozygous for the same region, delimited by D9S1872 and 9qter. The overlapping homozygous region of the three families contained 117 genes. Comparison of this region with a list of candidate genes for CDG-I glycosylation defects highlighted two candidate genes known to be involved in protein N-glycosylation, DOLPP1 and DOLK.
Analysis of the protein coding sequence of DOLK in family I showed a homozygous missense mutation (c.1222C>G; p.His408Asp; Figure 1A). The same mutation was identified in family II. The finding in two seemingly unrelated kindreds, who reside in two different villages in Northern Israel, and the presence of an identical 5 Mb interval haplotype including the same mutation, suggest a founder event among these Druze kindreds. In family III, a homozygous c.912G>T transition was identified resulting in a p.Trp304Cys amino acid change. Both His408 and Trp304 are fully conserved down to zebrafish (Figure S1) and both SIFT [16] and PolyPhen [17] programs predict these changes to be damaging for protein function. On basis of a similar clinical presentation, DOLK was sequenced in DNA of family IV. A third homozygous mutation (c.3G>A, Figure 1D) was identified that removes the initiator methionine residue (p.Met1Ile), which is conserved from human to zebrafish. All three mutations were not present in >1000 healthy Caucasian controls as shown by high resolution melting analysis, by exome sequencing, and by using data from the 1000 genomes project (www.1000genomes.com) (Text S1). Analysis of the protein coding sequence of DOLPP1 in families I and II did not show any sequence variations.
Activity of dolichol kinase was assessed in patient fibroblast homogenates using dolichol-19 as acceptor and γ32P- cytidine 5′-triphosphate (CTP) as phosphate donor. Analysis of 32P incorporation into dolichol-P clearly showed strongly reduced enzyme activity for five patients of all four families investigated (Figure 3B). Fibroblasts from CDG-I patients with a different genetic defect showed dolichol kinase activity comparable to controls.
SEC59 is the yeast ortholog of DOLK [18]. A sec59 yeast mutant that displays temperature sensitive lethality as well as an underglycosylation of glycoproteins at the restrictive temperature was used to further confirm the non-functionality of the mutations in our patients. In addition, we have compared the novel mutations with the previously reported mutations in the two DOLK-CDG patients. All strains showed comparable growth at the permissive temperatures of 25 or 32°C, whereas at the restrictive temperature of 37°C only wild-type DOLK supported growth (Figure 4A). N-glycosylation of the same mutant alleles was assessed by western blotting of the vacuolar glycoprotein carboxypeptidase Y (CPY), containing four N-glycan chains (Figure 4B). At the restrictive temperature CPY is underglycosylated in sec59 cells, visualized by the appearance of glycoforms lacking one to four N-glycan chains. Consistent with the cell growth results, wild-type DOLK was able to restore the glycosylation of CPY at 37°C, as evidenced by a shift to the more mature forms of CPY and a decrease of underglycosylated isoforms. All mutants failed to restore glycosylation to the same extent as wild-type DOLK. The p.Tyr441Ser and p.Cys99Ser mutants showed no or marginal improvement, respectively, as compared to sec59 cells. The three new mutants, however, improved the glycosylation to a higher extent as the glycosylation patterns showed a more prominent band of CPY with two N-glycans as compared to CPY with only one N-glycan.
To explain the tissue-restricted clinical phenotype in our patient group, we performed expression analysis of DOLK in fetal and adult tissue and biochemical analysis of the dolichol-phosphate dependent N-glycosylation and O-mannosylation. Highest expression levels of DOLK mRNA were found in fetal and adult brain, followed by skeletal muscle and heart in fetal tissue and heart in adult (Figure 5).
Dolichol-P is required for N-glycosylation in the ER. In addition, dolichol-P is converted to dolichol-P-mannose, the monosaccharide donor for N-glycosylation inside the ER lumen and for O-mannosylation of alpha-dystroglycan. O-mannosylation was assessed by direct immunofluorescence staining of a frozen heart biopsy with the IIH6 antibody directed against the O-mannosyl glycans of alpha-dystroglycan. Reduced and/or fragmented staining was observed, more pronounced in the right ventricle. The intensity of the sarcolemmal proteins beta-dystroglycan and beta-sarcoglycan was normal, while the intensity of intracellular desmin was somewhat increased (Figure 2). Western blotting was performed on heart muscle homogenates. IIH6 staining of WGA-enriched fractions was reduced (Figure 6A), which was confirmed in the laminin-overlay (LO) assay showing a reduction of the laminin-binding capacity of alpha-dystroglycan. To correct for muscle specific staining, western blotting was performed on non-enriched heart homogenates (Figure 6A) using anti-desmin and anti-β-sarcoglycan primary antibodies. Equal signals were observed for control and patient materials. However, the laminin-overlay assay clearly showed a reduction in signal intensity, similar to the results in WGA-enriched fractions. Additional control studies were performed in heart tissues of patients with idiopathic cardiomyopathy (Figure 6A, PC), with similar results as for the healthy controls (HC). N-glycosylation was analyzed by western blotting of the lysosomal glycoprotein CD63 (LAMP3). In fibroblasts of a DPM1-CDG patient, a clear shift was seen to a lower glycosylated CD63 isoform, indicating aberrant N-glycosylation of CD63 as compared to control fibroblasts. Glycosylation of CD63 in dolichol kinase deficient fibroblasts was comparable to controls (Figure 6B). Analysis of homogenized heart tissue showed a shift in CD63 isoforms in dolichol kinase deficient heart material as compared to control heart, indicating reduced N-glycosylation.
In a cohort of 11 patients, presenting primarily with nonsyndromic dilated cardiomyopathy at the age of 5–13 years, we identified three separate mutations in DOLK as the underlying cause of disease. Some of the patients showed mild additional clinical symptoms, such as ichthyosis, failure to thrive and mild neurological involvement. In contrast, the two families with DOLK mutations originally described by Kranz et al [9] showed a severe congenital multisystem phenotype including a variable presentation of cardiac failure, severe muscular hypotonia, and ichthyosis, with epilepsy due to hypsarrhythmia, microcephaly and visual impairment, leading to death within 6 months after birth. Dolichol kinase is an endoplasmic reticulum resident protein with a cytidine-5′-triphosphate (CTP) binding pocket in the C-terminal domain that is exposed to the cytoplasmic face [20]. The exact catalytic mechanism, the hydrophobic binding sites for dolichol, and a possible role in dolichol-P recycling have not been clarified as yet. Our and the previously identified mutations occur in or near transmembrane domains, not associated with a specific function (Figure S1). Functional investigation of these mutant alleles in the temperature sensitive yeast strain sec59, deficient in dolichol kinase activity, showed sustained reduced growth at 37°C. Moreover, a less severe underglycosylation of CPY was found in the three new mutants as compared to the two mutations from the previous report, which is in agreement with the milder clinical phenotype in our families.
DOLK mutations result in abnormal N-glycosylation as determined by analysis of serum transferrin glycosylation in our patients. Remarkably, only minor classical symptoms of CDG-I could be identified in the patients described here, such as increased liver transaminases in some and a slight decrease in coagulation parameters in all patients. No signs of cerebellar hypoplasia were observed, as commonly seen in the most frequent CDG subtype PMM2-CDG. On the other hand, dilated cardiomyopathy is uncommon in CDG patients with an N-glycosylation defect. A single case out of more than 40 known ALG6-CDG (MIM 603147) patients was reported with a multisystem presentation including DCM [21]. In DPM3-CDG (MIM 612937), DCM was reported as minor symptom compared to the muscular dystrophy [22]. As deduced from deficient IIH6 staining in skeletal muscle, both clinical symptoms were linked to deficient O-mannosylation of alpha-dystroglycan. In the disorders of dystroglycan O-mannosylation, a subgroup of the congenital muscular dystrophies, DCM is commonly observed in combination with limb-girdle muscular dystrophy at the milder end of the spectrum. Patients with dilated cardiomyopathy and no or minimal muscle involvement were reported with mutations in fukutin (FKTN, [23]) and fukutin-related protein (FKRP, [24]), showing reduced laminin binding capacity of alpha-dystroglycan in heart muscle biopsies. The involvement of dystroglycan O-mannosylation in the phenotype of our cohort of dolichol kinase deficient patients was shown by reduced IIH6 staining in frozen heart biopsy material. Western blot analysis showed a reduction in the laminin binding capacity of alpha-dystroglycan, thereby confirming a loss of alpha-dystroglycan function. Recently, the loss of functional alpha-dystroglycan as extracellular receptor in cardiac myocytes was shown to be the cause of dilated cardiomyopathy in mutant mice [25]. Dystroglycan was postulated as an important extracellular matrix receptor to limit the damage of cardiomyocyte membranes after exercise-induced stress to individual cells.
O-Mannosylation of alpha-dystroglycan (Figure 7) involves the protein O-mannosyltransferases POMT1 and POMT2 and the GlcNAc transferase POMGnT1. In addition fukutin, fukutin-related protein and LARGE have been shown to be involved in O-mannosylation, where LARGE is involved in a phosphorylation process of the O-mannosyl glycan [26]. Defects in these six genes have been described as cause for the dystroglycanopathies [27], explaining disease in only about 50% of the patients [28]. Defects in the biosynthetic genes of the sugar donor dolichol-P-mannose required for the O-mannosylation process, like DPM3 [22] and likely DPM1 [29], [30], result in abnormal dystroglycan O-mannosylation. Here, we show that mutations in dolichol kinase also lead to reduced dystroglycan O-mannosylation, likely via reduced availability of dolichol-P-mannose. This is supported by previous studies in yeast cells [31]: amphomycin, which binds to and inhibits the use of dolichol-phosphate, was shown to reduce the production of dolichol-P-mannose with a subsequent reduction of protein O-mannosylation. Interestingly, N-glycosylation of the O-mannosylating enzymes POMT1 and POMT2 was shown to be required for their activity [32]. This implies that in dolichol kinase deficiency, O-mannosylation could be reduced via two independent mechanisms, i.e. via reduced availability of dolichol-P-mannose and via reduced activity of O-mannosylation enzymes due to deficient N-glycosylation. Possibly, this leads to increased susceptibility of the O-mannosylation pathway in defects of dolichol-P or dolichol-P-mannose synthesis. In contrast, the clinical phenotype of CDG-If (MPDU1-CDG, MIM 609180) does not include muscular dystrophy or dilated cardiomyopathy in spite of the reduced availability of dolichol-P-mannose in the ER lumen in this disease [33], [34]. Also, the recently described polyprenol reductase SRD5A3-CDG (MIM 612379) does not show signs of a congenital muscular dystrophy [35], [36]. Apparently, additional factors play a role in determining the clinical outcome in deficiencies of dolichol-P-mannose synthesis or utilization. For SRD5A3, a by-pass synthesis route for dolichol was postulated [35], while both dolichol and dolichol-phosphate could have a function on their own in organelle membrane fluidity [37]. Clearly, many factors in dolichol and dolichol-phosphate homeostasis remain to be discovered [38], which could differentially affect the clinical outcome in dolichol cycle defects.
In conclusion, we have shown that dolichol kinase deficiency results in abnormal N-glycosylation and reduced O-mannosylation of alpha-dystroglycan, leading to a clinical phenotype of dilated cardiomyopathy. This new entity of cardiomyopathy warrants screening for glycosylation defects in any patient with idiopathic DCM. Dolichol kinase deficiency may initially present with mild or asymptomatic DCM which may deteriorate, underlining the necessity to follow these young patients closely.
Three families residing in the Galilee regions of Northern Israel and one Indian family were clinically and genetically investigated. Over the past five years, 11 children have been diagnosed as suffering from an autosomal recessive dilated cardiomyopathy associated with CDG type I transferrin isoelectric focusing profiles in serum (see pedigrees in Figure 1). Nine patients and 11 of their close relatives were included in the study. The study protocol was approved by the Institutional Ethics Review Committee and by the National Committee for Genetic Studies of the Israeli Ministry of Health. Informed consent was obtained from all participants and their legal guardians.
Transferrin isoelectric focusing was carried out as described before [39]. The clinical symptoms did not show any indication for the presence of fructosemia or galactosemia as possible secondary cause for CDG type I transferrin isoelectric focusing profiles. A protein polymorphism was excluded by neuraminidase digestion of the samples and by the normal profiles of both parents. Phosphomannomutase activity was measured in patient fibroblasts according to [40]. For analysis of dolichol kinase activity, fibroblast homogenates were incubated with [γ-32P]cytidine 5′-triphosphate and dolichol-19 and the formation of 32P-dolichol was measured according to [9] and described in detail in Text S1.
Genomic DNA was extracted from peripheral blood lymphocytes using standard salting out procedures [41]. Genotyping was performed using the Affymetrix NspI 250 K SNP array. All SNP array experiments were performed and analyzed according to manufacturer's protocols (Affymetrix, Santa Clara, CA, USA). Homozygosity mapping was performed using PLINK v1.06 [42], using a homozygous window of 50 SNPs tolerating two heterozygous SNPs and ten missing SNPs per window.
Primer sequences for amplification of the only exon of DOLK (GenBank ID NM_014908.3) are shown in Table S1. PCR products were sequenced using the ABI PRISM BigDye Terminator Cycle Sequencing V2.0 Ready Reaction Kit and analyzed with the ABI PRISM 3730 DNA analyzer (Applied Biosystems, Foster City, USA).
Immunohistochemistry was performed by incubation of heart tissue sections with monoclonal antibodies against alpha-dystroglycan, beta-dystroglycan, beta-sarcoglycan or desmin (Text S1).
WGA-enriched and non-enriched heart homogenates were used for western blotting of CD63, beta-dystroglycan, desmin, beta-sarcoglycan and alpha-dystroglycan and for the laminin overlay assay as described ([43] and Text S1).
For expression of DOLK, the following strain was used: MATa sec59 ura3-52. Cells were grown in selective YNB (yeast nitrogen base) medium (0.67% YNB, 0.5% casamino acids and 2% glucose) or in YPD medium (1% yeast extract, 2% bacto-peptone and 2%glucose). For growth on plates 2% agar was added. To construct the yeast expression plasmids, the DOLK open reading frame (encoded by a single exon) was PCR amplified (Phusion High-Fidelity DNA Polymerase, New England Biolabs) from the chromosomal DNA of patients I-3, III-1, and IV-2 and a healthy control with primers engineered with HindIII and BamHI restriction sites at the 5′and 3′ ends, respectively (Table S1). For patient IV-2, where the mutation is located within the primer region, the mutation was introduced in the primer. All four products were subcloned into the pCR4-TOPO vector (Invitrogen, Breda, The Netherlands). The two mutations described previously by Kranz et al. [9] were introduced in the WT-DOLK containing plasmid by site-directed mutagenesis. Subsequently, the WT and mutated forms of DOLK were digested with HindIII/BamH1 and ligated into the HindIII/BamHI digested vector pVT100-ZZ, thereby placing DOLK under the control of the constitutive ADH1 (alcohol dehydrogenase 1) promoter. The correct sequences were verified by sequencing the entire coding region of the constructs. Transformation into yeast cells was carried out using standard techniques [19].
|
10.1371/journal.ppat.0030119 | Genome Dynamics of Campylobacter jejuni in Response to Bacteriophage Predation | Campylobacter jejuni is a leading cause of food-borne illness. Although a natural reservoir of the pathogen is domestic poultry, the degree of genomic diversity exhibited by the species limits the application of epidemiological methods to trace specific infection sources. Bacteriophage predation is a common burden placed upon C. jejuni populations in the avian gut, and we show that amongst C. jejuni that survive bacteriophage predation in broiler chickens are bacteriophage-resistant types that display clear evidence of genomic rearrangements. These rearrangements were identified as intra-genomic inversions between Mu-like prophage DNA sequences to invert genomic segments up to 590 kb in size, the equivalent of one-third of the genome. The resulting strains exhibit three clear phenotypes: resistance to infection by virulent bacteriophage, inefficient colonisation of the broiler chicken intestine, and the production of infectious bacteriophage CampMu. These genotypes were recovered from chickens in the presence of virulent bacteriophage but not in vitro. Reintroduction of these strains into chickens in the absence of bacteriophage results in further genomic rearrangements at the same locations, leading to reversion to bacteriophage sensitivity and colonisation proficiency. These findings indicate a previously unsuspected method by which C. jejuni can generate genomic diversity associated with selective phenotypes. Genomic instability of C. jejuni in the avian gut has been adopted as a mechanism to temporarily survive bacteriophage predation and subsequent competition for resources, and would suggest that C. jejuni exists in vivo as families of related meta-genomes generated to survive local environmental pressures.
| Campylobacter jejuni is the major cause of bacterial food-borne illness worldwide. Predation of C. jejuni by virulent bacteriophage offers the prospect of controlling bacterial populations at source in poultry. We report that in chickens, bacteriophage resistance is infrequent because the mutants that escape bacteriophage are not proficient in poultry colonisation but readily revert back to colonisation-proficient phage-sensitive types. Bacteriophage resistance is generated by reversible genomic scale inversions, leading to the activation of an unrelated bacteriophage integrated into the bacterial genome. These data not only suggest that bacteriophage therapy of C. jejuni would remain a sustainable measure to reduce poultry contamination but also demonstrate how bacterial genomes can evolve under the strong and widespread pressure of bacteriophage predation in the environment.
| The Gram-negative bacterium Campylobacter jejuni is now recognised as a major cause of human gastroenteritis worldwide [1] and has been linked to serious neurological sequelae such as Guillain–Barré syndrome and Miller–Fisher syndrome [2]. Poultry are considered a major source of C. jejuni infections in humans, though numerous other risk factors have been proposed, including the consumption of pork, barbequing, living or working on farms, working in slaughterhouses, seasonal changes in flying insect populations, travel abroad, and the consumption of raw milk [3–8].
Targeted control of food-borne pathogens generally requires identification of the major route of transmission and thereby the most effective place to control infection. For C. jejuni, however, the ubiquitous presence of the organism in the environment, and the sporadic nature of the disease, coupled with the inherent genetic heterogeneity, make the task of tracing of individual strains, and thereby the source of infection, extremely difficult [9,10].
The molecular mechanisms behind this extensive diversity are not fully understood. However, C. jejuni exhibits slip-strand mutation within homopolymeric tracts, which is thought to alter the expression of a significant number of genes [11]. The majority of these genes have been identified as being involved in the production of surface structures, including key fitness determinants such as motility [12–14] and lipo-oligosaccharide synthesis [15]. C. jejuni is also known to be naturally competent under environmental conditions [16,17], though analyses of multi-locus sequence typing (MLST) profiles indicate that short lengths of DNA (less than 3 kb) are involved [18,19]. Intra-genomic recombination has been observed in C. jejuni [20,21] and C. fetus [22–24] but these events are reported to be highly localised and limited in size. Larger scale intra-genomic recombination events leading to genome diversity have, however, been reported for a wide range of other bacterial species [25–31].
As part of a study to investigate bacteriophage therapy and its impact on Campylobacter populations in poultry, we report that chromosomal inversions of up to 590 kb that include the origin of replication of C. jejuni arise in response to exposure to virulent bacteriophage. These inversions are associated with bacteriophage resistance, an inability to colonise chickens without reversion to bacteriophage sensitivity, and the production of a functional Mu-like bacteriophage. These data have profound implications with respect to the evolution of pathogen genomes under the strong and widespread pressure of bacteriophage predation in the environment, and the propagation of prophage under conditions in which host populations are falling.
Following bacteriophage CP34 treatment of chickens colonised by C. jejuni HPC5, a series of CP34-insensitive isolates were recovered and examined by Loc Carrillo et al. [32]. The frequency of resistance was found to be 4% from the intestinal contents of these birds. Pulsed-field gel electrophoresis (PFGE) analysis of these isolates following SmaI digestion indicated that a number of these strains had novel PFGE macro-restriction profiles (MRPs) compared to the parent strain HPC5, though some bands were clearly related. Two novel PFGE-MRPs were observed to be common, and an example of each type was selected for further analysis. These two isolates, R14 and R20, shared five of seven SmaI bands with HPC5, but contained two novel bands of approximately 420 kb and 240 kb in R14 and 170 kb and 125 kb in R20 (Figures 1 and S1). The combined size of the novel bands in R14 and R20 was approximately equal to those of the missing bands from HPC5, indicating that gross loss or gain of genetic material was unlikely.
To rule out the possibility of contamination, strains R14 and R20 were analysed by MLST. They were found to have identical MLST profiles to HPC5 (type 356), thus confirming their origin. R14 and R20 had indistinguishable growth characteristics in vitro compared to HPC5, where the resistant phenotypes were stable for five passages representing at least 100 generations. To examine the stability of R14 and R20 in vivo, the strains were administered to chickens in the absence of bacteriophage CP34. The colonisation potentials of R14 and R20 after 5 d following administration of log10 8 colony-forming units (CFU) of each strain were determined to be log10 7.1 (±0.5) CFU g−1 for R14 and log10 6.7 (±0.3) CFU g−1 for R20, not significantly different from that of HPC5 (log10 6.9 [±0.2] CFU g−1). However, of the recovered isolates tested, almost all (98%) had reverted to bacteriophage sensitivity (n = 100). Clearly, the resistance phenotype has a large fitness cost associated with it such that it is rapidly out-competed in chickens by a relatively low initial number of sensitive revertants.
PFGE of revertant strain DNAs demonstrated all to possess MRPs different from those of their respective parent strains, where the SmaI fragments involved were those that discriminated R14 or R20 from HPC5 to link the phenotype of bacteriophage sensitivity to the observed genomic changes (Figure S2). The MRPs of the revertant strains fell into several distinct classes, termed R14-A, R14-B, R20-A, R20-B, and R20-C, with reference to their parent strains. There was an unequal distribution of the MRPs, with a preponderance of R14-B (75% of R14 revertants) and R20-A isolates (80% of R20 revertants).
To determine whether similar genomic alterations could be observed in vitro, a number of growth curves were performed using C. jejuni HPC5 and bacteriophage CP34. After 24 h of growth, 91% of the isolates were resistant to CP34 (n = 148). However, genomic alterations were not observed in any of these, and it is assumed that resistance arose through a different mechanism, such as point mutation of the receptor. Motility assays indicated that these strains were essentially non-motile, identifying the flagella or motility as being involved in resistance. In contrast, strains R14 and R20 derived in vivo were as motile as the parent strain HPC5.
Binding assays were performed to determine whether the resistance observed in R14 and R20 was receptor mediated or an abortive infection. Bacteriophage CP34 was capable of binding to HPC5, exhibiting a 98.7% drop in titre following a 90-min period of co-incubation. When incubated with either R14 or R20, however, CP34 showed no reduction in titre at the end of the 90-min period (Figure 2A).
Initial colonisation experiments indicated that the bacteriophage-resistant strains R14 and R20 were compromised in their ability to colonise broiler chickens because they were found to revert to bacteriophage sensitivity. This was characterised further by examining the colonisation response of broiler chickens to a range of Campylobacter doses for each strain (Figure 3). These data show that when administered at higher doses, all of the strains tested achieved similar colonisation values at 48 h. However, at the lowest doses of log10 1.9 CFU and log10 2.8 CFU, the mean cecal colonisation values of R14 and R20 were determined to be log10 3.9 (±0.5) CFU g−1 and log10 3.7 (±0.6) CFU g−1, respectively. These were significantly different (p = 0.003 and p = 0.002, respectively) from those of the parent strain HPC5 (log10 5.7 [±0.7] CFU g−1) and the revertant strains R14-A (log10 6.3 [±0.5] CFU g−1), R14-B (log10 6.3 [±0.3] CFU g−1), R20-A (log10 5.8 [±0.7] CFU g−1), R20-B (log10 5.7 [±0.5] CFU g−1), and R20-C (log10 5.4 [±0.6] CFU g−1).
To identify the elements responsible for the MRP changes observed in the genomes of the HPC5 derivatives, a series of restriction maps were created for these strains. SmaI sites were located by the PCR amplification of genomic DNAs using primers designed on the basis of the SmaI sites present in the genome of C. jejuni NCTC11168, followed by digestion of the PCR product with SmaI. The presence of SmaI sites within specific gene sequences were found not to vary between strains, and thus point mutation within the SmaI sites was discounted as the reason for the MRP changes. Once identified, digoxigenin (DIG)-labelled probes for the genes immediately adjacent to the SmaI sites were created along with probes spanning the SmaI sites and used for Southern hybridisation against transferred SmaI-PFGE DNA. This essentially created a range of SmaI restriction maps. Comparison of these maps for the various strains indicated that the genomes were essentially co-linear, but could be divided into sections. The polarity of these sections with respect to each other varied between the strains, indicating that genomic rearrangements involving considerable regions of the genome had occurred (up to 590 kb in R14 and 220 kb in R20). The polarities of the 540-kb and 100-kb SmaI fragments present in the R14 change and the 190-kb and 100-kb SmaI fragments in the R20 change were reversed (Figure 1). Since the SmaI sites are asymmetrically distributed, this affected the observed MRP. It is noticeable that the generation of R14 involves a rearrangement about the origin of replication (located within the 130-kb SmaI fragment), whilst the rearrangement to generate R20 does not.
This procedure was similarly carried out on the R14-A, R14-B, R20-A, R20-B, and R20-C strains to demonstrate that these strains had undergone further genomic rearrangements, all utilising the SmaI-PFGE bands that were observed to change in the R14 and R20 genomic profiles (Figures 4 and S2). However, the rearrangements observed were not all simple reversions to the HPC5 MRP; rather, the most common isolates (R14-B and R20-A) were the result of two separate events. These data indicate that the HPC5 lineage contains three genomic locations capable of recombining with each other. Free recombination between these locations would result in eight genome configurations derived from HPC5. The strains described here represent four of eight of these potential arrangements. All of the rearrangements involve a central location within the 100-kb SmaI fragment of HPC5, which limits the permutations possible to four, all of which are observed. This central location appears key to the generation of the counter-selective phenotypes of bacteriophage resistance and inefficient chicken colonisation that are selected upon exposure to virulent bacteriophage.
To identify the sites of recombination, a system of chromosome walking using long-range PCR, Southern hybridisation, and direct sequencing from genomic DNA was developed. Using the genes adjacent to the SmaI sites as an anchor, long-range PCR was performed with the C. jejuni NCTC11168 genome as a guide. Once linked by PCR, DIG-labelled probes were created for individual genes to determine in which SmaI-PFGE band the gene was located. In places where the gene order in the HPC5 lineage diverged from that of NCTC11168, a system of sequencing directly from genomic DNA and inverse PCR was employed to determine the identity of the adjacent sequences. These DNA sequences appear in Figures S3–S8. The rearrangement end points were determined to be within copies of Mu-like prophages, similar to the prophage identified in C. jejuni RM1221 [33]. However, whereas RM1221 contained a single copy of the prophage, HPC5 and its daughter strains contained two complete copies and a partial copy (ORFs CjE0227 to CjE0241) at distinct genomic locations. In HPC5, the complete copies are between the 3′-end of Cj1470 (CtsF) and the 5′-end of Cj0167c in the 540-kb SmaI fragment (designated CampMu-I), and between the 3′-end of Cj0167 and the 3′-end of Cj1470c (CtsF) in the 100-kb SmaI fragment (CampMu-II), resulting in the disruption of both CtsF and Cj0167c. The partial copy is between an unknown gene and a paralog of CjE0225 in the 190-kb SmaI fragment (CampMu-III). The R14 rearrangement involved recombination between the two complete copies of the CampMu prophage, whilst the R20 rearrangement involved recombination between CampMu-II and CampMu-III. Similarly, the rearrangements to create the R14 and R20 derivative strains took place within these CampMu prophage DNA sequences (Figure 5A).
The discovery that the inversion sites featured CampMu prophage sequences led to studies of whether the CampMu lysogens could be induced to liberate bacteriophage particles. However, it was determined that both R14 and R20 were capable of producing a CampMu bacteriophage without the need for induction. Bacteriophages were produced at a rate of approximately one particle per 50 cells (R14 = 49, R20 = 61). These bacteriophages were examined by transmission electron microscopy (Figure 5B) and identified as corresponding to the CampMu prophage by PCR amplification of DNA extractions from the bacteriophage using CampMu primer pairs, but not from control C. jejuni 16s rDNA primers. Infectious CampMu bacteriophage particles could not be detected in supernatants from HPC5 or from the R14 and R20 revertant strains exhibiting phage sensitivity as tested by titration of the supernatants on all of the strains used in this study and a further panel of 139 independent C. jejuni isolates from broiler chickens, chicken meat, and humans.
The strains capable of supporting the replication of the virulent bacteriophage CP34 (HPC5 and the R14/R20-derived revertants) were also capable of supporting replication of bacteriophage R14-CampMu and R20-CampMu whilst R14 and R20 were resistant to CP34 and the bacteriophage they produce. To compare whether the resistance observed with R14-CampMu and R20-CampMu was due to the failure of the phage to bind the bacterial host in a similar way to CP34 or to abortive infection, binding assays of the CampMu phage were performed under similar conditions (Figure 2B and 2C). Bacteriophage R14-CampMu and R20-CampMu were capable of binding the progenitor host strain HPC5, exhibiting approximately 90% reductions in phage titre after a 90-min incubation. R14-CampMu showed no reduction in phage titre when incubated with the R14 strain producing it, but incubation with R20 produced a 95% fall in phage titre. R20-CampMu showed no evidence of binding to either R14 or R20.
Further evidence for differences between R14-CampMu and R20-CampMu became apparent upon testing the susceptibility of a variety of C. jejuni strains, which revealed that the CampMu bacteriophage exhibited different host ranges (Figure 5C), and that these were maintained following growth of the CampMu bacteriophage on susceptible strains not of the HPC5 lineage. It was also notable that R14-CampMu and R20-CampMu could replicate on independent strains carrying CampMu prophage genes (Figures S9 and S10).
One of the major fears concerning bacteriophage therapy is the potential for bacteriophage-induced genome evolution. Numerous examples exist where temperate bacteriophages are associated with virulence determinants, for example, the genes encoding the toxins of cholera, diphtheria, and verotoxigenic Escherichia coli [34–36]. However, it is generally assumed that using virulent bacteriophage will avoid this problem. This report indicates that virulent bacteriophage have the potential to activate dormant prophage, leading to rapid pathogen evolution; and via host recombination the evolution of temperate bacteriophage leading to the production of chimeric phage with novel phenotypes. However, we also show that whilst pathogen evolution can be rapid, resistance to the therapeutic bacteriophage is associated with a draconian fitness cost that renders the resistant strains non-competitive in the absence of the bacteriophage.
Clearly though, the primary benefit of bacteriophage therapy in this instance is to temporarily reduce the carriage of C. jejuni rather than to eliminate it. Indeed, the ability of C. jejuni to enter what is effectively a transient survival state is evidence of the unusual measures Campylobacter can employ to survive environmental pressures. Recent evidence suggests that Campylobacter is limited in stress response mechanisms [37] and can use genome alterations such as localised frame-shift mutations and slip-strand phase variation to modify gene expression as a substitute for the maintenance of structured regulatory mechanisms [11–15]. The evidence presented here indicates that C. jejuni can use specific genome inversions to survive adverse ecological conditions. Under these conditions, any given Campylobacter recovered is actually but a single representative of a larger family of related meta-genomes under continual flux, the relative proportions of which are dictated by local environmental pressures. Amongst derivate genomes are those in which the origin of replication has been inverted, which could give rise to yet wider changes in gene regulation. This form of chaotic genome regulation is a striking example of the extraordinary strategies adopted by C. jejuni to survive. This type of genomic scale regulation would also suggest that complementary typing methods are required to adequately differentiate C. jejuni strains; methods should be selected that sample the whole genome in parallel with those that are highly discriminatory for smaller sections of the genome. In this example, strains R14 and R20 cannot be differentiated from HPC5 by MLST alone despite the large phenotypic differences observed. A combination of MLST and PFGE methods are required to distinguish these strains and identify them as being different but closely related.
Intra-genomic rearrangements have been reported previously for the flagellin locus of C. jejuni [20,21] and the sap locus of C. fetus [22–24]. However, these rearrangements are relatively short (<5 kb) and highly localised, utilising areas of sequence homology (flagellin) or specific recombination pathways (sap locus). The genome sequence data available for C. jejuni are notable for their lack of repeated sequences, and the completed genomes of NCTC11168 and RM1221 are essentially co-linear; therefore, it would not be unreasonable to suggest that genome rearrangements of C. jejuni are either limited or, given the idea of Campylobacter as a meta-genomic organism, that the observed genome organisations are optimal for in vitro cultures. However, changes to PFGE-MRPs have been noted elsewhere [17,38,39], indicating that chromosomal rearrangements are possible for strains carrying repeated sequences as substrates for homologous recombination such as the prophage sequences documented here.
The observation that the R14 and R20 rearrangements occur in vivo rather than the generation of resistance through mutation of the receptor or a specific binding component is likely a consequence of the essential nature of these components. The frequency at which bacteriophage-resistant mutants are generated in vitro (91%) with HPC5 suggests that there are easier paths to escape bacteriophage predation. However, all the mutants selected in vitro were impaired in motility. Flagella components have been demonstrated to be dominant colonisation factors [40–42], and thus it is not surprising that resistant isolates lacking motility do not survive long in chickens. Bacteriophage CP34 appears to have selected an essential component of the bacteria's intestinal lifecycle, where dense host populations are likely to be most abundant. C. jejuni flagellin is known to be polymorphic and variably glycosylated, leading to differences in sero-specificity [43–46]. Bacteriophage predation may be the direct driving force behind the development of such antigenically variable flagellins rather than host immune evasion, as considered previously [47].
The spontaneous production of CampMu bacteriophages following bacteriophage therapy is of concern because Mu bacteriophages are potential agents of mutation. However, the influence of potential mutator phage needs to be considered against the mutation-driven lifestyle of C. jejuni, which does not carry a full complement of DNA repair mechanisms in the expectation that genomic variation will modify gene expression to overcome adverse conditions. Moreover, evidence suggests that Campylobacter populations are already exposed to CampMu bacteriophage through the mechanism outlined here. Virulent bacteriophages of the family Myoviridae, like CP34, are common in chickens harboring campylobacters. Isolation rates of around 20% in United Kingdom conventional broiler flocks, and more frequently in environmentally exposed free-range and organic flocks, have been reported [48–50]. A recent survey of C. jejuni and C. coli isolates found that 19 of 67 and two of 12 of the respective isolates contained at least one prophage gene [51]. This corresponds well with the four of 12 positive C. jejuni strains reported here. If these figures are representative of general Campylobacter populations, then the likelihood is that these processes are quite common.
The recombination events leading to the strain variants reported here are centred on a 9-kb region of DNA sequence that is shared between prophages CampMu-I, -II, and -III (genes CjE0227 to CjE0241). Recombination between CampMu-I and CampMu-II gives rise to the R14 genome that produces bacteriophage R14-CampMu, and recombination between CampMu-III and CampMu-II gives rise to the R20 genome that produces R20-CampMu. These events lead to the generation of chimeric CampMu prophage in which the genes CjE0242 to CjE0273 adjacent to the recombination resolution of R14 are exchanged, and genes CjE0215 to CjE0226 adjacent to the recombination resolution of R20 are exchanged. These exchanges enable functional excision of CampMu bacteriophages with different gene contents that are themselves distinguishable by their Campylobacter host range (Figure 5C).
What these events have in common is that they lead to resistance to the virulent phage CP34 that is unable to bind the host bacterium. The non-binding of CP34 may arise through two potential mechanisms: 1) changes in host surface structures that are required for phage adsorption; or 2) receptor saturation, if CP34 shares a receptor recognition site with the CampMu phages, and these sites are saturated in the R14 and R20 cultures that produce them. Changes in the surface structures of host bacteria leading to bacteriophage immunity often accompany the state of lysogeny and are mediated through the acquisition of additional genes, commonly known as morons, the control of which are generally independent of the regulation of prophage within which they are sited [52]. In the case of R14 and R20, the change in surface structure expression would have to be associated with the activation of the prophage, for although a wider set of genes other than those affected by the recombination resolution site could be influenced by the gross inversions, the second site reversion events that reinstate phage sensitivity as a consequence of the chromosome rearrangements would militate against the inversions themselves being responsible for the change in phenotype. A consequence of the prophage control of surface structures is that campylobacters carrying CampMu may be biased for certain phage types through the regulated expression of their receptors, or indeed for specific environments according to the need of the organism to express these surface structures. In the latter case, there would be strong selective pressure to inactivate the prophage even at the expense of inverting significant parts of the genome to reassert the control necessary to respond to alternative environments.
Bacteriophage R14-CampMu can bind strain R20, suggesting the receptor site for the phage is still available on this strain, but despite this the R14-CampMu phage does not form plaques on R20, indicating there is likely an underlying resistance mechanism that results in abortive infections. However, neither bacteriophage R14-CampMu nor R20-CampMu are able to bind the respective C. jejuni strains that produced them, and therefore receptor saturation remains a plausible mechanism by which R14 and R20 prevent super-infection.
Considering the above, it is of interest to contemplate how multiple prophage copies have been fixed within the HPC5 lineage. It is a general tenet that the state of lysogeny renders the bacteria immune to infection by homologous bacteriophage. Therefore, it is unlikely that the naïve HPC5 precursor was lysogenised by multiple copies of the same bacteriophage, though not of course impossible. Indeed, it is more likely that a single prophage was present and replicated itself by transposition during the first stages of prophage lytic multiplication. It is known that the position of DNA replication forks influences the location of transposition [53], and the equidistant spacing of the CampMu-I and CampMu-II copies about the origin of replication may indicate that these prophage inserted here as a result of the presence of replication forks symmetrically arranged around the origin. This does lead to the question as to why the replication of the putative CampMu phage was not carried through to completion, namely, the lysis of the host cell. A potential answer to this is a recombination event. The R14 genome structure is similar to that of C. jejuni NCTC11168 and RM1221, whereas HPC5 has a section of genome of reversed polarity. It is possible the R14 configuration represents the original bacteriophage-negative progenitor that became lysogenised by CampMu. When this prophage began to replicate by transposition, the sudden presence of extensive regions of homology allowed recombination within the genome of this strain. Presumably, this recombination led to strain HPC5, where the CampMu prophage was inactivated, and the cell survived. This is supported by the fact that the CampMu cannot be recovered from HPC5, and yet frequently exits the cell in R14 and R20, suggesting that in HPC5, the Mu is inactivated. If true, this is another example of how flexibility within the C. jejuni genome has enabled it to survive the induction of a lysogenic bacteriophage that should have resulted in cell death, and to capitalise on the outcome through evasion of virulent bacteriophage.
The genes present in the partial copy of the CampMu prophage (CampMu-III) have previously been identified as being present en masse in a variety of Campylobacter [51] that lack the other prophage genes (CjE0215 to CjE0226 and CjE0242 to CjE0273). Analysis of the unique sequences adjacent to CampMu-III in HPC5 indicates that these have similarity to bacteriophage genes from other sources, most notably to a phage major tail tube protein from C. jejuni 260.94. It would appear that genes CjE0227 to CjE0241 represent a module of a CampMu genome (CampMu-III) comprising a central region similar to that of the RM1221 CampMu, but flanked by novel prophage genes. Recombination between prophage genomes leads to exchange of these modules and the evolution of the prophage genome. Intra-chromosomal recombinations between the prophage in HPC5 are a direct example of such events, producing chimeric bacteriophage that can exploit differing host ranges.
Campylobacters were cultured on blood agar plates (blood agar base No. 2 with 5% defibrinated horse blood; Oxoid, http://www.oxoid.com/) in gas jars under microaerobic conditions (5% O2, 85% N2, 10% CO2) at 42 °C for 24 to 48 h. Growth curves were conducted by inoculating log10 7 CFU of the C. jejuni into 100 ml of nutrient broth No. 2 (Oxoid) and incubating at 42 °C under microaerobic conditions with 100 rpm orbital rotation.
Bacteriophage CP34 was propagated on C. jejuni HPC5 and recovered using a plate lysis method and stored at 4 °C in SM buffer [32]. Bacteriophage R14-CampMu and R20-CampMu were recovered from blood agar plate cultures of either C. jejuni R14 or R20 by swabbing into SM buffer and passage through a 0.2-μm filter to remove bacteria. Testing of Campylobacter strain susceptibility to bacteriophage was performed as described previously [48]. The susceptibility of Campylobacter strains to bacteriophage R14-CampMu and R20-CampMu was tested by growth of the appropriate Campylobacter strain in 100 ml of nutrient broth No. 2 in the presence of log10 3 plaque-forming units (PFU) ml−1. Samples were recovered and bacteriophage enumerated before and after growth for 24 h at 42 °C on HPC5.
Bacteriophage binding assays were performed to determine whether insensitivity to bacteriophage was due to surface or intracellular factors. Overnight Campylobacter growth from blood agar plates was swabbed into nutrient broth No. 2, centrifuged at 13,000g for 1 min, and the cell pellet resuspended in nutrient broth No. 2. The cells were washed in this manner twice more and, upon final resuspension, were adjusted to contain log10 10 CFU ml−1 as estimated from OD600. Bacteriophage was added at concentrations of log10 4–5 PFU ml−1 to the Campylobacter suspension and incubated at 42 °C with 100 rpm shaking under aerobic conditions for 90 min. Samples were taken at 0 and 90 min, filtered through a 0.2-μm filter, and stored at 4 °C until enumeration of the bacteriophage.
Campylobacter-free Ross broiler chickens were used to determine the colonisation of different Campylobacter strains in the presence and absence of bacteriophage. To ensure that the experimental birds remained free of naturally occurring infection, faeces and cloacal swabs were taken each day from hatch and tested for Campylobacter by direct plating on CCDA agar and for Salmonella by enrichment in Rappaport–Vassiliadis soya peptone broth (Oxoid), then plating on xylose-lysine desoxycholoate agar (Oxoid). Birds were dosed with Campylobacter at 21 d of age and with bacteriophage where applicable at 25 d of age. Following sacrifice, the ceca, upper (proximal small intestine) and lower intestines of the birds were first separated by ligature and then removed by sterile dissection. The lumenal contents were collected for Campylobacter and bacteriophage isolation as described previously [32].
MLST was performed as described previously [54] with reference to the C. jejuni MLST database (http://pubmlst.org/campylobacter/) to determine the sequence alleles.
PFGE was carried out on SmaI-digested genomic DNA and compared to the known profiles of the test strains [55].
Campylobacter DNA isolation was carried out by using GenElute Bacterial Genomic DNA purification kit (Sigma-Aldrich, http://www.sigmaaldrich.com/) or Wizard Genomic DNA purification kit (Promega, http://www.promega.com/). Bacteriophage genomic DNA isolation was performed according to standard procedure [32] using proteinase K digestion followed by phenol-chloroform extraction and precipitation.
Oligonucleotide primers were designed using the NCTC11168 and RM1221 sequences (Sigma-Genosys, http://www.sigmaaldrich.com/Brands/Sigma_Genosys.html). A list of primers used in this study can be found in Table S1. PCRs were performed in 50-μl volumes using a Techne Progene thermal cycler. Reactions consisted of 2.5 U AccuTaq DNA polymerase (Sigma-Aldrich), each dNTP at 500 μM (Promega), forward and reverse primers at 400 nM each, 2% v/v dimethyl sulphoxide, and 100–500 ng of genomic DNA as template in AccuTaq DNA polymerase buffer. DIG-labelled probes for Southern hybridisation were synthesised by PCR with the replacement of the 400 μM dTTP with dTTP at 368 μM and DIG-11-dUTP at 32 μM (Roche, http://www.roche.com/).
Sequencing of PCR products was carried out by MWG Biotech AG (http://www.mwg-biotech.com/) using the ValueRead system. Direct sequencing of genomic DNA was achieved using the same system but with 20 μg of genomic DNA prepared using the Wizard Genomic DNA purification kit.
DNA fragments separated in PFGE gels were transferred to Hybond N+ nylon membranes (Amersham Biosciences, http://www.gelifesciences.com/) using the capillary method. Hybridisation probes were synthesised by PCR as described above. Hybridisations were performed overnight at 42 °C using DIG Easy Hyb. Buffer (Roche). The membranes were blocked using 1% blocking reagent (Roche) before antibody binding with 150 mU ml−1 anti-DIG-AP in 1% blocking reagent. Colour development was performed by incubation in 100 mM tris-HCl (pH 9.5), 100 mM sodium chloride, 0.45 mg ml−1 nitro-blue tetrazolium chloride, and 0.175 mg ml−1 5-bromo-4-chloro-3-indolyl-phosphate, 4 toluidine salt.
Bacteriophage particles at log10 8 PFU ml−1 were absorbed onto a glow-discharged carbon-coated Pioloform grid and stained with uranyl acetate. These were examined using a JEOL 100CX transmission electron microscope (http://www.jeol.com/) operating at an acceleration voltage of 80 kV.
C. jejuni strains were grown on blood agar overnight. A loop of bacteria was inoculated into the centre of a motility plate (Mueller–Hinton broth with 0.4% agar) and grown micro-aerobically for 24 h. Motility was assessed as a function of the radius of the motility halo.
DNA sequences associated with this manuscript appear in the following supplementary figures with the corresponding GenBank (http://www.ncbi.nlm.nih.gov/Genbank/index.html) accession numbers: Figure S3, Cj1468-CjE0213 (EF581842); Figure S4, ORF0656-Cj0167 (EF581846); Figure S5, miaA-CjE0213 (EF581841); Figure S6, ORF0656-Cj1470 (EF581845); Figure S7, Unk-CjE0241 (EF581844); Figure S8, Unk-CjE0227 (EF581843). |
10.1371/journal.ppat.1005608 | Uropathogenic E. coli Exploit CEA to Promote Colonization of the Urogenital Tract Mucosa | Attachment to the host mucosa is a key step in bacterial pathogenesis. On the apical surface of epithelial cells, members of the human carcinoembryonic antigen (CEA) family are abundant glycoproteins involved in cell-cell adhesion and modulation of cell signaling. Interestingly, several gram-negative bacterial pathogens target these receptors by specialized adhesins. The prototype of a CEACAM-binding pathogen, Neisseria gonorrhoeae, utilizes colony opacity associated (Opa) proteins to engage CEA, as well as the CEA-related cell adhesion molecules CEACAM1 and CEACAM6 on human epithelial cells. By heterologous expression of neisserial Opa proteins in non-pathogenic E. coli we find that the Opa protein-CEA interaction is sufficient to alter gene expression, to increase integrin activity and to promote matrix adhesion of infected cervical carcinoma cells and immortalized vaginal epithelial cells in vitro. These CEA-triggered events translate in suppression of exfoliation and improved colonization of the urogenital tract by Opa protein-expressing E. coli in CEA-transgenic compared to wildtype mice. Interestingly, uropathogenic E. coli expressing an unrelated CEACAM-binding protein of the Afa/Dr adhesin family recapitulate the in vitro and in vivo phenotype. In contrast, an isogenic strain lacking the CEACAM-binding adhesin shows reduced colonization and does not suppress epithelial exfoliation. These results demonstrate that engagement of human CEACAMs by distinct bacterial adhesins is sufficient to blunt exfoliation and to promote host infection. Our findings provide novel insight into mucosal colonization by a common UPEC pathotype and help to explain why human CEACAMs are a preferred epithelial target structure for diverse gram-negative bacteria to establish a foothold on the human mucosa.
| Mucous surfaces are a hallmark of the nasal cavity and the throat as well as the intestinal and urogenital tracts. These surfaces serve as primary entry portals for a large number of pathogenic bacteria. To get a foothold on the mucosa, bacteria not only need to tightly attach to this tissue, but also need to overcome an intrinsic defence mechanism called exfoliation. During the exfoliation process, the outermost cell layer, together with attached bacteria, is released from the tissue surface reducing the microbial burden. A comprehensive understanding of the molecular strategies, which bacteria utilize to undermine this host defence, is currently lacking. Our results suggest that different bacterial pathogens have found a surprisingly similar answer to this problem by targeting a common set of proteins on the tissue surface. Accordingly, these bacteria express unrelated proteins that engage the same host receptors called CEA-related cell adhesion molecules (CEACAMs). Binding of microbes to CEACAMs triggers, via intracellular signaling pathways, an increased stickiness of the infected cells. Thereby, the pathogens suppress the release of superficial host cells from the tissue and effectively block exfoliation. Detailed mechanistic insight into this process and the ability to manipulate exfoliation might help to prevent or treat bacterial infections.
| During evolution bacteria have developed fascinating strategies to colonize multicellular organisms. A first critical step, which in many cases determines the outcome of the microbe-host encounter, is the ability of the microorganisms to establish themselves on mucosal surfaces [1,2]. Attachment to the mucosa is facilitated by specific bacterial adhesins, which firmly connect the microbe to the tissue [3,4]. Indeed, adhesin-mediated bacteria-host interactions prevent mechanical removal of the microbes via mucociliary cleansing or urinary flow, and can be seen as a prerequisite for efficient colonization. However, mucosal epithelia have several additional tissue-intrinsic defense mechanisms that protect the surface from adherent pathogens [5]. For example, in both stratified as well as single-layered epithelia the superficial cells are constantly replaced from a stem cell population. This tissue turnover also leads to shedding of cell-associated microbes from the epithelium reducing the bacterial burden. Epithelial tissue turnover can be very fast, as in the intestinal epithelium, where the superficial cells on the exposed villus folds are continuously replaced every day and where this process helps to maintain intestinal homeostasis. Indeed, slowing down tissue turnover in the intestinal tract can facilitate pathogen colonization [6,7]. Similar to the single-layered epithelium of the gut, stratified epithelia of the urogenital tract are also subject to continuous tissue renewal, albeit at a lower rate. However, exposure to high numbers of bacteria can trigger an accelerated turnover, whereby large amounts of superficial epithelial cells are released, a mechanism also known as exfoliation [8–12]. Exfoliation is an innate protective mechanism that, via rapid detachment and shedding of the infected superficial cells, limits colonization of the tissue by the microflora and ultimately prohibits further penetration of the bacteria [13]. By this process, even cell-associated bacteria can be removed from the tissue surface together with the infected cells.
Recently, we could show that specialized bacteria, which colonize the human urogenital tract, are able to suppress the exfoliation response [14]. These bacteria utilize outer membrane adhesins, the so-called OpaCEA proteins, to bind to members of the CEACAM family, a group of immunoglobulin-related gylcoproteins expressed on the apical membrane of mucosal epithelial cells (for review see [15]). CEACAM engagement by bacteria triggers activation of integrins, enhances matrix adhesion and reduces cell detachment of infected cells, ultimately facilitating bacterial colonization [14,16]. Besides Neisseria gonorrhoeae, which was studied in these previous investigations, also the closely related Neisseria meningitidis expresses CEACAM-binding Opa proteins and exploits mucosal CEACAMs such as CEACAM1 or CEA (the product of the CEACAM5 gene) for contacting host cells and for colonization of the nasopharynx [17–19]. In both instances, the pathogens selectively bind to human, but not other mammalian CEACAM family members [20], suggesting that this exquisite recognition mechanism is a result of the co-evolution of these microbes with their sole natural host. Indeed, by interfering with epithelial exfoliation OpaCEA-expressing Neisseria should have a clear advantage during colonization of the human mucosa. It is currently unclear if the expression of OpaCEA proteins is sufficient to counteract exfoliation or if additional neisserial virulence factors are involved in this process. This aspect is of particular interest, as several unrelated bacterial pathogens, such as Haemophilus influenzae, Moraxella catarrhalis, and pathogenic strains of Escherichia coli, have been shown to possess distinct CEACAM-binding adhesins (for an overview see [21]) and might trigger similar processes.
To address this question, we have investigated if engagement of human CEACAMs by a bacterial CEACAM-binding adhesin is sufficient to counteract exfoliation and to promote mucosal colonization. Indeed, we find that expression of a neisserial OpaCEA protein in non-pathogenic E. coli allows these bacteria to engage CEA on epithelial cells, to trigger increased integrin activity and cell-matrix adhesion, and to promote mucosal colonization. Importantly, uropathogenic E. coli (UPEC), harboring a CEACAM-binding adhesin of the Afa/Dr family, also exploit this mechanism to block epithelial exfoliation and to boost their ability to colonize the urogenital tract in vivo. Together, our results establish human CEACAMs as a preferred epithelial target structure, which allows diverse gram-negative bacterial pathogens to suppress exfoliation and to efficiently colonize the human mucosa.
Previous work demonstrated that a CEACAM-binding adhesin is necessary to allow Neisseria gonorrhoeae to increase extracellular matrix binding of infected cells [14,16]. To test if CEACAM-binding alone is sufficient to trigger this process, we separated the CEACAM-binding OpaCEA adhesin from other gonococcal virulence determinants by expressing this neisserial outer membrane protein in E. coli. In the used expression plasmid pTrc, OpaCEA-protein expression is under control of the IPTG-inducible lac promoter [22]. Due to the leakiness of this promoter, OpaCEA protein-expression was already observed under non-inducing conditions in OpaCEA-expressing E. coli (E. coli OpaCEA) compared to an E. coli strain harbouring the empty pTrc plasmid (S1A Fig). The OpaCEA protein expressed in E. coli showed a similar size to the native OpaCEA protein expressed in N. gonorrhoeae, but levels were only about 20% of that observed in gonococci (S1A Fig). Both E. coli strains showed a comparable growth pattern in liquid culture, indicating that expression of the OpaCEA protein in E. coli at this level did not interfere with growth (S1B Fig). In contrast, strong overexpression, e.g. upon IPTG induction, can retard E. coli growth (S1C Fig). When expressed in E. coli, the neisserial Opa protein adhesin was functional with regard to CEACAM-binding, as E. coli OpaCEA was able to associate with the GFP-tagged amino-terminal domain of CEA (CEA-N; Fig 1A). The control E. coli strain did not bind to soluble CEA-N, and no binding of either E. coli strain to the CEACAM8-amino-terminal domain (CEA8-N) was observed (Fig 1A). This binding pattern of the heterologously expressed OpaCEA protein is in agreement with the CEACAM binding profile observed for N. gonorrhoeae expressing this OpaCEA protein (Fig 1A). To investigate, whether E. coli OpaCEA interacts with CEA in a cellular context, we infected the human cervical epithelial cell line ME-180 for 2 h with the E. coli control strain or with E. coli OpaCEA. ME-180 cells endogenously express CEACAM1, CEACAM6, and CEA, which were located in the plasma membrane of uninfected cells (S2A and S2B Fig). Upon infection with the E. coli control strain, the distribution of CEACAMs was unchanged and these bacteria did not associate with ME-180 cells (Fig 1B and S2B Fig). Importantly, E. coli OpaCEA strongly adhered to ME-180 cells and triggered re-location and local concentration of CEACAMs at sites of bacteria-host cell contact (Fig 1B and S2B Fig; arrowheads). Similarly, OpaCEA protein-expressing gonococci adhered in large numbers to ME-180 cells and induced CEACAM clustering (Fig 1B and S2B Fig). Together, these results demonstrated that the neisserial adhesin is functionally expressed in E. coli, where it promotes interaction with CEA-expressing cells.
CEA engagement by pathogenic N. gonorrhoeae has been shown to trigger increased matrix adhesion and to counteract the detachment of epithelial cells [14]. Therefore, we infected confluent monolayers of ME-180 cells, grown on collagen, for 14 h or left the ME-180 cells uninfected. Next, monolayers were washed to remove detached and loosely adherent cells and the remaining cells were stained with crystal violet. Elution and quantification of the dye in a spectrophotometer served as a measure of the remaining adherent cells. Interestingly, in samples infected with OpaCEA protein-expressing N. gonorrhoeae as well as in samples infected with OpaCEA protein-expressing E. coli no reduction in the amount of adherent cells compared to uninfected cells was observed (Fig 1C). Indeed, ME-180 cells infected with CEACAM-binding bacteria even showed a slightly higher recovery than uninfected cells suggesting that cell-matrix adhesion is reinforced upon infection with OpaCEA protein expressing bacteria (Fig 1C). In contrast to CEACAM-binding bacteria, cells infected with the E. coli control strain or with piliated, non-CEACAM-binding gonococci (Ngo P+) showed pronounced detachment of cells (Fig 1C). Microscopic observation of the infected cell cultures corroborated the massive loss of epithelial cells from monolayers infected with control E. coli or Ngo P+, whereas cell numbers in samples infected for prolonged times with OpaCEA protein-expressing bacteria were even higher than in uninfected samples (Fig 1D). Similar results were obtained, when primary human vaginal epithelial (hVECs) cells were infected for 14 h with either OpaCEA protein-expressing bacteria or non-CEACAM-binding strains. Again, expression of a CEACAM-binding adhesin was able to block the infection-induced detachment of the primary epithelial cells (S3A and S3B Fig). Scanning electron microscopy revealed that ME-180 cells infected for 14 h with control E. coli reduced cell-cell-contacts and rounded up, whereas cells infected with OpaCEA protein-expressing E. coli remained well-spread on collagen similar to uninfected cultures and similar to cells infected with OpaCEA-protein expressing gonococci (Fig 1E). These results demonstrate that CEACAM engagement via a CEACAM-binding adhesin, even when expressed in a heterologous background, is sufficient to counteract the detachment of infected epithelial cells in vitro.
Both human primary vaginal epithelial cells as well as ME180 cervical carcinoma cells endogenously express members of the CEACAM family, in including CEA (S2B and S3C Figs). To rigorously test the contribution of CEACAMs to this process, we employed 293 cells, a human cell line that lacks endogenous CEACAM expression. Upon transient transfection with a CEA-encoding plasmid, about 40% of the cell population showed surface expression of the receptor (S4A Fig). Control transfected or CEA-expressing cells were then infected for 5 h with the indicated bacterial strains and used in cell adhesion assays on collagen. We again observed increased matrix binding of CEA-expressing cells upon infection with OpaCEA protein-expressing E. coli (S4B Fig). The adhesion of CEA-expressing cells infected with E. coli OpaCEA was comparable to the increased cell adhesion seen upon infection with OpaCEA protein-expressing gonococci. In contrast, incubation with the E. coli control strain did not alter matrix adhesion of the infected cells (S4B Fig). Importantly, 293 cells transfected with an empty control plasmid (pcDNA) did not display changes in adhesiveness, irrespective of the bacterial strain used for infection (S4B Fig). These results demonstrate that it is the adhesin-CEACAM interaction, which serves as the trigger for increased extracellular matrix adhesion of the infected cells.
Previously, CEACAM engagement by bacteria has been shown to induce CD105 expression in epithelial cells, which was a pre-requisite for enhanced matrix adhesion of infected cells [16]. Indeed, infection of CEA-expressing 293 cells with OpaCEA protein-expressing E. coli or N. gonorrhoeae resulted in the presence of CD105 on the surface of the cells, whereas uninfected cells or cells infected with control E. coli did not have detectable CD105 on their surface (S4C Fig). In line with the idea that CEACAM-triggered expression of CD105 is critical for increased cell adhesion, expression of CD105 in 293 cells led to strongly elevated cell-matrix adhesion in the absence of bacterial infection or CEACAM stimulation (S4B Fig).
Though bacterial infection or CD105 expression resulted in alterations in cell-matrix adhesion, the amount of surface exposed integrin β1 was unaltered in all infected samples compared to the uninfected control (S5A Fig). However, infection with CEACAM-binding gonococci or OpaCEA protein-expressing E. coli promoted a conformational change of integrin β1 as detected by the conformation-sensitive anti-integrin β1 antibody 9EG7 (S5B Fig). Increased labeling by antibody 9EG7 indicated that CEA-engagement by bacteria led to pronounced activation of integrins. In contrast, integrin activity remained low in uninfected cells or cells infected with non-CEACAM-binding bacteria (S5B Fig). The increased integrin activity clearly depended on the presence of CEA, as 293 cells transfected with the empty control vector did not alter the amount of surface integrin nor integrin activity upon infection with diverse bacteria (S5C and S5D Fig). As a further control, cells were stimulated with Mn2+, an exogenous activator of integrins. Upon Mn2+ addition, a similar, maximal integrin activity was observed in all samples demonstrating that the total activatable integrin levels on the cell surface were similar (S5B and S5D Fig). Together, these findings imply that E. coli OpaCEA, via engagement of CEACAMs, can trigger CD105 expression, which in turn enhances integrin activity in vitro.
Because OpaCEA-expressing E. coli was able to enhance integrin activity and suppress cell detachment in vitro, we next analyzed bacterial colonization of the urogenital tract. For this purpose, we used either wildtype mice or transgenic mice, expressing human CEA on all mucosal surfaces (CEAtg mice) [14,23]. Accordingly, 8–10 week old female mice were vaginally infected with 1 x 106 bacteria and colonizing bacteria were recovered by urogenital swabs 24 hours later. Only few bacterial colonies of the E. coli control strain could be isolated from wildtype mice and a slightly elevated (~3-fold) recovery of this non-CEACAM binding strain from CEA-tg mice was observed (Fig 2A). In contrast, more than 30-fold higher numbers of OpaCEA protein-expressing E. coli were recovered from CEAtg mice than from wildtype mice (Fig 2A). Analysis of re-isolated OpaCEA protein-expressing E. coli showed that expression of the Opa adhesin was unaltered by in vivo growth conditions (Fig 2B). Immunohistochemical staining of tissue sections from the urogenital tract revealed that OpaCEA-expressing E. coli were closely associated with the CEA-positive tissue surface (Fig 2C). In the case of E. coli control only few bacteria could be detected on the mucosal surface (Fig 2C). Importantly, CD105 was expressed by mucosal epithelial cells of CEAtg mice in contact with OpaCEA-protein expressing E. coli, whereas the E. coli control strain did not trigger CD105 expression (Fig 2D). Furthermore, E. coli did not trigger CD105 expression in wildtype mice, irrespective of the Opa protein status of the bacteria (S6 Fig). These data demonstrate that OpaCEA-expressing E. coli are able to associate with CEA-positive epithelial cells and trigger CD105 expression in the urogenital tract in vivo. Together, these data suggest that a CEACAM-binding adhesin is not only necessary, but also sufficient to promote colonization of the mucosal surface via stimulating CD105 expression and enhanced integrin activity in superficial epithelial cells.
Besides the neisserial OpaCEA proteins, additional CEACAM-binding bacterial proteins have been described. For example, uropathogenic E. coli harbouring the Afa/Dr locus of afimbrial adhesins have also been shown to engage CEACAMs including CEACAM1 and CEA [24]. Therefore, we used the uropathogenic E. coli (UPEC) strain A30, which expresses the AfaE-III adhesin (E. coli AfaE-III) encoded within the afa gene cluster on a large virulence plasmid [25]. To monitor AfaE-III-dependent events, we cured strain A30 from the virulence plasmid generating the AfaE-III-negative strain E. coli ΔAfaE-III, which lacks the afa gene cluster (S7A Fig). The wildtype UPEC strain and the ΔAfaE-III strain grew with similar growth kinetics (S7B Fig). Clearly, E. coli AfaE-III was able to associate with human CEACAM1 and CEA in the form of soluble GFP-tagged receptor domains and this property was lost in E. coli ΔAfaE-III (Fig 3A). Similar to other CEACAM-binding bacteria such as gonococci, E. coli AfaE-III selectively bound to human CEACAM1, but not to CEACAM1 orthologues from other mammalian species, including mouse, dog and cattle (Fig 3B). Furthermore, E. coli AfaE-III strongly associated with CEA-expressing ME-180 cells and clustered CEA on the surface of the infected cells, whereas E. coli ΔAfaE-III hardly attached to the cell surface (Fig 3C). These results suggest that E. coli pathovars associated with urogenital infections target human CEACAM family members present on epithelial cells of the urogenital tract via their AfaE-III adhesin.
To investigate, if a CEACAM-binding pathogenic E. coli strain is able to modulate the matrix-adhesion of infected epithelial cells, we again employed 293 cells transfected either with a GFP-encoding control vector or a CEA-encoding expression vector. As observed for CEACAM-binding gonococci, E. coli AfaE-III promoted cell-matrix adhesion in 293 cells expressing CEA, but not in control transfected cells (Fig 4A). Neither the E. coli control strain nor E. coli ΔAfaE-III led to enhanced extracellular matrix adhesion of infected cells (Fig 4A). In agreement with the enhanced extracellular matrix adhesion, CEA-expressing 293 cells did not detach upon infection with E. coli AfaE-III, whereas detachment of cells infected with non-CEACAM-binding E. coli (E. coli control strain or E. coli ΔAfaE-III), could be readily detected under the microscope (Fig 4B and 4C). Treatment of the infected cells with Mn2+, a general inducer of integrin activity, increased the matrix adhesion of cells infected with the E. coli control strain or E. coli ΔAfaE-III, but did not further enhance collagen binding of cells infected with CEACAM-binding E. coli AfaE-III (Fig 4D). Moreover, infection of CEA-expressing 293 cells with E. coli AfaE-III, but not E. coli ΔAfaE-III, triggered CD105 expression by 293 cells (Fig 4E). These results indicate that E. coli AfaE-III is able to exploit CEACAMs to modulate host cell adhesion and to counteract bacteria-induced cell detachment via CD105 expression and integrin activation. In line with this idea, infection with E. coli AfaE-III, but not with but not E. coli ΔAfaE-III, resulted in enhanced integrin β1 activity in CEA-expressing cells (Fig 4F and 4G). These results demonstrate that CEACAM-binding pathogenic strains of E. coli can modulate host cell adhesion and integrin activity on human epithelial cells.
Because CEACAM-binding by E. coli AfaE-III results in enhanced integrin activity in vitro, we wondered whether these pathogens have an advantage during colonization of the urogenital tract. Therefore, wildtype or CEAtg mice were vaginally infected for 24 h with 106 E. coli AfaE-III or with E. coli ΔAfaE-III. 24h after infection, colonization was analyzed by dilution plating. Whereas only low numbers of bacteria were recovered from the urogenital tract of wildtype mice, recovery of E. coli AfaE-III from CEAtg mice increased more than 80-fold (Fig 5A). In contrast, numbers of non-CEACAM-binding E. coli ΔAfaE-III were only slightly (~3-fold) elevated in CEAtg mice compared to wildtype mice (Fig 5A). Immunohistochemistry revealed that numerous E. coli AfaE-III were found associated with the epithelial surface of CEAtg mice, which stained positive for human CEA (Fig 5B). Furthermore, superficial epithelial cells of CEAtg mice infected with E. coli AfaE-III showed local expression of CD105 (Fig 5C). While occasionally non-CEACAM-binding E. coli were also detected on the mucosal surface of CEAtg mice, no increase in CD105 expression was evident (Fig 5C). Together, these results indicate that the enhanced colonization of CEAtg mice by E. coli AfaE-III might be due to modulation of epithelial cell-extracellular matrix adhesion via CD105 expression and integrin activation, which would provide the mechanistic explanation for the observed phenotype.
In the case of gonococcal infection, bacteria-triggered CD105 expression on the epithelial surface translates into suppression of host cell exfoliation in vivo [14]. To investigate the level of host cell exfoliation in response to bacterial infection, we used scanning electron microscopy (SEM) of the urogenital tract. In uninfected wildtype and CEAtg animals, the surface of the upper vaginal epithelium showed few detaching superficial cells, indicating low tissue turnover under these conditions (Fig 6A). However, infection of wildtype mice with E. coli AfaE-III or E. coli ΔAfaE-III triggered a dramatic increase in exfoliation of superficial epithelial cells (Fig 6B and 6C). In strong contrast, infection with the CEACAM-binding UPEC strain did not result in an increased exfoliation of epithelial cells in CEAtg mice (Fig 6B and 6C). Despite the presence of numerous E. coli AfaE-III as well as E. coli ΔAfaE-III on the vaginal epithelium, only the non-CEACAM-binding strain E. coli ΔAfaE-III led to a strong increase in exfoliation (Fig 6B and 6C). The low number of exfoliating cells in CEAtg mice upon infection with E. coli AfaE-III was comparable to the uninfected situation indicating that CEACAM-binding by the AfaE-III adhesin is able to completely suppress the exfoliation response (Fig 6B and 6C). Together, these results suggest that CEACAM-binding pathogenic E. coli, such as E. coli AfaE-III, can interfere with the detachment of infected epithelial cells in a manner reminiscent of OpaCEA-expressing Neisseria gonorrhoeae. Suppression of exfoliation clearly is a means to increase the likelihood of a successful and lasting colonization. Therefore, CEACAM-triggered interference with epithelial exfoliation seems to be more common amongst human pathogens than previously appreciated and appears to be an evolutionarily favourable strategy to colonize human mucosal surfaces.
Regulation of epithelial exfoliation is a particularly effective and rapid innate defense mechanism modulating mucosal colonization by microorganisms. However, there is only limited knowledge how pathogens themselves regulate this process and which molecular factors affect cell exfoliation during the course of an infection.
In this study we provide novel insight into the role of bacterial and host determinants, which modulate the exfoliation of epithelial cells. Based on prior observations with Neisseria gonorrhoeae, we were able to confirm that engagement of members of the CEACAM family on the mucosal surface of the upper vaginal epithelium results in suppression of exfoliation. Importantly, the observed host cell responses were independent of the bacterial background, in which the CEACAM-binding adhesin was expressed, as both OpaCEA protein-expressing gonococci as well as E. coli were able to block exfoliation. These results demonstrate that CEACAM engagement is not only necessary, but also sufficient to promote increased host cell-extracellular matrix adhesion and to counteract exfoliation. Interestingly, a pathogenic UTI isolate of E. coli, which expresses the Dra/AfaE CEACAM-binding adhesin, was able to induce a similar host cell phenotype in vitro, characterized by CEACAM-triggered upregulation of CD105, increased integrin activity, and enhanced host cell adhesion to the extracellular matrix. CEACAM engagement allowed these pathogens to blunt epithelial exfoliation leading to enhanced mucosal colonization in vivo. Based on these findings we propose that CEACAM-binding adhesins have independently evolved in multiple gram-negative bacterial pathogens, including pathogenic Neisseriae, E. coli pathovars, Haemophilus influenzae and Moraxella catarrhalis, as a means to facilitate the initial, species-specific contact with the mucosa of an appropriate host organism and to counteract the detachment of superficial cells.
One of the best studied examples of bacteria-induced exfoliation is taking place in the bladder, where incoming bacteria trigger massive shedding of the superficial umbrella cells, a specialized cell type covering the luminal surface of the bladder urothelium [10,26]. Indeed, while in some organs the epithelium regenerates constantly, the mammalian urinary bladder can shift from a physiological mode of slow tissue turnover to a highly proliferative status as a result of epithelial injury [27,28]. UPEC strains that infect the bladder epithelium are characterized by the possession of specific adhesin gene clusters, such as the operons encoding for the type 1 pilus, the P pilus, or the Afa/Dr family of adhesins, as well as the secretion of toxins such as α-hemolysin (HlyA) [29,30]. In particular, the FimH adhesin-mediated contact of UPEC with bladder cells has been shown to induce pronounced exfoliation and tissue renewal [10]. Detachment of the large superficial urothelial cells, which seems to be accompanied by apoptosis, then affords the pathogen access to deeper strata of the bladder epithelium, where some bacteria invade, multiply, and persist in undifferentiated epithelial cells [31,32]. Accordingly, a small fraction of FimH-expressing UPEC seems to profit from exfoliation, as increased urothelial stem and early progenitor cell proliferation provide an expanded protective niche for UPEC in the bladder [27]. In line with the idea that in some instances the bacteria might benefit from the host tissue response, Dhakal and Mulvey recently identified the pore-forming toxin α-hemolysin (HlyA) as the bacterial effector that induces exfoliation upon infection with UPEC [33]. At sublethal concentrations, HlyA activates host cell proteases resulting in breakdown of integrin-associated proteins such as paxillin, thereby weakening cell-matrix attachment.
Interestingly, the finding by Dhakal and Mulvey that a secreted factor promotes cell detachment, also demonstrates that host cell contact is not a pre-requisite for the induction of epithelial exfoliation. This is also reflected in our current and prior studies, where non-adherent bacteria were able to trigger this process in vitro and in vivo [14, 16]. Clearly, the non-pathogenic E. coli or non-opaque gonococci used in these studies do not secrete toxins, pointing to the existence of additional soluble triggers for host cell detachment. It has been speculated that conserved bacterial factors such as LPS could initiate this host response [8]. However, infection of TLR-4-deficient mice with E. coli also resulted in a strong increase in exfoliation, suggesting that LPS, or more precisely LPS sensing by TLR-4, is not required to trigger exfoliation (S8 Fig). Therefore, future efforts should be directed towards identifying the relevant (and most likely conserved) bacterial feature(s), which initiate epithelial exfoliation.
Amongst UTI isolates of E. coli, type 1 pilus or P pilus expression in combination with the secretion of HlyA prevails, which is in line with the idea that these pathotypes profit from epithelial exfoliation [13]. However, a further pathotype of UTI has been described (pathotype V), which lacks HlyA secretion and expresses Afa/Dr-related adhesins [30]. Indeed, non-hemolytic, Afa/Dr-possessing E. coli are found in less than 4% of fecal isolates, but make up almost 10% of UTI isolates suggesting a prominent enrichment of strains with this genetic makeup during urogenital colonization [30]. Even higher rates of Afa/Dr-expressing strains have been reported from cystitis cases in children and pyelonephritis cases of pregnant women [34,35]. Our results with a non-hemolytic, AfaE-III-expressing strain now demonstrate that this pathotype can engage human CEACAMs on the surface of the urogenital mucosa to suppress exfoliation. Though our model relies on infection of the upper vaginal epithelium in mice, rather than the bladder epithelium, pathogenic E. coli can also infect the human genital tract mucosa of both male and female and can be acquired similar to gonococci by sexual transmission [36]. Therefore, our results indicate that amongst isolates some strains may, in contrast to HlyA-expressing strains, be able to suppress exfoliation via the possession of CEACAM-binding Afa/Dr adhesins.
Clearly, CEACAM-binding is only found in a subfamily of Afa/Dr adhesins expressed by E. coli, the so-called Afa/Dr-I family, including Dr, F1845 and AfaE-III adhesins [24,37]. The Afa/Dr adhesins belong to the large group of chaperone/usher (CU) type adhesins, which can be located on fimbriae or which can occur in the form of an afimbrial surface sheath (such as in the case of Afa adhesins) [38,39]. Similar to other CU systems, the Afa adhesins are part of a small gene cluster (afaA–afaF), which, besides the adhesin, encodes the accessory proteins required for surface expression of the adhesin proper [38]. The afa gene clusters comprise genes of transcriptional regulators (afaA, afaF), a periplasmic chaperone (afaB), and an integral outer-membrane protein (afaC), which serves as the assembly platform, the so-called usher, for the surface display of the adhesin [40]. In the case of the afa-3 gene cluster, there are two genes (afaD and afaE), which encode proteins with adhesive and invasive properties towards mammalian cells [41–44]. Isolated, recombinant AfaE-III derived from the afa-3 operon of the UPEC strain A30 has been shown to bind to CD55 as well as members of the CEACAM family including CEACAM1, CEA, and CEACAM6 [37]. However, AfaE orthologues from other pathogenic E. coli strains, such as AfaE-I from strain KS52, seemingly lack CEACAM binding [37]. For AfaE-III, the CEACAM binding interface of the adhesin has been structurally defined [45]. Though the six amino acids of AfaE-III engaged in the binding interface with CEA seem to be largely conserved in the CEACAM-binding Dr protein from strain IH11128, the CEACAM-binding adhesin F1845 encoded by the daaE gene from strain C1845 does not share a single identical amino acid at the corresponding positions [37,46,47].
The lack of a defined CEACAM-binding motif currently prohibits the use of the vast sequence information about E. coli adhesins to predict possible binding interactions with human CEACAMs. Furthermore, a comprehensive functional analysis of CEACAM-binding properties among UTI-causing E. coli isolates and a correlation with the clinical manifestations is currently lacking. It is therefore interesting to note, that a recent survey by Qin et al. found a strong association between the presence of the afa gene cluster and recurrent infections of the lower urogenital tract, which were characterized by a lack of systemic symptoms [48]. In contrast, the afa operon was not detected in strains isolated from acute pyleonephritis or cystitis patients [48]. These findings are in line with the idea that some pathogenic strains of E. coli, e.g. afa-3 harboring strains, by the help of CEACAM-binding adhesins, are able to sustain long-term accommodation in the urogenital tract without causing systemic pathology. Interestingly, also many asymptomatic bacteriuria (ABU) E. coli isolates are hlyA-negative and carry afa gene clusters [49]. Based on our previous study, we have performed an additional extended PCR screening of 126 ABU isolates from various sources and find that 9.3% of these isolates encode afa/dra genes. These ABU strains can colonize the urinary tract with high bacterial numbers for many weeks or months without provoking an overt innate immune response [50]. Accordingly, suppression of exfoliation via Afa/Dr binding to CEACAM may also promote the stable asymptomatic colonization of the bladder by ABU strains. Given the HlyA-triggered induction of exfoliation versus Afa/Dr-mediated suppression of exfoliation, it can be envisioned that the presence of the respective virulence factors might dictate the distinct behaviour of these strains and the clinical outcome.
Importantly, the majority of CEACAM-binding bacteria characterized to date does not colonize the urogenital tract, but rather inhabits the nasopharynx, such as Neisseria meningitidis, Moraxella catarrhalis and Haemophilus influenzae [51–53]. In each case, the bacteria employ structurally unrelated adhesins to contact CEACAMs on the luminal surface of the host tissue, where CEACAM-binding adhesins, including AfaE-III, bind to the protein part and not the glycan part of the receptor [24, 45,54–56]. It is easily conceivable that all these human-restricted pathogens exploit CEACAMs for securing successful colonization of their sole natural host. Indeed, recent experiments with model organisms have demonstrated that adhesin-CEACAM interactions contribute to improved recovery of these bacteria after experimental infection of the nasopharyngeal mucosa [19, 57]. Our in vivo experiments with CEAtg-mice indicate that the recovery of non-CEACAM binding E. coli from CEA-transgenic animals is also slightly, but consistently elevated compared to wildtype animals. It could be speculated that by expressing human CEA in mice, an additional glycoprotein is added to the murine mucosa that could allow additional low affinity, glycan-based interactions. Though such weak interactions might not be apparent in in vitro binding assays with soluble CEA-domains, they could contribute to a slightly improved colonization in vivo. Clearly, there is a strong further increase in colonization of the urogenital tract by CEACAM-binding E. coli and this correlates with the ability of these strains to suppress exfoliation in vivo, an effect not observed for non-CEACAM-binding bacteria. This suggests that potential low affinity, glycan-mediated binding interactions are not sufficient to mediate suppression of epithelial exfoliation. In contrast to the situation in the urogenital tract, it is currently unknown, if the improved colonization of the nasal cavity of CEACAM1 transgenic mice observed for N. meningitidis or the CEACAM-dependent colonization of chinchillas by Haemophilus influenzae is linked to a suppression of exfoliation. However, at least in stratified regions of the nasopharyngeal epithelium similar processes might occur as observed for AfaE-III-expressing E. coli or OpaCEA protein expressing N. gonorrhoeae in the urogenital tract. Together, a detailed understanding of the molecular processes initiating the exfoliation of epithelial cells and the host-tailored countermeasures by successful pathogens holds the promise to provide novel avenues to interfere with or to protect from bacterial colonization right from the start of the bacterial host cell encounter.
The E. coli and E. coli OpaCEA strains were derived from DH5α and have been described previously [22]. The heterologous expression of Opa proteins in E. coli allows these strains to interact with human CEACAMs in a manner analogous to Opa protein expressing gonococci and meningococci [18]. The opaque phenotype was regularly controlled by Western blotting of bacterial lysates with the monoclonal anti-Opa protein antibody (clone 4B12/C11). The uropathogenic E. coli harbouring the afa-3 gene cluster (strain A30; isolated from a cystitis patient [25,58] is a non-hemolytic, serotype O75 strain expressing the CEACAM-binding AfaE-III adhesin (E. coli AfaE-III). All E. coli strains were grown on LB agar plates with or without adequate antibiotics and cultured at 37°C. To select for afa-deficient variants of E. coli AfaE-III, bacteria were sequentially cultured overnight at 37°C, then for 24 h at 40°C, at 42°C, and at 45°C. Serial dilutions of the final culture were plated on LB agar and individual colonies were tested by colony-PCR for the presence of the afa locus (primer pair afa-f: 5’-ggcagagggccggcaacaggc-3’; afa-r: 5’-cccgtaacgcgccagcatctc-3‘) and the K5 capsule determinant (K5-f: 5‘-cagtatcagcaatcgttctgta-3‘ / kpsII-r: 5’-catccagacgataagcatgagca-3’) as described [59].
In addition, we have attempted to generate a complemented strain, but have not succeeded in re-expressing the afa-gene cluster of E. coli strain A30, despite subcloning into low-copy or inducible vectors. Sequencing of subcloned gene clusters in multiple independent clones revealed that all the tested subclones carried point mutations resulting in premature stop codons or frame shifts within the coding regions of the afa-gene cluster, indicating that the maintenance of heterologous fimbrial gene clusters in multiple copies in E. coli laboratory strains may select for subcloned PCR products with accumulated mutations preventing expression. For determining growth curves of E. coli strains, LB broth cultures were initiated and grown at 37°C for 4 h. These cultures were used as 1:25 inoculation into 100 ml LB medium at 37°C. Absorbance at 600nm was recorded every 60 min over a course of 10–25 h using a Libra S4 spectrophotometer (Biochrom, Cambridge, UK). For infection, bacteria were suspended in DMEM, the optical density of the suspension at 600 nm (OD600) was used to estimate the number of bacteria according to a standard curve, and the bacteria were added to the cells at the indicated multiplicity of infection (MOI).
Neisseria gonorrhoeae strains used in this study were described previously and were derived from strain MS11 [16]. The gonococci were either non-piliated and expressed a CEACAM-binding Opa protein (Ngo OpaCEA; strain N309), were non-piliated and expressed a heparansulphate proteoglycan-binding Opa protein (Ngo OpaHSPG; strain N303) [22], or the bacteria did not express an Opa protein (non-opaque phenotype), but expressed pili to bind to human cells (Ngo P+, strain N280, a non-opaque derivative of MS11-F3 [60]). The opacity and the piliation status of the used bacteria were regularly monitored by colony morphology as well as Western Blotting with monoclonal anti-Opa antibodies. Gonococci were grown on GC agar plates (Difco BRL, Paisley, UK) supplemented with vitamins, chloramphenicol (10 μg/ml) and erythromycin (7 μg/ml) at 37°C, 5% CO2 and subcultured daily.
The human cervix carcinoma cell line ME-180 (ATCC, Rockville, MD) and the embryonic kidney cell line 293T (293 cells; ACC-635, DSMZ, Braunschweig, Germany) were cultured in DMEM containing 10% calf serum (293 cells) or DMEM containing 10% FCS (ME-180 cells) at 37°C in 5% CO2 and subcultured every second to third day. 293 cells were transfected by calciumphosphate co-precipitation using a total of 5 μg of plasmid DNA for each 10 cm culture dish and employed in experiments 2 days after transfection. In some cases, 293 cells were serum-starved in DMEM containing 0.5% CS. The human vaginal epithelial cell (hVEC) line MS74 was obtained from A.J. Schaeffer (Feinberg School of Medicine, Northwestern University, Chicago, IL), cultured on gelatine-coated dishes in DMEM containing 10% fetal calf serum (FCS), and subcultured every third day.
The expression plasmids used in this study included the commercially available vectors pcDNA3.1 Hygro (pcDNA; Invitrogen, Karlsruhe, Germany), pEGFP-N1, pLPS-3’-EGFP and pDsRed2-C1 (Clontech, Palo Alto, CA). Plasmid pcDNA3.1 CEA (pcDNA-CEA) was described previously [61]. The expression plasmid pcDNA3.1 CEACAM1-4L-HA (pcDNA-CEACAM1) was constructed by PCR amplification of the human CEACAM1-4L cDNA (generous gift of W. Zimmermann, LMU München, Germany) with primers CEACAM1-HA-sense (5’-GGGAAGCTTGCCATGGGGCACCTCTCAGCCCCACTTCAC-3’) and CEACAM1-HA-anti (5’-GGGGACGTCATAGGGATACTGCTTTTTTACTTCTGAATAAATTATTTCTG-3’) and was cloned into the HindIII-AatII—digested plasmid pBluescript CEACAM3-HA [62] before further subcloning via HindIII-NotI into pcDNA3.1 Hygro (Invitrogen). The vector pLPS-3'-RFP2 was constructed by PCR amplifying the RFP2-coding sequence from vector pDsRed2-C1 with primers RFP2-AgeI-sense 5’-ATAACCGGTCGCCTCCTCCGAGAACGTCATCACC-3’ and RFP2-NotI-anti 5’-ATAGCGGCCGCTTACAGGAACAGGTGGTGGCGGCC-3’ and subcloning into the AgeI/NotI sites of pLPS-3’-EGFP resulting in the exchange of the GFP by the DsRed2 coding sequence. The human CD105 cDNA was transferred by Cre-mediated recombination from pDNR-dual CD105 [16] into pLPS-3'-RFP2 resulting in RFP2 fused to the carboxy-terminus of full-length CD105.
GFP-fusion proteins of the human CEACAM1, CEA, and CEACAM8 amino-terminal IgV-like domains as well as the corresponding constructs encoding the amino-terminal IgV-like domains of CEACAM1 orthologues from dog (cCEACAM1), the two CEACAM1 alleles from cattle (bCEACAM1a; bCEACAMb), and from mouse (mCEACAM1) have been constructed and employed previously [20,63].
293 cells were transfected with 5 μg plasmid DNA encoding secreted GFP-fusion proteins of the N-terminal domains of human or mammalian CEACAM family members. After 24 h, the medium of the transfected 293 cell cultures was replaced by serum-reduced OptiMEM (Life Technologies, Darmstadt, Germany). Two days later the culture supernatants (supe) containing the secreted fusion proteins were collected, centrifuged for 15 min at 5000 rpm and either stored at -20°C or used immediately for bacterial pull-down experiments. The GFP-derived fluorescence was analysed using a Varioskan Flash reader (Thermo Scientific) to adjust equal amounts of the secreted fusion proteins. For pull-down experiments, indicated bacteria were suspended in PBS and binding to the indicated receptor protein contained in cell culture supernatants was determined essentially as described [63].
Immunofluorescence staining, cell lysis and Western blotting were performed as described previously [62] using mAbs against CEACAMs (clone D14HD11) against CEACAM1 (clone GM-8G5), against CEACAM6 (clone 9A6) (all from Aldevron, Freiburg, Germany), against CEA (clone COL-1; Zymed, San Francisco, CA), against green fluorescent protein (GFP; clone JL-8; BD Biosciences), against E. coli LPS (AbD Serotec, Oxford, UK), or against murine CD105 (clone MJ7/18; Southern-Biotech, Birmingham, USA). Mouse monoclonal antibodies against human CD105 (clone P4A4; provided by Developmental Studies Hybridoma Bank (DSHB), University of Iowa) or against Opa proteins (clone 4B12/C11; generous gift of Marc Achtman, University of Warwick, UK) as well as rat monoclonal anti-integrin β1 (clone 9EG7; provided by D. Vestweber (MPI for Molecular Medicine, Münster, Germany)) and rat monoclonal anti-integrin β1 (clone AIIB2; DSHB) were purified from hybridoma culture supernatants. Further, rabbit polyclonal antisera raised against paraformaldehyde-fixed N. gonorrhoeae MS11 (IG-511) was custom produced by Immunoglobe (Himmelstadt, Germany). All secondary antibodies were from Jackson Immuno-Research (West Grove, PA).
For scanning electron microscopy, ME-180 cells were seeded at 2.5 x 105 cells/well in 24-well plates on acid-washed glass coverslips coated with 25 μg/ml collagen and grown to confluency. Medium was replaced with DMEM, 0.5% FCS for 8 h. Then cells were infected for 14 h at a MOI of 20 or left uninfected. Samples were fixed in situ for at least 1 h at 4°C. The samples were washed and dehydrated in a graded series of aceton on ice. After critical point drying samples were sputter-coated with 5 nm gold-palladium in a BAL-TEC SCD 030 and examined at 15 kV accelerating voltage in a Philipps 505 scanning electron microscope using the secondary electron detector. Images were digitally recorded with a DISS 5 system (remX GmbH, Bruchsal, Germany) and processed in Adobe Photoshop 6.
C57BL/6J mice transgenic for human CEA (CEAtg mice) have been described before [14,23]. Wildtype C57BL/6J mice (originally obtained from Elevage Janvier, Le Genest Saint Isle, France) and CEAtg mice as well as TLR4+/+(HeN) and TLR-/- (HeJ) female mice were maintained under specified pathogen-free conditions under a 12-h light cycle in the animal facility of University of Konstanz in accordance with the institutional guidelines. The CEAtg mice were kept heterozygous for the transgene by crossing male CEAtg mice with female WT mice. Offspring of these crosses (age 3–4 weeks old) was genotyped by PCR.
Experiments involving animals were performed in accordance with the German Law for the Protection of Animal Welfare (Tierschutzgesetz). The animal care and use protocol, including the protocol of experimental vaginal infection of female mice, was approved by the appropriate state ethics committee and state authorities regulating animal experiments (Regierungspräsidium Freiburg, Germany) under the permit file numbers G-10/108 and G-15/43.
Experimental vaginal infection of female mice with E. coli strains was performed as previously described for N. gonorrhoeae [14]. Briefly, CEAtg and wildtype mice were subcutaneously injected with 17-β-estradiol 4 days prior to infection. The drinking water was supplemented with ampicillin (1 mg/ml) to reduce the overgrowth of commensal bacteria during hormone treatment. Ampicillin containing water was changed to normal drinking water 1 day before infection. Mice were inoculated intravaginally with 106 CFU of the different E. coli strains suspended in 20 μl of PBS. 24h later, the mucosa-associated bacteria were re-isolated by cotton swaps. Serial dilutions of re-isolated bacteria were plated on LB agar (E. coli AfaE-III) or LB agar containing ampicillin (E. coli OpaCEA and E. coli). In order to analyse the Opa protein profile of re-isolated bacteria, single colonies were expanded on agar plates, lysed and analysed by Western blotting using Opa specific antibodies.
The genital tract of infected animals was excised and the longitudinally opened vaginal and uteral tissue was mounted and analysed by scanning electron microscopy essentially as described previously [14]. For immunohistochemistry, tissue samples were immediately fixed with 4% paraformaldehyde for at least 24 h and transferred to 10% sucrose, 0.05% cacodylic acid for 1 h at 4°C. Next, samples were transferred to 20% sucrose for 1 h and then into 30% sucrose at 4°C over night. Organs were mounted in embedding medium (Cryo-M-Bed; Bright Instrument, Huntingdon, UK) and frozen at -20°C. 10 μm thick sections were cut at -20°C using a cryostat (Vacutom HM500, Microm, Germany). Sections were incubated with a mouse monoclonal antibody against CEA (clone COL-1; dilution 1:200) or a rat monoclonal antibody against murine CD105 (clone MJ7/18; dilution 1:1200) together with a polyclonal rabbit antibody against N. gonorrhoeae MS11 (dilution 1:100) or a polyclonal rabbit antibody against E.coli (dilution 1:200). Detection of the primary antibodies was accomplished by incubation with a combination of Cy5-conjugated goat-anti-rabbit antibody (1:250) and rhodamine-conjugated goat-anti-rat antibody (1:250; in the case of CD105 detection) or Cy3-conjugated goat-anti-mouse antibody (1:250; in the case of CEA detection). Cell nuclei were visualized by the addition of Hoechst 33342 (1:30,000; Life Technologies, Darmstadt, Germany) in the final staining step. Samples were analysed with a TCS SP5 confocal laser scanning microscope (Leica, Mannheim, Germany). Images were digitally processed with Photoshop CS (Adobe Systems, Mountain View, CA) and merged to yield pseudo-coloured images.
293 cells were transfected with pcDNA3.1 Hygro (pcDNA), pcDNA-CEA, or pLPS-3’-RFP2-CD105. Cells were infected or not with E. coli, E. coli OpaCEA, E. coli AfaE-III, or gonococci for 14 h. After infection, cells were stained with monoclonal antibodies against CEA (clone COL-1) or against human CD105 (clone P4A4) for 1 h at 4°C. Following washing, samples were stained with a Cy2-conjugated goat-anti-mouse antibody for 30 min, 4°C. ME180 cells and hVEC cells were analysed for the presence of endogenous CEACAMs using mouse monoclonal antibodies specific for CEACAM1 (clone GM-8G5), CEACAM6 (clone 9A6), or CEA (clone COL-1). Stained samples were analysed for Cy2-derived fluorescence by flow cytometry on an LSRII (BD Biosciences) using FACS Diva software.
Cell adhesion and cell detachment assays were performed essentially as described [14,16]. Briefly, the wells of 96-well plates were coated with PBS containing collagen type 1 from calf skin (ICN Biomedicals, Irvine, CA) for 24 hours at 4°C. 293 cells were transfected with pEGFP-N1 (GFP), CEA or pLPS3’-CD105 (CD105). After serum starvation, cells were infected or not with the indicated bacterial strains at a MOI of 30 for 8 hours. Following infection, the cells were detached and kept in suspension medium (DMEM, 0.2% BSA) with or without 1 mM Mn2+ (1 h at 37C°), and then replated at 4 x 104 cells/well onto collagen-coated wells. Cells were allowed to adhere for 90 min in the presence or absence of 1 mM Mn2+ at 37°C, before non-adherent cells were removed by gentle washing with PBS. For the Mn2+-treated samples, the washing buffer also contained 1 mM Mn2+. Adherent cells were fixed and stained for 60 min with 0.1% crystal violet in 0.1 M borate, pH 9. After washing and drying, the crystal violet was eluted in 10 mM acetic acid and the staining intensity was measured at 550 nm with a Varioskan Flash (Thermo Fisher Scientific Oy Microplate Instrumentation (Vantaa, Finland).
For measuring integrin activity, cells were serum-starved over-night, before they were infected or not with the indicated bacteria at an MOI of 30 for 8 h. Following infection, cells were detached by limited trypsin/EDTA digestion that was stopped by addition of soybean trypsin inhibitor (0.5 mg/ml in DMEM). Detached cells were kept in suspension medium (DMEM, 0.2% BSA) for 1 h at 37°C, and then replated at 5 x 104 cells/well into wells coated with collagen type 1. After 75 min, wells were treated or not with 1 mM MnCl2, incubated for 5 min and transferred to ice. Cells were fixed with 4% paraformaldehyde in PBS for at least 30 min, washed with PBS and permeabilized with Triton X-100 (0.1% in PBS) for 15 min. Cells were washed with PBS and blocked with 2% BSA in PBS (blocking buffer) for 20 min, before incubation for 1 h with rat monoclonal antibodies against active integrin (clone 9EG7; dilution 1:600) or against total integrin (clone AIIB2; dilution 1:750) in blocking buffer as described [14]. After washing and incubation with ProteinA/G-HRP (1:250), 100 μl/well of substrate solution (substrate solution was prepared by mixing 10 ml of 2.4 mg/ml tetramethylbenzidine in 10% acetone, 90% ethanol with 0.5 ml of 30 mM potassium citrate, pH 4.1) were added. The enzymatic colour reaction was stopped using 2 M H2SO4 (100 μl/well) and the absorbance was determined at 450 nm in a Varioskan Flash (Thermo Scientific) microplate reader.
For cell adhesion, cell detachment, and integrin activity assays, values were analysed for normal distribution and mean values were compared by two-tailed unpaired t-test. For in vivo infection assays, including enumeration of exfoliating cells, differences between samples were assessed using the Mann-Whitney U-test. Differences between samples with p<0.001 are indicated by ***.
|
10.1371/journal.pgen.1004907 | Nur1 Dephosphorylation Confers Positive Feedback to Mitotic Exit Phosphatase Activation in Budding Yeast | Substrate dephosphorylation by the cyclin-dependent kinase (Cdk)-opposing phosphatase, Cdc14, is vital for many events during budding yeast mitotic exit. Cdc14 is sequestered in the nucleolus through inhibitory binding to Net1, from which it is released in anaphase following Net1 phosphorylation. Initial Net1 phosphorylation depends on Cdk itself, in conjunction with proteins of the Cdc14 Early Anaphase Release (FEAR) network. Later on, the Mitotic Exit Network (MEN) signaling cascade maintains Cdc14 release. An important unresolved question is how Cdc14 activity can increase in early anaphase, while Cdk activity, that is required for Net1 phosphorylation, decreases and the MEN is not yet active. Here we show that the nuclear rim protein Nur1 interacts with Net1 and, in its Cdk phosphorylated form, inhibits Cdc14 release. Nur1 is dephosphorylated by Cdc14 in early anaphase, relieving the inhibition and promoting further Cdc14 release. Nur1 dephosphorylation thus describes a positive feedback loop in Cdc14 phosphatase activation during mitotic exit, required for faithful chromosome segregation and completion of the cell division cycle.
| During the cell cycle, a specific sequence of events leads to the formation of two daughter cells from one mother cell. Progression through the cell cycle is tightly controlled, with events occurring in the right place at the right time. Exactly how this is achieved is still being elucidated. In budding yeast, the events occurring during the final cell cycle phase – “mitotic exit” – are controlled by the phosphatase Cdc14. It is kept sequestered and inactive until it is needed for mitotic exit, at which time it is rapidly released. In this study, we have identified a new regulator of Cdc14 activity, the protein Nur1. In a series of experiments, we saw that Nur1 acts both upstream and downstream of Cdc14 activation, thereby creating a positive feedback loop. On the one hand, Nur1 contributes to inhibiting Cdc14 until the start of mitotic exit. On the other hand, through the actions of Cdc14 itself, Nur1 is disabled as an opponent of the phosphatase. This creates a robust system, rapidly switching between two opposing states and thus driving forward the mitotic exit transition.
| Cellular reproduction is a highly regulated process that is controlled on a multiplicity of levels, ensuring orderly progression through the different phases of the cell cycle and accurate partitioning of the genome. At the heart of eukaryotic cell cycle control lie cyclin-dependent kinases (Cdks) and their opposing phosphatases [1], [2]. In Saccharomyces cerevisiae, the Cdk subunit Cdc28 associates with a series of cell cycle stage-specific cyclins to bring about Cdk activity. It is opposed by the main Cdk-counteracting phosphatase Cdc14, which reverses Cdk phosphorylation events during mitotic exit [3]. The changing balance between Cdk and Cdc14 phosphatase activities at this stage of the cell cycle serves to order mitotic exit events, such as spindle elongation and chromosome segregation followed by spindle disassembly and ultimately cytokinesis [2], [4]. So far, only a few of the Cdk substrates, whose dephosphorylation brings about mitotic exit, have been characterized [5]–[9]. During mitotic exit, Cdc14 is also essential for the downregulation of Cdk activity, on the one hand by promoting mitotic cyclin degradation, via dephosphorylation of the Anaphase Promoting Complex (APC) activator Cdh1, and on the other by promoting accumulation of the Cdk inhibitor Sic1 [3], [10], [11].
Cdc14 activity is stringently regulated. During most cell cycle phases, Cdc14 is sequestered in the nucleolus, and thus inactive, through inhibitory binding to Net1 [12]–[14]. It is thought that Cdc14 release from Net1 occurs following phosphorylation of the latter, which can be achieved by a series of kinases including Cdk, Polo and MEN kinases, thus reducing Net1 affinity for Cdc14 [15]–[17]. Net1 phosphorylation, and thus Cdc14 release, is prevented until early anaphase by the action of the phosphatase PP2ACdc55, which keeps Net1 under-phosphorylated. At anaphase onset, the protease separase is activated after APC-mediated destruction of its inhibitor securin. Separase now cleaves the chromosomal cohesin complex to trigger sister chromatid segregation. At the same time, separase uses a non-proteolytic activity to downregulate PP2ACdc55 [18], [19]. This swings the phosphorylation balance on Net1 towards phosphorylation by mitotic Cdk which, with additional help from components of the FEAR network [17], [20], initiates Cdc14 release.
While Cdc14 release during the early stages of anaphase depends on mitotic Cdk activity [17], [19], mitotic cyclins are being degraded at this time and Cdk activity is in decline. It is thought that declining Cdk and increasing Cdc14 contribute to activation of the MEN, a G-protein coupled signaling cascade consisting of the GTPase Tem1, its regulators Lte1 and Bub2/Bfa1, and its downstream kinases Cdc15 and Dbf2/Mob1 [21]–[24]. Both Cdc15 and Mob1 are Cdk targets and their dephosphorylation in mid anaphase contributes to MEN activation [20], [25], [26]. Active MEN kinases in turn are thought to have the potential to maintain Net1 phosphorylation. However, how Cdc14 release is sustained while Cdk activity declines between anaphase onset and MEN activation has remained poorly understood.
Timely Cdc14 activation in early anaphase is important for successful chromosome segregation. It is required to stabilize the anaphase spindle and is also required for completion of chromosome segregation, in particular telomeres and the late segregating rDNA locus [6], [7], [27], [28]. Cdc14 promotes the condensation and segregation of the repetitive rDNA region during anaphase and cdc14 mutants display rDNA segregation failure despite unobstructed cohesin cleavage. How Cdc14 promotes rDNA segregation is still being debated. Condensin is recruited to the rDNA in anaphase in a Cdc14-dependent manner, where it appears to promote decatenation of the locus, making the condensin complex a prime candidate for Cdc14 regulation [28]–[30]. It has also been suggested that Cdc14 downregulates rDNA transcription by RNA polymerase I, which could facilitate condensin access to the locus [31], [32]. On the other hand, rRNA synthesis continues unabated during mitotic exit, making this hypothesis appear less likely [33]. In any event, a Cdc14 target that is dephosphorylated to promote rDNA condensation and segregation in anaphase remains unknown.
In this study, we take advantage of our recent phosphoproteome analysis of budding yeast mitotic exit [34]. In the search for Cdc14 targets that have a role in regulating rDNA segregation, we identified the nuclear rim protein Nur1 as a Cdc14 substrate. Failure to dephosphorylate Nur1 causes rDNA missegregation, however, this turns out to be the consequence of compromised Cdc14 activation rather than a specific rDNA segregation defect. This leads us to discover that Nur1 has a previously uncharacterized role in Cdc14 inhibition, and that its inhibitory activity is phosphorylation-dependent. Constitutive Nur1 phosphorylation delays Cdc14 activation, while non-phosphorylatable Nur1 causes premature Cdc14 activation. Thus, Cdc14-dependent Nur1 dephosphorylation in early anaphase forms a positive feedback loop to promote further Cdc14 release, with important implications for faithful chromosome segregation.
In the search for Cdc14 targets that promote rDNA condensation and segregation during anaphase, we reviewed the phosphoproteome of budding yeast mitotic exit. Cells were arrested in metaphase by depletion of the APC coactivator Cdc20, then synchronous mitotic exit progression was induced by ectopic expression of the Cdc14 phosphatase. Mass spectrometry was used to survey the disappearance of phosphopeptides over the course of mitotic exit, with the original intention to identify proteins whose dephosphorylation controls cytokinesis [34].
Among the proteins that were dephosphorylated in response to Cdc14 expression, in addition to cytokinesis regulators, we identified the nuclear rim protein Nur1 (Fig. 1A, B). Along with its binding partner Src1, Nur1 is involved in tethering rDNA and telomeres to the nuclear envelope, its absence leading to decreased rDNA repeat stability, unequal rDNA segregation, as well as loss of telomere stability and silencing [35]–[37]. Nur1 was previously identified as a Cdk target, containing nine putative Cdk phosphorylation sites, of which four have been confirmed in mass spectrometry studies (Fig. 1A) [34], [38]. Of these four, our phosphoproteome analysis covered three phosphorylation sites on two phosphopeptides. These disappeared with early to intermediate timing, relative to the phosphopeptides of all detected proteins, during Cdc14 induced mitotic exit (Fig. 1B).
To confirm cell cycle-dependent Nur1 phosphorylation, and the role of Cdc14 in its dephosphorylation, we monitored the electrophoretic mobility of Nur1 using Phos-tag gels during synchronous cell cycle progression. Cells were arrested in G1 by pheromone α-factor treatment, before release to progress through the cell cycle at 35.5°C and either rearrest in the next G1 phase by readdition of α-factor, or arrest in late mitosis following Cdc14 inactivation using the temperature sensitive cdc14-1 allele [39]. Protein extracts were prepared at the indicated times and cell cycle progression was monitored by FACS analysis of DNA content (Fig 1C). Phos-tag gel analysis of the protein extracts revealed the appearance of slower migrating Nur1 isoforms 30 minutes after release from G1 arrest, coincident with the time of S-phase, presumably due to phosphorylation (Fig. 1C). During undisturbed cell cycle progression, Nur1 reached its slowest migration at 45 minutes, in G2/M, before faster migrating forms appeared again at 60–75 minutes. This pattern is consistent with dephosphorylation during early anaphase, before cells completed cytokinesis at 90 minutes. When Cdc14 was inactivated, in the cdc14-1 strain, dephosphorylation was no longer observed and Nur1 accumulated in slow migrating forms in the late mitotic arrest. To confirm that the mobility shift observed on the Phos-tag gels is indeed due to Nur1 phosphorylation, Nur1 was immunoprecipitated from cells arrested in mitosis by nocodazole treatment and incubated in a control buffer, or in the presence of either λ-phosphatase or purified recombinant Cdc14 [4]. Incubation with either phosphatase, but not control incubation, led to conversion to faster migrating species on the Phos-tag gel, confirming that the slower migrating forms are the consequence of phosphorylation (Fig. 1D).
In order to examine the importance of Nur1 dephosphorylation, we took advantage of a strategy to create constitutively Cdk phosphorylated proteins by covalent fusion to a mitotic cyclin (Fig. 1E) [34], [40]. In brief, gene targeting was used to fuse the NUR1 gene at its genomic locus with a mitotic cyclin Clb2 tagging cassette. Clb2 is modified to lack localization and destruction signals, so as not to interfere with Nur1 function. After gene targeting, NUR1 remains initially separated from the cyclin tag by a selectable marker, flanked by loxP recognition sites for the Cre recombinase. The marker is then removed through the action of β-estradiol-activatable Cre-ER recombinase, following hormone addition to the growth medium. As a control, a similar tag was created in which Clb2 harbors three additional point mutations that prevent it from interacting with and recruiting the Cdc28 kinase subunit (denoted Clb2ΔCdk, see Materials and Methods for details). An HA epitope is also included in the tag to facilitate detection, both before and after excision of the marker.
We found there was virtually complete conversion of Nur1 to Nur1-Clb2 and Nur1-Clb2ΔCdk, respectively, four hours following β-estradiol addition (Fig. 1F). Furthermore, Western blotting revealed a haze of slower migrating forms in case of the Nur1-Clb2 fusion protein, but a sharp, faster migrating band in case of Nur1-Clb2ΔCdk, consistent with increased Nur1 phosphorylation due to the Clb2 fusion.
We next examined the effect of the fusions on cell survival. Cells expressing Nur1-Clb2 are viable at 25°C but unable to grow at a higher temperature of 36°C (Fig. 1G). In comparison, Clb2ΔCdk fusion did not affect cell growth at the high temperature, suggesting that temperature sensitivity is caused by the continuous presence of Cdk activity close to Nur1. As the Nur1-Clb2 fusion appears to cause increased Nur1 phosphorylation, constitutive phosphorylation could be the cause of temperature sensitive growth. Alternatively, phosphorylation of proteins in the vicinity of Nur1, due to the increased local Cdk concentration, could cause the temperature sensitivity. To differentiate between these possibilities, we created a version of Nur1, in which its 9 Cdk consensus phosphorylation sites were replaced by alanines (Nur1(9A)). We then repeated the process of generating Nur1(9A)-Clb2 and Nur1(9A)-Clb2ΔCdk fusions. The absence of Cdk phosphorylation sites on Nur1 restored temperature resistant growth following Clb2 fusion (Fig. 1G). It also prevented the Nur1 mobility shift following Clb2 fusion. (Fig. 1H). This suggests that indeed persistent Nur1 phosphorylation on its Cdk phosphorylation sites is the cause for a temperature sensitive growth defect.
In order to study the effect of persistent Nur1 phosphorylation on rDNA segregation, we tagged the rDNA binding protein Net1 with YFP to visualize the behavior of the rDNA locus. We synchronized a cell population by α-factor block and release at 36°C and monitored rDNA segregation as cells progressed through mitosis. As an internal marker for segregation timing, we recorded the length of the elongating anaphase spindle in each cell, as well as whether or not the rDNA had separated and segregated into two opposite cell halves. In both a wild type control strain, as well as in the Nur1-Clb2ΔCdk or Nur1(9A)-Clb2 strain, rDNA segregation started at a spindle length of 5–6 µm and was complete by the time spindles reached 8 µm in length. In contrast, in Nur1-Clb2 cells, rDNA segregation only started at spindle lengths of 7–8 µm and never reached completion even when spindles were fully elongated (Fig. 2A). Cytological observation of the rDNA locus showed that it reached its expected tightly condensed state in opposite cell halves in wild type and Nur1-Clb2ΔCdk cells (Fig. 2B). In contrast, the rDNA often appeared stretched and uncondensed in Nur1-Clb2 cells. These observations are consistent with the possibility that Nur1 dephosphorylation promotes rDNA condensation, which in turn is required for its timely segregation. Notably, an rDNA segregation defect of similar extent to that in Nur1-Clb2 cells is seen after inactivation of the chromosomal condensin complex [27]–[30]. An rDNA segregation defect was also observed, albeit less pronounced, in Nur1-Clb2 cells at 25°C (S1 Fig.). Taken together, we conclude that constitutive phosphorylation of Nur1 leads to delayed rDNA segregation, coincident with defective rDNA condensation.
A consequence of rDNA segregation defects is unequal sister chromatid exchange and consequent chromosomal instability. As a measure for rDNA stability, we assessed the loss rate of an ADE2 marker within the rDNA repeats [36], [41]. ADE2 loss causes accumulation of a red intermediate metabolite in the adenine biosynthesis pathway, thus causing the colony color to turn red. ADE2 loss during the first cell division after plating on agar medium will generate half red-sectored colonies. We therefore counted the fraction of half red-sectored colonies, among all colonies, as a measure for the ADE2 loss rate from the rDNA. The loss rate was approximately five-fold elevated in the Nur1-Clb2 strain at 25°C, compared to a wild type and Nur1-Clb2ΔCdk controls (Fig. 2C). This is indicative of rDNA instability, probably as the consequence of rDNA segregation defects caused by the Nur1-Clb2 fusion, even at a permissive temperature.
Nur1 is a nuclear rim protein that makes contact with the rDNA, so its phosphorylation status could directly affect rDNA condensation. Alternatively, Nur1 phosphorylation could indirectly affect rDNA condensation and segregation. In particular, delayed rDNA segregation is a hallmark of mitotic exit defects, as for instance observed in FEAR pathway mutants. In order to differentiate between these possibilities, we monitored cell cycle progression of Nur1-Clb2ΔCdk and Nur1-Clb2 cells. Cells were synchronized in G1 by α-factor treatment, released to pass through a synchronous cell cycle at 36°C, before being rearrested in the following G1 by α-factor readdition. FACS analysis of DNA content showed that Nur1-Clb2 cells spent at least 20 minutes longer with a 2C DNA content (Fig. 3A), i.e. show delayed cell cycle progression between G2 and cytokinesis, compared to the control.
To delineate where Nur1-Clb2 cells are delayed in cell cycle progression, we analyzed several cell cycle markers using Western blotting at frequent time intervals during the time course (Fig. 3B). Appearance of securin and Clb2 at around the time of S-phase was indistinguishable between the control and Nur1-Clb2 cells, as was phosphorylation of Orc6 that takes place at that time. However, destruction of both securin and Clb2 were markedly delayed in Nur1-Clb2 cells, as was Orc6 dephosphorylation during mitotic exit and reaccumulation of the Cdk inhibitor Sic1. This pattern suggests a delay of Nur1-Clb2 cells during mitosis and mitotic exit. To distinguish whether the mitotic delay takes place before or after the metaphase to anaphase transition, we monitored spindle morphology during cell cycle progression (Fig. 3C). This revealed that Nur1-Clb2 cells persisted for somewhat longer with short metaphase spindles (1–3 µm in length), as well as a pronounced prolongation of the time that cell persisted with long anaphase spindles (>3 µm in length). These observations are consistent with a delay of Nur1-Clb2 cells both at the metaphase to anaphase transition as well as during mitotic exit. A similar delay in mitotic progression, albeit less pronounced, was observed in Nur1-Clb2 cells at 25°C (S1 Fig.). To confirm that the mitotic delay due to Nur1-Clb2 was caused by persistent Nur1 phosphorylation, we used cells harboring Nur1(9A). Clb2 fusion to Nur1(9A) no longer delayed cell cycle progression (S2 Fig.), indicating that indeed persistent Nur1 phosphorylation slows down mitotic progression. These findings further open the possibility that rDNA condensation and segregation defects seen in Nur1-Clb2 cells are a consequence of a mitotic exit delay.
The mitotic delay observed in Nur1-Clb2 cells is reminiscent of that seen in cells with an inactive FEAR pathway [42]. In those cells, delayed Cdc14 phosphatase activation slows mitotic exit progression and rDNA segregation [27], [28]. Given the phenotypic similarities, we examined the timing of Cdc14 release in Nur1-Clb2, compared to wild type and Nur1-Clb2ΔCdk, cells. Again, we synchronized cells in G1 and then measured the timing of Cdc14 release as a function of spindle length as cells passed through mitosis. In several repeats of this experiment, we detected a small but statistically significant delay to Cdc14 release in the Nur1-Clb2 strain (Fig. 4A). This suggests that phosphorylated Nur1 impedes Cdc14 activation, a likely cause for delayed mitotic exit and rDNA segregation defects.
To investigate whether delayed Cdc14 activation is indeed the cause for mitotic defects in Nur1-Clb2 cells, we tested whether the dominant active CDC14TAB6-1 allele, which leads to Cdc14 being less tightly bound by Net1 [43], restores cell survival at high temperature. Indeed, we found that CDC14TAB6-1 almost completely restored temperature-resistant growth of the Nur1-Clb2 strain (Fig. 4B). As a control, we confirmed that CDC14TAB6-1 did not cause dephosphorylation of Nur1-Clb2, e.g. due to the presence of higher than normal levels of Cdc14 activity. Western blotting revealed that Nur1-Clb2 mobility, in particular its slower migrating forms, remained unaffected by CDC14TAB6-1 (Fig. 4C). This confirms that the temperature sensitive growth due to persistent Cdk phosphorylation of Nur1 is caused by defective Cdc14 activation.
We also monitored the dynamics of cell cycle progression in the Nur1-Clb2 strain rescued by the CDC14TAB6-1 allele. This revealed that the mitotic delay caused by Nur1-Clb2 is largely reduced in cells carrying the CDC14TAB6-1 allele (Fig. 4D). Both the delays at the metaphase to anaphase transition, as well as the delay during mitotic exit, were ameliorated. FACS analysis of DNA content as well as Western blotting analysis of Clb2 levels and of Orc6 dephosphorylation during a synchronous cell cycle confirmed that Nur1-Clb2 no longer affects mitotic progression in CDC14TAB6-1 cells. This indicates that the mitotic defects and associated loss of viability at high temperature in Nur1-Clb2 cells are caused by defective Cdc14 activation.
If Nur1, especially the Cdk phosphorylated form, prevents Cdc14 release in early anaphase, then eliminating Nur1 or removing its Cdk phosphorylation sites, should facilitate Cdc14 release. To investigate this possibility, we compared Cdc14 release kinetics in synchronized populations of wild type, nur1Δ and nur1(9A) cells. Strikingly, Cdc14 was released in over half of nur1Δ and nur1(9A) cells in early anaphase at spindle lengths below 3 µm, when Cdc14 release is almost never seen in wild type cells (Fig. 5A). Even in metaphase cells with short (1-2 µm) spindles, when Cdc14 is normally tightly sequestered, a substantial fraction of nur1Δ and nur1(9A) cells displayed released Cdc14 (Fig. 5B). This suggests that phosphorylated Nur1 restricts Cdc14 release in early anaphase.
Cdc14 activation is promoted by the FEAR network in early anaphase. To further study the impact of Nur1 on Cdc14 release at this time, we introduced the nur1Δ and nur1(9A) alleles into a spo12Δ strain, lacking a key component of the FEAR network [20]. In the spo12Δ background, Cdc14 release is delayed until spindles reach about 6 µm in length. Deletion of nur1, or its replacement with nur1(9A), restored Cdc14 early anaphase release (Fig. 5C). The Cdc14 release profile in the nur1Δ spo12Δ and nur1(9A) spo12Δ double mutant strains approached that of wild type cells. However, at short spindle lengths, nur1Δ spo12Δ and nur1(9A) spo12Δ cells still showed premature Cdc14 release, as compared to wild type, while at longer spindle lengths the rescue did not fully match wild type release levels. These findings confirm that phosphorylated Nur1 is a potent inhibitor of Cdc14 release in early anaphase and that the FEAR network acts to overcome Cdc14 inhibition by Nur1. However, the incomplete rescue of Cdc14 release in spo12Δ cells by Nur1 ablation suggests that the FEAR network acts at least in part by a mechanism different from inactivating Nur1.
We next compared the kinetics of mitotic progression between wild type, nur1Δ and nur1(9A) strains. Despite the advanced Cdc14 release in the latter strains, the timing of progression through mitosis, as measured by the fractions of cells displaying metaphase and anaphase spindles at each time point, was indistinguishable from wild type (Fig. 5D). In a spo12Δ background, cells showed approximately a 20 minute delay in anaphase. In this case, nur1Δ or nur1(9A) restored the kinetics of mitotic progression close to wild type (Fig. 5D). This confirms that Nur1, specifically its phosphorylated form, can delay mitotic progression.
Inactivation of Nur1, by deletion or mutation of its Cdk phosphorylation sites, augments Cdc14 release in early anaphase and almost completely compensates for loss of FEAR network components. This could be because phospho-regulation of Nur1 has a role specific to early anaphase. Alternatively, Nur1 might be a general Cdc14 inhibitor at all stages of mitotic exit. To differentiate between these two possibilities, we tested whether Nur1 inactivation can also compensate for partial loss of MEN signaling. We used conditional thermosensitive alleles in two MEN kinases cdc15-2 and dbf2-2 and asked whether nur1 deletion would improve cell growth at a semi-permissive temperature when MEN signaling is partially compromised. However, cdc15-2 nur1Δ and dbf2-2 nur1Δ double mutants lost viability at increasing temperatures in a fashion indistinguishable from the parental cdc15-2 and dbf2-2 strains (Fig. 6A). Thus, Nur1 does not appear to act later during mitotic exit, when the MEN takes control of Cdc14 release. Instead, Nur1 function as a Cdc14 inhibitor appears to be restricted to early anaphase. This conclusion is consistent with the Nur1 phosphorylation pattern during mitotic exit. Nur1 is Cdk phosphorylated during early mitosis, when it counteracts Cdc14. In anaphase, Nur1 becomes dephosphorylated due to Cdc14 action and thus looses its inhibitory effect on Cdc14.
To investigate the mechanism of how Nur1 counteracts Cdc14, we asked whether Nur1 is physically linked to components of mitotic exit control. Affinity purified fractions of Nur1, analyzed by sensitive mixture mass spectrometry, contained Heh1 and mitotic monopolin, as well as a Net1 peptide [36]. However, while mitotic monopolin resides in the nucleolus, Nur1 could not be detected on the rDNA together with Net1 by chromatin immunoprecipitation [36], [44]. To clarify whether Nur1 interacts with Net1, we performed a co-immunoprecipitation experiment. A Net1-myc strain was synchronized in G1 and cell extracts prepared at regular time intervals following release into synchronous cell cycle progression. Net1 was immunoprecipitated from the extracts using an α-myc antibody. Nur1 coprecipitated with Net1 at all stages of the cell cycle, with little fluctuation to the efficiency of the interaction (Fig. 6B). No Nur1 was recovered in a parallel control immunoprecipitation from extracts of a strain lacking the Net1 myc epitope. In addition, we detected Cdc14 in immunoprecipitates of Nur1 throughout the cell cycle. While we do not currently know with which of Net1 and/or Cdc14 Nur1 makes direct contact, this finding opens the possibility that Nur1 directly influences Cdc14 inhibition in conjunction with Net1.
A key event during Cdc14 activation in early anaphase is Cdk phosphorylation of Net1 on at least six Cdk consensus phosphorylation sites. It could therefore be that Nur1 antagonizes Cdc14 by counteracting Net1 phosphorylation. As a start to investigate this, we took advantage of the net1-6Cdk allele that lacks these six Cdk phosphorylation sites and, as a consequence, delays Cdc14 activation and causes a short mitotic exit delay [17]. If Nur1 impacts on Cdc14 by counteracting Net1 phosphorylation, then Nur1 inactivation will not be able to correct the mitotic exit delay of net1-6Cdk cells. In contrast, if Nur1 counteracts Cdc14 in a pathway different from Net1 phosphorylation, then its deletion should be able to advance Cdc14 activation in net1-6Cdk cells, just as it did in the spo12Δ background. However, nur1 deletion did not improve Cdc14 release nor reduce the mitotic exit delay of net1-6Cdk cells (S3 Fig.). This suggests that Nur1 contributes to Cdc14 regulation most likely by influencing Cdk phosphorylation of Net1. It will be an important task to directly study the influence of Nur1 on Net1 phosphorylation.
The paramount importance of the Cdc14 phosphatase during budding yeast mitotic exit is well established, with the list of its targets and functions increasing. For instance, the incompletely understood role of Cdc14 in cytokinesis has recently come into focus [34], [45], [46]. Concomitantly, our understanding of the regulation of nucleolar release and activation of this phosphatase is deepening [17], [19], [47]–[50].
In this study, we set out to further our molecular knowledge of the role of Cdc14 in rDNA condensation and segregation. A candidate Cdc14 target with potential to impact on rDNA segregation was identified as the nuclear rim protein Nur1 in our recent phosphoproteome screen of budding yeast mitotic exit. Nur1 has been previously linked to regulating rDNA stability. We confirmed Nur1 as Cdc14 substrate in vivo and in vitro, that is dephosphorylated in early anaphase. To study the role of Nur1 dephosphorylation, we created a constitutively phosphorylated form of Nur1, by fusing it to a cyclin Clb2 moiety. The fusion led to an rDNA segregation delay and reduced viability, which was dependent on the Cdk phosphorylation sites, especially at high temperatures. This confirms Nur1 as a Cdc14 target whose dephosphorylation is critical for successful mitotic exit progression.
Further investigation revealed that the overt rDNA segregation defect in Nur1-Clb2 cells was likely a secondary consequence of a primary defect in Cdc14 activation. While the Cdc14 substrate(s) that govern rDNA condensation, resolution and segregation in anaphase therefore remain elusive, we have uncovered a previously unknown Cdc14 substrate that acts to sustain Cdc14 activation, thus creating positive feedback of Cdc14 release in early anaphase (Fig. 7).
While inhibition of Cdc14 release by Nur1 is normally restricted to early anaphase while Nur1 is phosphorylated, our Nur1-Clb2 fusion likely extends that period, potentially throughout mitotic exit. Nur1-Clb2 may thus elicit phenotypic consequences beyond the delayed Cdc14 release characteristic of FEAR network mutants. An example of a difference is the temperature sensitive growth of Nur1-Clb2 cells, which is not typically shared by FEAR network mutants. A higher temperature causes cells to progress faster through the cell cycle, presumably rendering them less tolerant to delays or alterations to cell cycle signaling pathways. Phenotypes of late mitotic exit mutants, in particular those involved in cytokinesis, are known to be exacerbated at high temperatures [34], consistent with an impact of Nur1-Clb2 at later stages.
Intriguingly, defective Cdc14 activation in Nur1-Clb2 cells caused not only a mitotic exit delay, but also a delay at the metaphase to anaphase transition. A small amount of Cdc14 is likely to be active by the time of metaphase, when it contributes to dephosphorylation of proteins like Dsn1 and possibly Spo12 [49], [51]. Furthermore, securin dephosphorylation by Cdc14 has been suggested to sharpen the timing of anaphase [52]. Nur1-dependence of regulating these earliest Cdc14 targets might play a role at the metaphase to anaphase transition. We cannot exclude that Nur1 performs additional functions at this time, that might be affected by Nur1-Clb2 fusion and that might contribute to explaining the lethality of Nur1-Clb2 fusion strains at high temperatures.
Our results demonstrate that Nur1, through regulating Cdc14, plays an important role in promoting the accurate timing of mitotic exit events. Nur1 was previously characterized as chromosome linkage protein, connecting chromatin and the inner nuclear membrane, with functions in the maintenance of genome stability and replicative life span [36], [37]. Are these independent functions of this protein, or are they related?
Constitutive Nur1 phosphorylation, caused by the Nur1-Clb2 fusion, increased marker loss at the rDNA, likely due to unequal sister chromatid exchange. A similar level of marker loss at the rDNA was previously reported in nur1Δ cells [36]. It will be therefore interesting to examine whether Nur1's role in tethering the rDNA to the nuclear envelope is regulated by its phosphorylation status. It is as yet unknown whether rDNA tethering is modulated during the cell cycle, for instance during rDNA condensation and segregation, and how the rDNA moves relative to the nuclear envelope during anaphase. It has been speculated that cell cycle-dependent Nur1 phosphorylation within a putative nuclear localization signal might affect the nucleocytoplasmic shuttling of the protein, with possible implications both for Cdc14 regulation and rDNA tethering [37]. Our initial observations of Nur1 localization revealed faint Nur1 enrichment at the nuclear envelope, consistent with previous reports [36], [53]. This localization did not noticeably change during cell cycle progression and was not altered as consequence of Clb2 fusion or Cdk phosphosite mutations. Nevertheless, the possibility of Nur1 regulation by localization merits further investigation.
Nur1 interacts with Net1 and, in addition to Net1's role as a Cdc14 inhibitor, one of the phenotypes reported for net1-1 mutants is a high rate of chromosome loss [54]. This indicates that Net1 plays a role in accurate chromosome segregation that might go beyond its role as cell cycle regulator. The relationships between Net1, Nur1, the phosphoregulation of both proteins and their roles in Cdc14 phosphatase regulation and genome stability are an important topic for further studies.
Removing Nur1 (nur1Δ), or its Cdk phosphorylation sites (nur1(9A)), resulted in premature Cdc14 release. Phosphorylated Nur1 therefore fulfills a role in attenuating Cdc14 activation in early anaphase. In a quantitative model in which Cdc14 substrate dephosphorylation timing is determined by the ratio of phosphatase activity versus Cdk kinase activity [2], [4], phospho-Nur1 ensures that Cdc14 activity is kept low and only its earliest targets are dephosphorylated in early anaphase.
While preventing Nur1 phosphorylation causes premature Cdc14 activation, maintaining persistent Nur1 phosphorylation delays Cdc14 release and mitotic exit. Nur1 dephosphorylation by Cdc14 in early anaphase thus engages a positive feedback loop in which phospho-Nur1 as a Cdc14 inhibitor is removed by the very action of Cdc14 itself (Fig. 7). Whether Nur1 dephosphorylation actually stimulates release of Cdc14, or just stops Nur1 from counteracting it, is still an open question. In either case, failure to turn off phospho-Nur1 results in a mitotic exit delay, compromised rDNA segregation and inability to survive at higher temperature. Thus, Nur1 dephosphorylation plays an important role in shaping the Cdc14 activation pattern in early anaphase, until the MEN takes over to sustain Cdc14 release. Additional players might contribute to Cdc14 feedback regulation, including Tof2 [50] or Spo12 [49].
Although we have found that Nur1 interacts with Net1 and Cdc14, the molecular mechanism by which Nur1 counteracts Cdc14 release is not yet understood. The fact that Nur1 no longer influences Cdc14 activation in a net1-6Cdk strain suggest that Nur1 acts at the level of Cdk-dependent Net1 phosphorylation. A possible mechanism that is consistent with this observation takes into consideration that Net1 itself is a Cdc14 target in vivo and in vitro [12], [25]. Cdc14 could prevent its own release if it were allowed access to Cdk phosphorylation sites on Net1. A scenario can be envisaged in which phospho-Nur1 promotes dephosphorylation of Net1 by Cdc14. The required change in accessibility of Net1 phosphorylation sites could be accomplished through conformational changes within a Net1/Nur1/Cdc14 protein complex, depending on the Nur1 phosphorylation state.
In summary, we describe a positive feedback loop, in which Nur1 attenuates Cdc14 release until Nur1 itself is dephosphorylated by Cdc14. This feedback imposes a requirement for the FEAR network to initiate Cdc14 release. Once Cdc14 becomes active, Nur1 dephosphorylation helps to sustain Cdc14 release while Cdk activity declines, until the MEN pathway takes over. In this model, Nur1 can be seen as a bridge between the FEAR network and the MEN signaling cascade.
All strains were of the W303 background and are listed in S1 Table. A strain containing ADE2 integrated within the rDNA repeats was a kind gift from M. Kaeberlein [41]. Epitope tagging of endogenous genes and gene deletions were performed by gene targeting using polymerase chain reaction (PCR) products [55], [56]. The nur1(9A) mutant was engineered by endogenous gene replacement using an integrative plasmid, based on a synthetic DNA construct (GeneArt, Life Technologies). The conditional Nur1-Clb2 and Nur1-Clb2ΔCdk fusions were created as described [34]. In brief, Clb2 lacking its destruction and KEN-boxes [57], as well as its nuclear localization sequence [58], was fused to Nur1, separated by an unstructured 10-mer GGSGTGGSGT linker. In addition, the Clb2ΔCdk mutant contained further 3 point mutations that prevent it from interacting with the Cdc28 kinase subunit, as described [59]. Strains harboring the conditional Clb2-fusion cassettes were grown on medium lacking uracil to maintain the selectable marker and to prevent spontaneous recombination. Marker loop-out was then induced by β-estradiol-dependent activation of Cre recombinase fused to an estradiol binding domain (Cre-EBD78; [60]), by addition of 1 µM β-estradiol to the growth medium.
Yeast cultures were grown in rich YP medium supplemented with 2% glucose [61]. Cell synchronization using α-factor was as described [62]. For cell synchronization at higher temperatures, using the cdc14-1 or NUR1-CLB2 backgrounds, cultures were shifted to the higher temperatures at the time of release from α-factor arrest.
Protein extracts for Western blotting were prepared following cell fixation using trichloroacetic acid, as described [63], and analyzed by SDS-polyacrylamide gel electrophoresis (SDS-PAGE). Antibodies used for Western detection were, α-Clb2 (Santa Cruz, sc9071), α-Orc6 (clone SB49); α-Sic1 (Santa Cruz, sc50441), α-Tub1 (clone YOL1/34, AbD Serotec), α-HA (clone 12CA5), α-myc (clone 9E10), α-Pk (clone SV5-Pk1, AbD Serotec). Phos-tag was purchased from Wako Chemicals and added to SDS-polyacrylamide gels along with MnCl2 according to the manufacturer's instructions.
For immunoprecipitation, cell extracts were prepared in EBXG buffer (50 mM HEPES pH 8.0, 100 mM KCl, 2.5 mM MgCl2, 10% glycerol, 0.25% Triton X-100, 1 mM DTT, protease inhibitors) using glass bead breakage in a Multi Bead Shocker (Yasui Kikai). Extracts were precleared, incubated with antibody and finally adsorbed to Protein A Dynabeads. Beads were washed and elution was carried out in SDS-PAGE loading buffer. For the in vitro Nur1 dephosphorylation assay, immunoprecipitation was performed as above, then beads were resuspended in phosphatase buffer and 1 µg λ phosphatase (New England Biolabs), or 8 µg purified recombinant Cdc14 [4], were added, followed by incubation at 30°C for 30 minutes before the reaction was stopped and proteins eluted by addition of SDS-PAGE loading buffer.
Indirect immunofluorescence was performed on formaldehyde-fixed cells using the following antibodies, α-GFP, (clone TP401, Torrey Pines Biolabs or ab6556, Abcam), α-Tub1 (clone YOL1/34, AbD Serotec) and FITC and Cy3-dye labeled secondary antibodies (Sigma and Chemicon, respectively). Cells were counterstained with the DNA binding dye 4',6-diamidino-2-phenylindole (DAPI). Fluorescent images were acquired using an Axioplan 2 imaging microscope (Zeiss) equipped with a 100x (NA = 1.45) Plan-Neofluar objective and an ORCA-ER camera (Hamamatsu). Spindle length measurements were carried out in ImageJ.
|
10.1371/journal.pntd.0007047 | Dengue and chikungunya among outpatients with acute undifferentiated fever in Kinshasa, Democratic Republic of Congo: A cross-sectional study | Pathogens causing acute fever, with the exception of malaria, remain largely unidentified in sub-Saharan Africa, given the local unavailability of diagnostic tests and the broad differential diagnosis.
We conducted a cross-sectional study including outpatient acute undifferentiated fever in both children and adults, between November 2015 and June 2016 in Kinshasa, Democratic Republic of Congo. Serological and molecular diagnostic tests for selected arboviral infections were performed on blood, including PCR, NS1-RDT, ELISA and IFA for acute, and ELISA and IFA for past infections.
Investigation among 342 patients, aged 2 to 68 years (mean age of 21 years), with acute undifferentiated fever (having no clear focus of infection) revealed 19 (8.1%) acute dengue–caused by DENV-1 and/or DENV-2 –and 2 (0.9%) acute chikungunya infections. Furthermore, 30.2% and 26.4% of participants had been infected in the past with dengue and chikungunya, respectively. We found no evidence of acute Zika nor yellow fever virus infections. 45.3% of patients tested positive on malaria Rapid Diagnostic Test, 87.7% received antimalarial treatment and 64.3% received antibacterial treatment.
Chikungunya outbreaks have been reported in the study area in the past, so the high seroprevalence is not surprising. However, scarce evidence exists on dengue transmission in Kinshasa and based on our data, circulation is more important than previously reported. Furthermore, our study shows that the prescription of antibiotics, both antibacterial and antimalarial drugs, is rampant. Studies like this one, elucidating the causes of acute fever, may lead to a more considerate and rigorous use of antibiotics. This will not only stem the ever-increasing problem of antimicrobial resistance, but will–ultimately and hopefully–improve the clinical care of outpatients in low-resource settings.
ClinicalTrials.gov NCT02656862.
| Malaria remains one of the most important causes of fever in sub-Saharan Africa. However, its share is declining, since the diagnosis and treatment of malaria have improved significantly over the years. Hence leading to an increase in the number of patients presenting with non-malarial fever. Often, obvious clinical signs and symptoms like cough or diarrhea are absent, probing the question: “What causes the fever?” Previous studies have shown that the burden of arboviral infections–like dengue and chikungunya–in sub-Saharan Africa is underestimated, which is why we screened for four common arboviral infections in patients presenting with ‘undifferentiated fever’ at an outpatient clinic in suburban Kinshasa, Democratic Republic of Congo. Among the patients tested, we found that one in ten presented with an acute arboviral infection and that almost one in three patients had been infected in the past. These findings suggest that clinicians should think about arboviral infections more often, thereby refraining from the prescription of antibiotics, a practice increasingly problematic given the global rise of antimicrobial resistance.
| Acute fever is one of the main reasons for healthcare seeking worldwide. In tropical settings, and especially sub-Saharan Africa, malaria is the first cause to be ruled out, which is done increasingly so following the World Health Organization’s testing before treating policy of 2010 –through microscopic blood slide examination or a rapid diagnostic test (RDT) [1]. Following the introduction of this policy, together with the roll-out of the highly efficacious artemisinin-combination therapy as first-line malaria treatment and efficacious vector control, the overall malaria burden declined over the last decade [1]. Accordingly, clinicians face a relatively higher number of malaria-negative patients for whom they do not have a clear diagnosis [2–4]. In sub-Saharan Africa, where healthcare settings are often resource-limited, healthcare providers face the daunting challenge pinpointing the causing agent of acute fever in an adequate and timely fashion, with little to no diagnostic means other than a malaria-RDT. They mostly rely on history taking and physical examination to determine the focus and cause of infection, of which acute respiratory infection (ARI), gastroenteritis (GE) and urinary tract infection (UTI) are the three most prevalent syndromes reported [4]. However, for some patients presenting with acute fever no focus of infection can be found, thus labeling them as ‘undifferentiated’–knowing that their differential diagnosis is broad, ranging from viral, bacterial, parasitic to fungal infections. Some studies have found that this ‘undifferentiated’ group among fever patients represents 20 to 40% of the grand total [4]. Although viral illnesses are often suspected, both prescription and over-the-counter usage of antimicrobials is rampant in this group globally and their licentious usage in low-resource settings fuels the global burden of antimicrobial resistance [5,6]. More insight in the exact causes of this group of ‘undifferentiated fevers’ may help curb the usage of antimicrobials and improve the clinical care of patients in low-resource settings more broadly [7]. Still, evidence on the causes of ‘undifferentiated acute fever syndromes’ is scarce and is coming almost entirely from inpatient settings. Indeed, to our knowledge only one study in sub-Saharan Africa, namely in Sierra Leone in 2012–2013, looked in a prospective way at the etiologies of acute fever–using RDTs–in patients with self-reported or clinically confirmed fever with a maximum duration of 7 days, finding 5% acute dengue virus (DENV) infection and 39% acute chikungunya virus (CHIKV) or other alphavirus infection [8]. Nonetheless, limited outbreaks and sporadic clinical cases of DENV have been reported over the last 50 years in 22 African countries [9]. Seroprevalence studies have demonstrated DENV IgG-antibodies, indicating past-infection, in 12.5% of study participants in Cameroon, 36% in Burkina Faso and 45% in Nigeria [9], although in other areas seropositivity remained zero [10]. In Tanzania past infection rates are higher, reaching 50.6% in health-facility based studies and 11% in community-based studies [11]. Despite the presence of all four DENV serotypes, severe disease epidemics are rarely reported in Africa [12]. The DENV burden in Africa is, based on modeling, estimated at 16 million symptomatic clinical infections or 16% of the global total [13]. In East-Africa, CHIKV outbreaks and circulation are described, such as in Kenya with a past-infection rate of 67% [14] and the reports of epidemics in 2004 in Kenya [15], in 2013 in Tanzania [16] and in 2018 in Mozambique [17].
These viral vector-borne diseases are also circulating in the Central African region. This is illustrated by CHIKV outbreaks in Kinshasa in both 2000 [18] and 2012 [19] and Brazzaville in 2011 [20], and the 2013 DENV [21] and 2016 yellow fever virus (YFV) outbreak in Angola [22], along with the first Zika virus (ZIKV) case reported in Angola in 2017 [23]. These outbreaks are only possible because the Aedes mosquito, the vector of the aforementioned arboviruses, thrives in this region. Furthermore, although not conclusive for vector competence and local transmission capability, alphaviruses (chikungunya) and flaviviruses (species not specified) were demonstrated by RT-PCR in Aedes mosquitoes in Kinshasa in 2014 [24].
In the Democratic Republic of Congo (DRC) the circulating pathogens causing uncomplicated acute undifferentiated fever, are unknown [25]. However, outside the above documented epidemics, CHIKV and DENV probably circulate continuously. Of travelers returning from Africa (2007–2012) and attending the outpatient clinic of the Institute of Tropical Medicine in Antwerp, Belgium, 22% of those diagnosed with a CHIKV infection came from DRC [26] and up to that point there was an increasing number of confirmed DENV infections in travelers coming from a large set of African countries, including DRC [27]. In Eastern DRC, a few DENV cases have been found during an outbreak of West-Nile fever in 1998 [28] and between 2003 and 2012 when testing samples negative for YFV [29].
In this study in DRC, we aim to quantify the importance of four major arboviruses as a cause of acute undifferentiated fever. Furthermore, we aim to describe the case presentation and the presence of arbovirus/malaria co-infections, since Plasmodium falciparum is still responsible for an estimated 25 million cases nationwide–with 97% of the country being a ‘high transmission’ region, ranking DRC among the top 3 countries in sub-Saharan Africa with the highest malaria burden [30].
This study was approved by the ethical review boards of the School of Public Health of the University of Kinshasa (DRC), the Institute of Tropical Medicine of Antwerp (Belgium) and the University Hospital of Antwerp (Belgium). The study was registered in a public repository (https://www.clinicaltrials.gov/ct2/show/NCT02656862). Written informed consent was obtained from every adult or–in case of minors–from their caretaker. This study was conducted in full compliance with the principles of the latest amended Declaration of Helsinki and of the International Conference Harmonization (ICH) guidelines, plus adhering to local laws and regulations.
The study took place in Lisungi health center in Pumbu, an area of about 14,000 inhabitants, belonging to the peri-urban health district Mont Ngafula 1, at the southern side of Kinshasa. The climate is tropical with a rainy season between October and May, and a dry season from June to September. The Lisungi health center is the only public health facility in the area, with a medical staff of 40 persons averaging 250 patient encounters per week–which are provided for a small out-of-pocket contribution, as is commonplace throughout DRC. It treats mainly outpatients, but several inpatient beds are available for short time follow-up of more complicated cases. Over the years, around 70% of patients mention fever as the reason for medical care seeking, of whom half tested positive for malaria on RDT (personal communication with Dr. Blaise Fungula). There are no means for other microbiological testing. The Lisungi health center has recently performed a Good Clinical Laboratory Practice compliant trial and has been involved in other febrile illness investigations, specifically on malaria [31].
The study was designed as a cross-sectional study with prospective patient inclusion. As the proportion of pathogens can change over time, especially for epidemic-prone diseases, we included patients proportionally from November 2015 to June 2016. Only patients of at least 2 years old, presenting at the outpatient department with a history of acute fever (i.e. ≥ 2 days and ≤ 7 days) or having an axillary temperature of ≥ 37.5°C, were eligible. Patients with any history of an acute injury, trauma or poisoning, suspicion of meningitis/encephalitis, recent hospitalization or women who gave birth in the preceding two weeks, were excluded. Reported recent intake of antimicrobials was not an exclusion criterion, but was recorded accordingly. There were two categories of patients included: the ‘undifferentiated fevers’, with as case definition history of acute fever and without any clear clinical focus of infection and the ‘differentiated fevers’, with a history of fever and with acute respiratory infection (ARI), gastroenteritis (GE) or urinary tract infection (UTI) categorized on clinical grounds. Of the first group, a maximum of 6 patients per day (3 children and 3 adults), and of the second group a maximum of 2 patients per day (1 child and 1 adult) were to be included. The latter category was included given the non-specificity of the signs and symptoms of a possible arboviral infection and in order to estimate the burden of co-infection–which was also the reason not to exclude the confirmed malaria cases based on laboratory analysis. Our main interest was the distribution of viral pathogens among the undifferentiated fevers. To be able to detect the presence of a disease whose prevalence is 5% with a precision of 2.5% at a confidence level of 95%, 290 patients needed to be included. Increased with 10% for incomplete data or loss of biological samples, we came to a minimal sample size of 320.
Over the period November 2015 to June 2016, 342 patients were included from whom clinical data and on the spot malaria-RDT results were recorded. Clinical diagnoses were as follows: 70.2% undifferentiated fever, 17.0% ARI, 6.1% GE and 6.7% UTI–all three further labeled as ‘differentiated fevers’. The study population (Table 1) consisted of 183 (53.5%) female participants and 180 (53.1%) were under the age of 18. Only 10 (2.9%) reported having a chronic disease such as diabetes or sickle cell anemia and 50 (14.6%) patients were in need of hospitalization. The reported YFV vaccination rate was as low as 1%. The vast majority (77.8%) of patients presenting in the four first days after the onset of fever, came from the commune Mont Ngafula, and a minority from neighbouring Selembao and Ngaliema–acute and past arbovirus infections were detected in patients from these three communes (Table 2). Before going to the health center, 15.5% of the participants already self-medicated with one or more tablets of an anti-malarial drug. At presentation, malaria RDT for Plasmodium falciparum was positive in 155 (45.3%) of the total number of participants (47.9% and 39.2% in undifferentiated and differentiated fever groups, respectively, p = 0.139). However, 300 (87.7%) received an antimalarial treatment, which was significantly more in the undifferentiated (92.1%) than the differentiated fever group (77.5%) (p<0.001). Antibacterial drugs were frequently prescribed: 64.3% of participants received at least one antibacterial drug, significantly more in the differentiated (72.5%) than the undifferentiated group (60.8%) (p = 0.039) and 11.7% received even more than one antibacterial drug (29.4% in the differentiated and 9.6% in the undifferentiated fever group, p<0.001).
In the 235 participants further tested for arboviruses, 19 (8.1%) fulfilled the criteria of an acute DENV infection, of which 14 were confirmed by RT-PCR. Both serotypes DENV1 and DENV2 were detected (Table 3). Five participants were presumptively infected with DENV, based on the presence of IgM antibodies alone. All NS1 positive patients were RT-PCR positive. In contrast, only two acute CHIKV infections were suspected based on the presence of IgM antibodies. The majority of CHIKV IgM ELISA positive samples was not confirmed by IFA. On these non-congruent samples, PCR to detect Plasmodium was performed and revealed an actual malaria infection in 18 of the 22 CHIKV IgM ELISA positive /IgM IFA negative. There was a temporal heterogeneity in the appearance of DENV infections (Fig 1). In June, a dry season month which had less than 25mm of rain (www.infoclimat.fr), there was an increased risk of 6.13 (adjusted OR 95% CI 2.24–17.81) of presenting with acute DENV in comparison to the rainy season (S1 Table). Of the acute DENV and CHIKV cases, 31.6% and 0% had a positive malaria RDT, respectively. When focusing on the malaria negative cases, we observed that 8.7% (13/149) tested positive for acute DENV, 1.3% (2/149) for acute CHIKV, none for acute ZIKV, nor for YFV.
With regard to the clinical presentation of both DENV and CHIKV infections, we found no specific signs or symptoms to be statistically significantly–let alone clinically relevant–associated with acute DENV or CHIKV versus the other febrile patients (S2 Table). None of the acute DENV or CHIKV cases were clinically severe enough to require hospitalization and there was no apparent leucopenia or hemoconcentration (as often seen in severe DENV cases).
We found 71 (30.2%) patients with anti-DENV IgG antibodies, of which 60 (75.0%) contained relatively high levels of anti-DENV antibody (O.D. ≥ 1.9), the latter not depending on age (S1 Fig). The PRNT on the subsample was only positive in 5.6% and 66.7% for DENV, and 25% and 20% for YFV, on samples with IgG ratio below 1.9 and above 1.9,respectively (Table 3). For these past infections, all 4 serotypes were detected: DENV1, DENV2, DENV3 and DENV4 in 4, 2, 1 and 1 patient, respectively (of the 6 positively tested DENV PRNT)– 2 patients were reactive to all 4 serotypes, the other 4 only to 1 serotype. Past exposure to CHIKV was suspected with IFA IgG in 26.4% of the study participants. When taking the DENV and CHIKV positive IgG samples together, 56.6% of the study participants were suspected having been exposed to at least one arbovirus. The prevalence of past DENV and CHIKV infections increased with age, raising from 18.9% and 5.4% under 5 years of age, to 80% and 40% over 65 years of age, respectively (Fig 2). The association of age is statistically significant for the past infections with CHIKV (p = 0.01) and DENV (p<0.01) (S2 Table). Having been exposed to DENV was also statistically significant associated with recent travel (p = 0.01). The congruence between the RDT and PCR/ELISA results for DENV was variable: in comparison to PCR the sensitivity of NS1 was 90%, in comparison to IgM ELISA the IgM RDT had a sensitivity of 30% and in comparison to IgG ELISA the IgG RDT had a sensitivity of 7.6%. The specificities were all above 99.3%.
Although no large epidemics were reported recently, our study showed ongoing transmission of arboviruses in Kinshasa, DRC. Acute DENV, caused by DENV1 and DENV2, and CHIKV infection was demonstrated in 8.1% and 0.9% of the patients attending a first line health center with acute undifferentiated fever, respectively. Importantly, neither DENV nor CHIKV was clinically suspected, nor considered in the clinical differential diagnosis and 64.3% of patients were treated with at least one antibacterial drug of whom almost one in eight (11.7%) received dual or triple antimicrobial therapy. A possible explanation of the apparent absence of clinical and/or severe acute DENV cases in our study, and in other African settings too, is that African heritage is described to genetically protect against severe DENV. More specifically, the lower OSBPL10 expression profile in Africans is protective against viral hemorrhagic fever and dengue shock syndrome [36,37].
However, diagnostic testing for arboviruses has several shortcomings. Hereafter, we will highlight the limitations of the tests used for acute and past infections. Five out of 19 acute DENV and 2 out of 2 acute CHIKV infections were diagnosed based on the presence of IgM antibodies only and were therefore only presumptive infections. This could have led to an overestimation of the number of acute infections. In addition, IgM antibodies can be present for several months and it is therefore possible that DENV/CHIKV was not the cause of the fever at the moment of presentation. Nonetheless, the presence of IgM antibodies suggests that DENV/CHIKV was recently circulating in the area. On the other hand, as we did not have repeated measurements or convalescent samples to demonstrate seroconversion to IgG or a four-fold increase in IgG titer, we may have missed some acute secondary dengue cases, which may have undetectable IgM antibody levels [38].
The number of positive CHIKV IgM ELISA test results was unexpectedly high. As the results could not be confirmed with IFA, a technique which is considered to be more specific, interference with malaria was suspected based on our experience. Indeed, false-positive reactions as a result of polyclonal B-cell activation is a phenomenon that we experienced before with ZIKV ELISA [39]. The detection of Plasmodium by PCR in the CHIKV ELISA positive/IFA negative samples strongly supported this hypothesis. The use of IFA for IgM detection may result in false-positive reactions although the risk is limited in case of experienced readers [40]. Cross-reaction of anti-CHIKV antibodies with antibodies against other members of the Semliki-forest serogroup, notably O’nyong nyongvirus, could not be excluded as no neutralization assays were performed. ZIKV was not suspected to be circulating in the area and indeed, no RT-PCR positive cases were found. YFV was actively circulating in the region at the time of study, transgressing the border with Angola [22], but we did not detect any RT-PCR positive case. Detecting YFV and ZIKV only based on RT-PCR diagnostics could have resulted in an underestimation of acute ZIKV and YFV infections. However, ZIKV and YFV IgM testing was not done, because almost 80% of the patients included in this study presented in the first four days after the onset of fever, which is the period with the highest probability to detect the virus with molecular methods and there are several shortcomings with the IgM testing for these viruses [41,42]. Nowadays, there is evidence that in urine samples ZIKV is longer detectable by ZIKV PCR, but at the time of study this was not known. Thus, no such samples were collected [43].
We demonstrated past exposure to arboviruses too, with 30.2% of the participants having detectable IgG against DENV, which is on the higher end of the spectrum compared to other studies in Africa reporting an overall flavivirus seroprevalence ranging between 0 and 35% with a mean of 18.1% [9,25]. The IgG-seroprevalence increased with age, thus suggesting a continuous exposure to flaviviruses over time. The 26.4% past CHIKV infection rate was in line with the estimated seroprevalence of 34.4% in Congo Brazzaville before the outbreak of 2011 [20] and was on par with other African sites reporting an overall alphavirus seroprevalence oscillating between 0 and 72% [25]. Remarkably, CHIKV IgG was also detected in small children, born after the 2011 epidemic, pointing towards an endemic circulation of the virus. DENV IgG testing was done with an ELISA test. Although it is widely known that there is cross-reactivity among flaviviruses, ELISA is still the most affordable–hence most commonly used–test [25]. Since YFV vaccination is an expected cause of cross-reaction with DENV IgG, we performed PRNT for DENV and YFV in a subset of samples, and found that the majority of samples negative with PRNT for DENV, were also negative with PRNT for YFV, suggesting that the high flavivirus IgG positivity is not likely to be the result of YFV vaccination. Due to operational reasons and the scope of the study, we did not further test for other flaviviruses and hence, we cannot rule out that the DENV IgG positivity in our study is due to other flaviviruses exposure [44]. We noted that the congruence between DENV ELISA and PRNT was lower than in American and Asian settings [45–49], but similar to the observation in other studies in DRC [50] and in Ethiopia [51]. The report of YFV vaccination was very low (below 1%), but as YFV vaccination is included in the childhood vaccination program over the last decade, this may indicate that the population is not aware of which vaccines their children get. However, the increasing prevalence of flavivirus IgG antibodies with age is congruent with a history of increasing exposure to the pathogens over lifetime and thus could not be explained by YFV vaccination. The CHIKV IgG testing was done with two tests: screening with ELISA and confirmation with IFA. We cannot exclude that positive results are due to cross-reactivity with other known (e.g. O’nyong nyong virus, Semliki Forest virus or Sindbis virus) or unknown togaviruses.
In a recent study conducted at the Lisungi health center, it was reported that 62% of patients–both children and adults–with acute fever had neither malaria nor bacteremia [52]. For the first time we were able to demonstrate the fact that arboviruses, more specifically DENV and CHIKV, circulate in the capital of DRC. The highest number of acute cases was reported in June (a dry season month), but cases were also confirmed in the other months, indicating that despite an epidemic profile, transmission persists over the rainy season in Kinshasa. This finding is consistent with observations over the past decades in Asia [53] and Latin-America [54] and adds to mounting–although still scarce–evidence that arboviruses are endemic in large parts of sub-Saharan Africa [9,17].
Furthermore, we were able to document the common practice of over-prescription of antimicrobials, including antimalarial drugs, in malaria RDT-negative patients, as is apparently the case nationwide in DRC as recently shown by Ntamabyaliro et al [55]. Indeed, while not even half of the patients (45.3%) tested positive for malaria–a figure just below average national RDT positivity rates [30], close to 90% received antimalarial treatment, in addition to 15.5% of patients treated with over-the-counter antimalarials prior to presentation at the clinic. It could be questioned whether the rigorous implementation and usage of RDTs has any added benefit. A recent meta-analysis, including studies from Afghanistan, Cameroon, Ghana, Nigeria, Tanzania, and Uganda, evaluated data from over half a million children and adults and showed that the introduction of a malaria RDT simply shifted the antimicrobial overuse from one antimicrobial class to the other, mainly from antimalarial to antibacterial and anthelmintic drugs [56]. Consequently, the increasing prescription rate of antimicrobials–including antibacterial, anthelmintic and antimalarial drugs, is extremely worrisome in terms of the growing global problem of antimicrobial resistance, including against Plasmodium falciparum [57].
Although the sampling design of this study was adequate to evaluate the proportion of arboviruses causing acute undifferentiated fever, sample size was small and patients were only recruited from a single health center. However, the Lisungi health center is well visited by the surrounding population and all ages were represented in our study population. Moreover, the median age of our study population is 17 years, which approximates the median age of 18.6 years in the DRC (UNDESA 2017 and CIA World Factbook 2017). Another limitation in our study was the impossibility due to operational reasons to include participants over an entire year. The study was halted at the end of June, which was apparently the month with the highest number of infections. We did not investigate bacterial causes of fever through culture of blood or other bodily fluids, tests typically done in hospital settings, hereby possibly underestimating the burden of concomitant bacterial (super)infection. We therefore encourage further research elucidating the broad range of pathogens causing acute undifferentiated fever and the distribution of the insect vectors involved in arboviral transmission in urban and rural sub-Saharan African settings. Based on our findings, we recommend to include arboviral infections, namely DENV and CHIKV, in the differential diagnoses of acute fever presentation in Kinshasa.
In conclusion, we state that among undifferentiated acute fever cases in a peri-urban health center of Kinshasa, dengue–both DENV-1 and DENV-2 –and chikungunya infections were demonstrated, but no acute cases of Zika or yellow fever were detected. Apart from these acute infections, we showed that about one third of participants showed evidence of past arboviral exposure, as evidenced by positive IgG antibodies titers.
|
10.1371/journal.pntd.0001772 | An Analytical Method for Assessing Stage-Specific Drug Activity in Plasmodium vivax Malaria: Implications for Ex Vivo Drug Susceptibility Testing | The emergence of highly chloroquine (CQ) resistant P. vivax in Southeast Asia has created an urgent need for an improved understanding of the mechanisms of drug resistance in these parasites, the development of robust tools for defining the spread of resistance, and the discovery of new antimalarial agents. The ex vivo Schizont Maturation Test (SMT), originally developed for the study of P. falciparum, has been modified for P. vivax. We retrospectively analysed the results from 760 parasite isolates assessed by the modified SMT to investigate the relationship between parasite growth dynamics and parasite susceptibility to antimalarial drugs. Previous observations of the stage-specific activity of CQ against P. vivax were confirmed, and shown to have profound consequences for interpretation of the assay. Using a nonlinear model we show increased duration of the assay and a higher proportion of ring stages in the initial blood sample were associated with decreased effective concentration (EC50) values of CQ, and identify a threshold where these associations no longer hold. Thus, starting composition of parasites in the SMT and duration of the assay can have a profound effect on the calculated EC50 for CQ. Our findings indicate that EC50 values from assays with a duration less than 34 hours do not truly reflect the sensitivity of the parasite to CQ, nor an assay where the proportion of ring stage parasites at the start of the assay does not exceed 66%. Application of this threshold modelling approach suggests that similar issues may occur for susceptibility testing of amodiaquine and mefloquine. The statistical methodology which has been developed also provides a novel means of detecting stage-specific drug activity for new antimalarials.
| The schizont maturation test (SMT) was developed to monitor drug resistance in malaria parasites. The SMT examines differences in the rate of parasite development when exposed to different drug concentrations, providing an estimate of drug efficacy. While the assay is effective when examining resistance in Plasmodium falciparum, there are concerns regarding its suitability for testing other malaria species, particularly if the drug only targets particular life-cycle stages of the parasite. Blood samples taken from Plasmodium vivax infected individuals exhibit significant heterogeneity in the parasite life-cycle stages present. If a drug targets the early stage parasites, but only late stage parasites are present in the sample, the test will show an erroneously high degree of resistance. In this study, we estimate thresholds which can be used to identify when test results can be considered accurate should the drug being tested only affect specific life stages of the parasites. Chloroquine is used as a case study but the method developed also allows the identification of stage-specific activity in other malarial drugs in P. vivax. For field researchers, this threshold modelling approach will allow for increased confidence in the reliability of P. vivax resistance results and provides a novel means of detecting stage-specific drug activity for new antimalarials.
| Malaria continues to pose a significant threat to human health globally. Currently, as many as 2.6 billion people are at risk of P. vivax infection, with an estimated 72–390 million cases per year [1], [2]. Historically, malaria research has focussed on P. falciparum, due to its reputation as the most lethal human malaria parasite. Although P. vivax is less pathogenic than P. falciparum, it still poses a serious public health burden and is being increasingly recognised as a cause of severe and fatal disease, particularly in children and pregnant women [2], [3].
A number of recent publications have highlighted the increasing recognition of the clinical importance of P. vivax and the renewed emphasis placed on research into this species [2], [4], [5], [6]. Further, the emergence of highly chloroquine (CQ) resistant P. vivax in Southeast Asia (CQ remains a front-line treatment for vivax malaria as it is affordable, well tolerated and safe, and its long half-life ensures protection from early relapses [7], [8]) has created an urgent need for an improved understanding of the mechanisms of drug resistance in these parasites, the development of robust tools for defining the spread of resistance, and the discovery of new antimalarial agents. To glean insights into the development of resistance in P. vivax, a modified form of the ex vivo Schizont Maturation Test (SMT) has been developed and applied to fresh isolates directly from patients [9], [10], [11], [12].
The central tenant of the SMT is that drug activity on susceptible parasites will completely stop or slow the growth in a dose-dependent manner, with reduced susceptibility being manifest by an ability of parasites to mature to schizont stages in the presence of higher concentrations of drug. The standard assay is conducted for 30 hours, the time required for parasites to reach maturation without drug, and the proportion of schizonts at the conclusion of the assay is used as an indicator of parasite maturation.
The SMT was initially developed for testing drug susceptibility in P. falciparum [13], [14], where almost all parasites in the peripheral circulation are at the immature ring stage. However, in infections due to non-falciparum species, trophozoite and schizont stages are commonly present in the peripheral circulation. To accommodate this, a modified SMT has been developed in which the control wells are monitored until the number of schizonts exceeds 40% of parasites prior to harvest (i.e., assays are conducted for variable lengths of time) [15].
For the results of the SMT to be valid, a sample, irrespective of the drug resistance phenotype of the parasites, must have had sufficient exposure to the drug to affect a response. The diversity in parasite life cycle stages in P. vivax infections creates a significant confounding factor, particularly since there is an apparent marked variation of drug susceptibility in different erythrocytic life cycle stages. In previous work, it has been demonstrated that the trophozoite stages of P. vivax are almost completely resistant to CQ, and continue to mature no matter how high the concentration of drug [15], [16]. It follows that, if a drug only acts on ring stage parasites, but the sample contains a majority of trophozoites and schizonts, the parasite is likely to be erroneously categorised as resistant simply because there were no susceptible life cycle stages present in the assay.
In this study we develop a statistical methodology to identify stage specific drug activity in the SMT. We use CQ against P. vivax as a case study and demonstrate that stage specific drug activity has profound consequences for the interpretation of SMT results. We also examine stage-specific drug effects for other commonly used drugs and simulate the growth dynamics of P. vivax parasites within the SMT to provide recommendations on how to improve the reliability of SMT results.
Ethical approval for the collection of blood samples for drug susceptibility testing was obtained from the ethics committees of the National Institute of Health Research and Development, Ministry of Health (Jakarta, Indonesia), and the Human Research Ethics Committee of NT Department of Health & Families and Menzies School of Health Research (Darwin, Australia). Written informed consent was obtained from adult patients and parents and/or guardians of enrolled children.
Blood samples were collected from patients attending outpatient clinics in Timika, Papua Province, Indonesia, as previously described [15]. Only patients infected with a single species of Plasmodia were included in the study; the majority of samples contained P. vivax or P. falciparum, although a limited number of P. malariae and P. ovale isolates were also available [17].
SMTs were conducted on these samples following World Health Organisation guidelines for drug susceptibility testing [18], with modifications developed to aid the application of the test to P. vivax [15]. Venous blood (5 mL) was collected by venipuncture, and after removal of host white blood cells using a CF11 column, 800 µL of packed infected red blood cells (IRBC) were used for the SMT. Assays conducted for up to 7 drugs: CQ, artesunate, amodiaquine, lumefantrine, mefloquine, piperaquine and pyronaridine [15].
The proportion of parasite life cycle stages (i.e., rings, trophozoites and schizonts, as defined by Russell et al [12], [15]) in each isolate was assessed at 0 hours, 24 hours, and then at non-uniform times until 40% of the parasites in the control well reached mature schizonts. At this point, the assay was terminated, and wells under serial drug concentrations were harvested. The proportion of parasites in each stage of the erythrocytic cycle at time of harvest was reported for each drug concentration, as was the duration of the assay.
The estimated drug response (R) of each isolate was derived from the ratio between the proportion of schizonts at harvest in the treatment well compared to that in the control well. Only data satisfying the following criteria were included in the dose response modelling:
The sigmoid Emax dose-response curve (equation 1) was fitted to all available data simultaneously, with mixed-effects modelling used for CQ data.
Rij represents the drug mediated growth inhibition response (the ratio between the proportion of schizonts at harvest in the treatment well compared to the control well) for the ith isolate at the jth concentration, cij represents the drug concentration. Emax and E0 represent the maxima and minima of the dose response curve and γ the slope of this curve. The EC50 value represents the effective concentration at which 50% of the parasite population exhibit a response to the drug. Inter-isolate variability was included for Emax, EC50 and γ. The drug plate batch was also incorporated as a random effect, to control for batch-to-batch variability. Predicted values for EC50 for each isolate were calculated from Empirical Bayes estimates.
EC50 values for parasite samples collected from April 2004 and May 2007 have previously been reported (Russell 2008). However the methodology used to calculate EC50 for CQ differs between the previous report and the current study due to the use of mixed-effects modelling.
For each of the seven drugs tested, linear regression models were used to characterise the relationship between the EC50 values derived for each isolate and the following independent variables: assay duration, the proportion of rings at 0 hours, and delay between venepuncture and assay. The EC50 values were not normally distributed, therefore ln(EC50) was used as the dependent variable.
Where a relationship between one of the independent variables and ln(EC50) was found a non-linear threshold model was constructed to determine the threshold at which the association ceased to exist (Equation 2).
a represents the rate of decline in EC50 when the duration of the assay is less than the threshold, b the mean EC50 values for samples where the duration exceeds the threshold, and c the threshold. A threshold model was preferred over other non-linear models due to the greater interpretability of parameters.
A threshold model of the same form was also fitted to define the relationship between the proportion of rings at time 0 and EC50. The threshold models were fitted to the data using nonlinear regression.
A simulation approach was used to estimate the duration of each stage of the erythrocytic life cycle in both P. vivax and P. falciparum. Estimates were also made of stage durations using the limited data available for P. malariae and P. ovale to validate the methodology. Poisson distributions were selected to represent the duration of each stage, as random deviates produced by sampling the distribution are always positive, and the parameters of the distributions allow for biologically meaningful interpretation. We assumed the duration of each stage could be drawn from a probability distribution, and estimate the mean of each Poisson distribution (λ).
One hundred simulations of the growth dynamics within a culture well, each containing 200 parasites, were conducted. It was assumed each parasite began life as a ring, transitioned to a trophozoite, then to a schizont. The length of time spent in each stage was determined by sampling from the relevant probability distribution. At selected time points, the proportion of parasites in each stage was calculated. As parasites completed the schizont stage, they were assumed to die, and subsequently, proportions were calculated based only on the remaining surviving parasites.
A systematic search of the parameter space for the mean duration in each life cycle stage (λ) was conducted to determine the optimal values for each Poisson distribution. The mean model fit from three simulation experiments (totalling 300 simulations) for each combination of parameters was used and the search examined potential parameters in increments of 0.5 hours.
The optimal fit was determined by minimising the sums-of-squares between the simulation results and data on proportion of parasite at each life cycle stage from the control wells, for a subset of the original field samples. The subset of field samples used had 1) 100% rings at 0 hours, 2) data for at least 3 time points, and 3) an assay duration >42 hours. This subset was used to ensure only samples with young ring stage parasites were included in the fitting process. A penalty equivalent to a difference of 20 between data and simulation results was applied for any later time points where the death of the entire simulated parasite population meant that no direct comparison could be made to the data.
All statistical analyses and simulations were conducted using the R statistical computing software package [19].
SMT results for parasites sourced from 784 patients with single-species infections of either P. vivax (n = 345) or P. falciparum (n = 439) were analysed; 289 (84%) P. vivax and 331 (75%) P. falciparum isolates met the inclusion criteria for statistical analysis. Among these, 141 P. vivax (49%) and 216 P. falciparum (65%) isolates reached the 40% schizont threshold at harvest. The time between venepuncture and start of the assay (the ‘delay’) was significantly correlated with the duration of the assay for isolates of P. falciparum (correlation co-efficient (r) = −0.211, p<0.001), but not P. vivax (r = −0.065, p = 0.226).
There was a significantly higher mean proportion of ring stage parasites in samples at the start of the assay (0 hours) for P. falciparum (0.922) compared to P. vivax (0.588) (Wilcoxon rank sum test, p<0.001). 87.5% (189/216) of P. falciparum isolates contained 100% ring stage parasites at 0 hours, compared to only 2.8% (4/141) of P. vivax isolates.
There was a significant negative association between the CQ assay duration (hours) and the ln (EC50) values for P. vivax (r2 = 0.219, p<0.001; Figure 1a). A similar, but weaker relationship was observed for P. falciparum (r2 = 0.097, p<0.001; Figure 1b). Figure 1 suggests some bimodality in the distribution of results, but this is predominantly due to the lack of sampling between ∼32 and 40 hours. The proportion of rings at the start of the assay was also significantly negatively associated with the ln(EC50) values for both P. vivax (r2 = 0.245, p<0.001) and P. falciparum isolates (r2 = 0.206, p<0.001; Figure 2). In both species there was a significant negative correlation between the proportion of rings at the onset of assay and assay duration (r2 = 0.621, p<0.001; r2 = 0.185, p<0.001, for P. vivax and P. falciparum, respectively).
The threshold model for assay duration was successfully fit to the P. vivax data (Figure 3, Table 1). The threshold value, c, represents the point at which the assay duration was no longer significantly associated with estimates of EC50. We interpret this threshold as the point at which assay duration was sufficiently long to guarantee that the target parasite stage/s were present and exposed to the drug. Although there was a relationship between assay duration and EC50 for P. falciparum we were unable to fit a threshold model as fitting procedures did not identify an appropriate non-linear model and associated threshold point.
A threshold model to determine the relationship between the initial proportion of rings and the EC50 was also only possible for P. vivax; we were unable to fit a threshold model to the P. falciparum data. For P. vivax the proportion of rings at the onset of assay was found to have a significant non-linear relationship with EC50 (Figure 4; Table 2). Samples with less than 65% ring stage parasites at time 0 had higher and more variable EC50 values than those isolates with a greater proportion of ring stage parasites.
Of the six other drugs tested, threshold models were successfully fitted to data for amodiaquine and mefloquine in P. vivax. Confidence intervals around the estimated threshold points showed distinct overlap, suggesting no significant differences between the two drugs, or CQ (Table 3).
In the simulations of the growth dynamics under ex vivo assay conditions, the λ parameter for the Poisson distributions represented the duration of each parasite life cycle stage in P. vivax. The mean λ was 19.1 hours (95% CI 11, 29) for rings, 23.1 hours (95% CI 11, 36) for trophozoites, and 2.2 hours (95% CI 0, 7) for the schizonts. The total duration of the P. vivax erythrocytic cycle was estimated as 44.5 hours (95% CI 29, 62). Similar simulations of P. falciparum growth gave mean estimates of the λ parameter of 23.3 hours (95% CI 14, 33) for ring stage parasites, 21.4 hours (95% CI 11, 33) for trophozoites, and 3.7 hours (95% CI 0, 10) for schizonts; estimates produce a total duration of 48.4 hours (95% CI 32, 66). These estimates of an ex vivo life cycle do not include the age of the rings before the start of the assay and the growth time of schizonts after the end of the assay, thus the true life cycle is expected to be longer than reported.
To validate the methodology, the growth dynamics of a small number of SMT results were also simulated for P. malariae (n = 39) and P. ovale (n = 13) isolates. The total duration of the erythrocytic cycle was estimated to be 75.6 hours (95% CI 59, 93) in P. malariae and 53.9 hours (95% CI 34, 79) in P. ovale.
Despite being developed initially for P. falciparum parasites, the Schizont Maturation Test (SMT) has seen considerable use in testing drug susceptibility of P. vivax. However, little consideration has previously been given to whether the application of the SMT to P. vivax is appropriate. In this study, we demonstrate that the stage-specific dynamics of drug activity may impact on the validity of SMT results. If a particular drug targets ring stage parasites, but is tested in an isolate predominantly containing mature stage parasites, the drug will appear to be ineffective, resulting in an overestimate of resistance. Assay duration and the proportion of rings in the initial sample both provide proxy indicators of the likelihood that early stage parasites will be exposed to drugs. Previous experimental work has identified that trophozoite stage P. vivax parasites are insensitive to CQ [15], [16]; our results support this finding. We apply statistical analysis techniques to show that the heterogeneity of erythrocytic life cycle stages present in peripheral blood samples taken from patients infected with P. vivax necessitates additional criteria be applied to the SMT to ensure the validity of the results.
The significant negative relationship between the time of venepuncture and start of the SMT and duration of the SMT for P. falciparum was not unexpected. The onset of fever is usually associated with the start of a new erythrocytic cycle, meaning the sum of the time elapsed between establishment of the SMT and SMT duration will be more indicative of total duration of the erythrocytic cycle than the SMT duration alone. Hence, parasites which have a delay between venepuncture and start of SMT will enter the SMT in a more advanced state, requiring less time to develop to 40% schizonts. This effect is most likely more dominant in P. falciparum samples due to the high synchronicity in parasites when obtained from the patient, compared to P. vivax.
For both P. vivax and P. falciparum isolates, there were significant negative relationships between the duration of the assay and estimated EC50 values (i.e., short duration assays were associated with reduced susceptibility to drugs). There are a number of possible explanations for these results, none of which is mutually exclusive and all of which are specific to the drug/parasite combination:
When stage-specific drug activity is a consideration, as with CQ against P. vivax [15], [16], we hypothesised that samples more advanced in their development, as characterised by short assay duration or a low proportion of ring stage parasites in the initial blood sample, would appear to be less susceptible to a drug because a significant number of parasites present in the assay had developed beyond the target stage. Support for this hypothesis is provided by the threshold modelling. While the linear models we have described are appropriate for the P. falciparum SMTs, examination of the trends in P. vivax suggest a distinctly non-linear pattern for some drugs, including CQ. The threshold modelling indicates that the EC50 for CQ in P. vivax parasites stabilises once the sample has been exposed to the drug for at least 33.7 hours (CI 28.2, 39.3). The simulation modelling of parasite development time suggests that P. vivax parasites spend an average of 25.3 hours as trophozoites and schizonts before termination of the assay. If the SMT is terminated when the control well reaches 40% schizonts, and this occurs after 33.7 hours (mean duration threshold), it follows that the SMT must have exposed sufficient ring stage parasites to the drug, resulting in EC50 reaching its minimum. Combining the results from the threshold and life cycle modelling, an assay lasting 33.7 hours will have exposed 86.5% of the parasites to CQ for at least one hour at ring stage. Using the upper 95% confidence limit for mean assay duration, a more conservative threshold estimate of 39 hours, we can predict that 95% of the parasites were exposed to CQ for at least 2 hours at ring stage. We propose that this threshold represents the duration of the assay which is sufficient to guarantee that a significant proportion of the ring stage P. vivax parasites have been exposed to the drug (i.e. there is no longer any association between duration and EC50). Assay durations shorter than this threshold expose a greater proportion of tolerant mature stage parasites to the drug, and thus the EC50 values derived from these samples will be artificially elevated.
Similar dynamics are apparent when examining the composition of the initial blood samples. The disparity in the degree of developmental stage heterogeneity in the initial samples between P. vivax and P. falciparum is striking. P. falciparum isolates are markedly more synchronous, presumably because the mature stages (trophozoites and schizonts) are sequestered in deep tissues and organs, rather than in the peripheral circulation. Nearly all P. falciparum samples contain only rings, and are, therefore, ideal for SMT. In contrast, P. vivax isolates tend to show significantly greater heterogeneity in the life cycle stages, with trophozoites and schizonts regularly occurring. The presence of these advanced stages at the onset of SMT makes the interpretation of drug susceptibility results more difficult. Our modelling suggests that the initial sample should contain a minimum of 66% ring stage parasites, preferably >90% ring stage parasites (upper 95% confidence interval of the threshold parameter), to ensure the target life cycle stages are sufficiently present in a sample. However, this significantly reduces the number of samples from which drug sensitivity data can be obtained potentially introducing a sampling bias. It should be noted that, by selecting only those samples that achieve 40% schizonts, we also introduce bias against those parasites that do not grow well in culture.
Processes for synchronising parasitemia have also been proposed as a means of decreasing stage heterogeneity of Plasmodium isolates for ex vivo characterisation [20], [21]. Although such methods may have utility and permit testing of some field isolates that would otherwise be excluded from testing, the removal of mature trophozoite parasites inevitably results in a reduction in parasite count, which is itself another important parameter for reliable quantification of parasite growth.
An alternate approach to specifying a definitive threshold is to apply the same types of threshold models which we have developed here using all available data. Such an approach would have three advantages. First, it would allow all the field samples to be used, thus reducing the potential for bias in the SMT samples. Second, it would allow the development of resistance to be monitored over time by looking for changes in the threshold duration and minimum EC50. Third, it can be used to look for stage-specific drug action in current and new antimalarial drugs. Similar patterns and threshold values were found for CQ, amodiaquine and mefloquine, suggesting all of these drugs have their main effect on ring stage P. vivax parasites. Such a relationship was not observed for the other antimalarials investigated (i.e., artesunate, lumefantrine, piperaquine and pyronaridine). Differences between the stage specific activity of each drug and its variation between parasite species may prove highly informative in elucidating the mechanisms of drug action as well as innate and acquired drug resistance. While it is always possible to investigate stage-specific drug activity using carefully planned laboratory experiments, as reported by Russell et al. [15], the methodology presented here can identify stage specificity through far less laborious means, and can use previously collated results.
Our simulation of parasite development in the SMT and subsequent estimate for the duration of each life cycle stage is the first attempt to model parasite development times for P. vivax. It is important to note that the estimates are relative to the parasite development in the restricted conditions of the SMT control well and may not represent the length of the life stages in vivo, or indeed in potential in vitro culture. It should also be expected that the estimated duration of the ring stage underestimates the true duration due to the delay in obtaining the blood sample after parasite rupture and establishing the SMT. More expansive sampling over the first 24 hours of the assay would likely reduce the confidence intervals of the estimated development times.
In summary, a threshold modelling approach was applied to data from a modified SMT to investigate resistance to CQ in P. vivax. We identified patterns which suggest a non-linear relationship between drug susceptibility in the parasite and both the duration of an assay and the proportion of ring stage parasites in the initial sample, which signifies tolerance of late stage parasites to CQ. Consequently, we recommend that P. vivax isolates should contain a minimum of 66% ring stage life cycle stages, and that assay duration should exceed 34 hours to ensure this stage-specific effect does not artificially inflate the reported EC50. More conservative thresholds would require a minimum of 90% ring stage parasites and a minimum assay duration of 40 hours. An alternative approach would be to use the statistical methodology which has been developed. For field researchers, this threshold modelling approach will allow for increased confidence in the reliability of resistance results. This approach also provides a novel means of detecting stage-specific drug activity for new antimalarials, as demonstrated by our analysis of the susceptibility to amodiaquine and mefloquine.
|
10.1371/journal.pbio.1001364 | Topology and Dynamics of the Zebrafish Segmentation Clock Core Circuit | During vertebrate embryogenesis, the rhythmic and sequential segmentation of the body axis is regulated by an oscillating genetic network termed the segmentation clock. We describe a new dynamic model for the core pace-making circuit of the zebrafish segmentation clock based on a systematic biochemical investigation of the network's topology and precise measurements of somitogenesis dynamics in novel genetic mutants. We show that the core pace-making circuit consists of two distinct negative feedback loops, one with Her1 homodimers and the other with Her7:Hes6 heterodimers, operating in parallel. To explain the observed single and double mutant phenotypes of her1, her7, and hes6 mutant embryos in our dynamic model, we postulate that the availability and effective stability of the dimers with DNA binding activity is controlled in a “dimer cloud” that contains all possible dimeric combinations between the three factors. This feature of our model predicts that Hes6 protein levels should oscillate despite constant hes6 mRNA production, which we confirm experimentally using novel Hes6 antibodies. The control of the circuit's dynamics by a population of dimers with and without DNA binding activity is a new principle for the segmentation clock and may be relevant to other biological clocks and transcriptional regulatory networks.
| The segmented pattern of the vertebral column, one of the defining features of the vertebrate body, is established during embryogenesis. The embryo's segments, called somites, form sequentially and rhythmically from head to tail. The periodicity of somite formation is regulated by the segmentation clock, a genetic oscillator that ticks in the posterior-most embryonic tissue: for each tick of the clock, one new bilateral pair of segments is made. The period of the clock appears to determine the number and the length of segments, but what controls this periodicity? In this article, we have investigated the interactions of three transcription factors that form the core of the clock's regulatory circuit, and have measured how the period of segmentation changes when these factors are mutated alone or in combination. We find that these three factors contribute to a “dimer cloud” that contains all possible dimeric combinations; however, only two dimers in this cloud can bind DNA, which allows them to directly regulate the oscillatory gene expression that underpins the periodicity of segment formation. Nevertheless, a mathematical model of the clock's dynamics based on our experimental findings indicates that the non-DNA-binding dimers also influence the stability, and hence the function, of the two DNA-binding dimers controlling the segmentation clock's period. Such involvement of non-DNA-binding dimers is a novel regulatory principle for the segmentation clock, which might also be a general mechanism that operates in other biological clocks.
| Rhythmic phenomena are widespread in biology and the control of their timing is fundamental to many processes. Yet how the dynamics of genetic circuits that control rhythmicity are regulated remains poorly understood. The segmentation clock is an attractive model system to address this question. This gene regulatory network operates in the presomitic mesoderm (PSM) of developing vertebrate embryos and generates transcriptional oscillations that direct the rhythmic and sequential formation of body segments in concert with embryonic elongation [1]–[3]. Many components of the segmentation clock have been identified in the last decade [4],[5], but how they interact to produce oscillations remains unclear. The oscillations of the segmentation clock are most easily observed at the tissue level [6], but they arise on the level of single cells [7].
Current models for the origin of single cell oscillations in the zebrafish segmentation clock posit a negative feedback loop involving the her1 and her7 genes [8]–[10], which encode members of the Hairy and enhancer of split related (Hes/Her) family of basic helix loop helix (bHLH) transcriptional repressor proteins. Specifically, it has been proposed that oscillations arise through the auto-repression of these genes via a mix of Her1 and Her7 homo- and heterodimers, all of which have identical properties. This model is consistent with the reported redundant functions of her1 and her7 in somitogenesis [11],[12] and the observation that overexpression of either her1 or her7 leads to repression of both genes [13]. Furthermore, direct binding of Her1 homodimers to sites in the her1 promoter has recently been shown in vitro [14]. However, biochemical evidence for the other regulatory interactions proposed in this model is still lacking. It is also not clear how the proposed promiscuous protein-protein interactions and the equivalent functions of the resulting dimers in the current model can be reconciled with the reported distinct loss-of-function phenotypes of either gene. Knockdown of her1 results in segmentation defects preferentially located in the anterior trunk, while her7 knockdown in contrast leads to fully penetrant posterior segmentation defects [11],[12].
The period of the oscillations of the segmentation clock, in concert with embryonic elongation, determines the number of embryonic and adult segments, and is therefore of key importance in determining a species-specific body plan [15]–[17]. How the period of single-cell oscillations is controlled molecularly is not known. The cyclically expressed genes her1 and her7 have been proposed to differentially regulate the period as a consequence of different protein production delays [10], but this has not been tested experimentally. The only bHLH factor gene for which there is experimental evidence for a role in controlling the period of oscillations is hes6 [17]. hes6, in contrast to her1 and her7, is not cyclically expressed, but displays an FGF-dependent posterior-to-anterior expression gradient in the PSM [18]. In addition to its role in setting the period of the segmentation clock, hes6 contributes to stabilizing the transcriptional oscillations of her7 and her1 in the PSM [17],[18]. The Hes6 protein physically interacts with Her1 [18], but the interactions of Hes6 with other cyclic clock components have not been explored, and consequently, the molecular mechanism by which Hes6 controls the period of the oscillations is not understood.
Here we map the topology of the regulatory interactions between Her1, Her7, and Hes6 and DNA sequences in the promoters of cyclically expressed genes. We find that all of the possible dimers between Her1, Her7 and Hes6 form, but only Her1 homodimers and Her7:Hes6 heterodimers have strong DNA binding activity and target similar DNA sites. Using our experimentally determined network topology, we develop a simple mathematical model that can account for single and double mutant phenotypes that we observe. In this model, sequestration of monomers into dimers without DNA binding activity underlies the observed distinct phenotypes of genetic mutants. A surprising prediction of this model is that Hes6 protein levels oscillate post-transcriptionally, and we confirm this with a novel Hes6 antibody. Together, our results lead to a major revision of the current model of the core circuitry of the zebrafish segmentation clock and emphasize the importance of the properties of the Hes/Her protein-protein interaction network in controlling the clock's dynamics.
To investigate the topology of the regulatory network formed by Her1, Her7, and Hes6 we first asked which dimers form between the three factors. Co-immunoprecipitation suggested that all possible dimers form rather promiscuously, and even the bHLH containing factor MyoD, but not the negative control non-bHLH protein PPARγ, was co-purified by Her1, Her7, and Hes6 to a similar extent (Figure S1). To verify these findings in an independent setup and to investigate the DNA binding activities of the different dimers, we employed a microfluidic platform and mechanically induced trapping of molecular interactions (MITOMI, schematically depicted in Figure 1A–D) [19]. In this system, GFP-tagged Her proteins are immobilized on the surface of a microfluidic chip, and the pulldown of mCherry-tagged proteins is used to assess protein-protein interactions. The DNA binding activity of the resulting dimers can be investigated in the same assay by adding a DNA fragment that is labeled with a different fluorophore. We therefore coupled GFP-tagged versions of each of our three Hes/Her proteins to the chip surface, expressed different mCherry-tagged bHLH proteins in individual chambers of the chip [20],[21], and added to the expression mix a Cy5-labeled DNA fragment that contains the 12 mer CGACACGTGCTC from the her1 promoter. We chose this sequence because it has previously been shown to interact with Her1 in electrophoretic mobility shift assays [14]. Varying the amount of expression template spotted in the reaction chambers allowed us to titrate the concentration of mCherry-tagged protein in each chamber (Figure 1E,F). Protein concentrations reached by on-chip expression are below the dissociation constant Kd reported for the prototypical bHLH proteins E12 and MyoD [22]. At these concentrations, the amount of bound protein for any concentration of free protein is approximately inversely proportional to the dissociation constant [23], and therefore the slope of the linear fit to the plots of the free mCherry signal against the normalized signal from bound mCherry (Figure 1F) can be used as a measure for the relative affinities of protein-protein interactions.
When we coupled GFP-tagged Her7 or Hes6 to the chip surface, the four mCherry-tagged bHLH proteins showed a 4- to 20-fold higher relative affinity to the immobilized protein compared to the negative control PPARγ. The differences in relative binding affinity between bHLH proteins, however, were at most 2-fold (Figure 1F,G), in agreement with our immunoprecipitation experiments. In this assay, GFP-tagged Her1 coupled to the chip formed mainly homodimers. Heterodimerization of Her1 with Her7 and MyoD occurred with approximately 3-fold lower relative affinity, but was still significantly stronger than binding of PPARγ to Her1 (p<0.01), and only the strength of Hes6-mCherry with Her1-GFP was not significantly different from that of the PPARγ–Her interaction in this assay (Figure 1G). Because this latter finding is in contrast to the results of our co-immunoprecipitation experiments, and the interactions of Hes6-GFP with Her1-mCherry in MITOMI experiments, we suspect that steric factors hinder the formation of Hes6:Her1 heterodimers when Her1 is coupled to the chip surface. Taken together, our protein-protein interaction studies indicate that interactions are non-selective between the Her1, Her7, Hes6, and MyoD bHLH proteins.
As a next step we analyzed the DNA binding activity of all homo- and heterodimers detected in our protein-protein interaction experiments. In the presence of mCherry-tagged Her7 or MyoD, GFP-tagged Her7 immobilized on the chip surface did not bind the DNA sequence derived from the her1 promoter, whereas the presence of Her1-mCherry conferred weak binding, and the presence of Hes6-mCherry conferred strong DNA binding activity (Figure 1E,H,I). A similar situation was observed for Hes6-GFP, which bound the same DNA fragment strongly in the presence of Her7-mCherry, weakly in the presence of Her1-mCherry, but did not bind DNA in the presence of Hes6-mCherry or MyoD-mCherry (Figure 1I). Her1-GFP coupled to the chip surface bound DNA in the absence of any coexpression partner, presumably as a homodimer (unpublished data), and this binding was further increased by the presence of Her1-mCherry, but not Her7-mCherry or Hes6-mCherry (Figure 1G). These findings suggest that, while protein-protein interactions between Hes/Her factors are promiscuous, the DNA binding activity of the resulting dimers is restricted: Of the dimers formed between the three transcription factors investigated here, Her1 homodimers and Her7:Hes6 heterodimers bind most strongly to the target sequence from the her1 promoter.
To identify additional potential Hes/Her binding sites in cyclic gene promoters and to test whether Her1 homodimers and Her7:Hes6 heterodimers have similar or distinct DNA binding preferences, we sought to systematically investigate the DNA binding specificity of the two dimer species. We did this again using MITOMI, but now deposited increasing concentrations of different labeled DNA sequences in the chambers of the microfluidic chip. For low DNA concentrations, the amount of bound DNA for any concentration of free DNA is inversely proportional to the dissociation constant Kd of the complex [23]. The slopes of the linear fits to the data points for individual sites can therefore be used to compare relative binding affinities of a set of sequences to a given protein. To identify the consensus binding sequence of zebrafish Hes/Her factors, we first tested binding of all 64 permutations of the sequence CACNNN by Her1 homodimers and Her7:Hes6 heterodimers. We found that both dimers prefer the consensus site CACGNG (with N = T conferring stronger binding than A, G, and C) over all other sequences (Figure 2A,B). We term this common consensus binding site of Hes/Her proteins the H-box.
Although both Her1 homodimers and Her7:Hes6 heterodimers prefer binding to the H-box consensus site, it is possible that the two dimers prefer distinct bases flanking the core hexamer and thereby preferentially regulate different genes in the segmentation clock. To test this idea, we created a library comprising all H-box sequences flanked by three nucleotides 3′ and 5′ as they occur in the genomic context within 20 Kb of the her1/her7 locus and 12 Kb of the dlc cyclic gene locus (see Table S1 for sequences and localization of these H-boxes). We first measured relative binding affinities of the sequences in this library towards Her1 homodimers. We detected a characteristic profile of sites with a range of affinities (Figure 2C, Table S1). In this library, sequences that contain the core hexamer CACGGG or CACGAG were bound at most 5-fold stronger than control sequences that lacked an H-box consensus, whereas all H-box sequences with a more than 5-fold stronger relative affinity to Her1 homodimers compared to control sequences contain either a CACGTG or a CACGCG core hexamer (Table S1). This suggests that the relevant H-box in vivo is CACG[T/C]G. Furthermore, we found that the bases flanking the core hexamer influence binding affinity. Certain flanking bases reduced the affinity of sites containing the optimal CACGTG consensus by more than an order of magnitude (compare sequences 2 (TGGCACGTGTCC) and 7 (CATCACGTGAAA) from the dlc promoter, Table S1). In the appropriate sequence contexts, the CACGCG hexamer was bound almost as strongly as the highest affinity CACGTG-sites (e.g., sequence 26 (GGGCGCGTGCCG) from the her7 promoter, Table S1). With a few exceptions, H-boxes flanked by G or C were generally bound more strongly than those flanked by A or T. Importantly, we found that several H-box sites from the dlc, her1, and her7 promoters were bound by Her1 homodimers with comparably high affinity (Figure 2C), suggesting that Her1 can potentially regulate all three genes.
To test whether this was also the case for Her7:Hes6 heterodimers, we determined the relative binding affinities of the same sequences to Her7:Hes6 heterodimers and plotted their values against those determined for binding to Her1 homodimers (Figure 2D). If the target site specificity of Her7:Hes6 heterodimers was distinct from that of Her1 homodimers, and Her7:Hes6 heterodimers preferentially regulated one of the three cyclic genes, we would expect that the datapoints corresponding to sequences derived from different promoters cluster together. However, this was not the case: the datapoints were evenly distributed along the diagonal of the plot, irrespective of their origin.
Finally, motivated by our finding that Hes6:Her1 and Her7:Her1 heterodimers weakly bound an H-box sequence from the her1-promoter (Figure 1), we wanted to test whether these heterodimers target a distinct subset of H-boxes. We therefore co-expressed Hes6 or Her7 with Her1 coupled to the chip. This did not change the affinity profile of Her1 (Figure S2), which suggests that Her1:Hes6 or Her1:Her7 heterodimers do not bind to a subset of H-boxes distinct from the ones bound by Her1 homodimers. Combined, these results indicate that the strongest DNA binding is from Her1 homodimers and Her7:Hes6 heterodimers; these have similar DNA binding specificity, and each dimer has the potential to directly repress dlc, her1, and her7.
To validate our findings from these biochemical assays in a more physiological setting, we employed a yeast one-hybrid (Y1H) assay to assess binding of Her proteins to cyclic gene promoter fragments in the context of a eukaryotic nucleus. We selected 18 promoter fragments from approximately 100 bp to 1 Kb in length that cover between 3 and 4 Kb upstream of the transcriptional start sites of her1, her7, and dlc, cloned them upstream of the lacZ and the His3 reporter genes and stably integrated them into the yeast genome. We first used the pDESTAD vector system [24],[25] to express individual Hes/Her proteins tagged with the Gal4 activation domain (AD) as protein prey in our DNA bait strains. In this assay we found interaction of Her1, but not Her7 or Hes6, with a number of DNA baits (Figure 3A–C). This is consistent with the results of our MITOMI assays and confirms that Her1 homodimers, but not Her7 or Hes6 homodimers, have DNA binding activity. Seven of the nine fragments that displayed interactions with Her1 in this Y1H assay contain H-box sites that showed medium or high affinity binding to Her1 in the MITOMI assay (red or yellow bars in Figure 3D), whereas the majority of fragments that were negative in the Y1H contain no H-box sites or H-box sites with low affinity (black bars in Figure 3D). In the two fragments from the dlc and the her7 promoter that contain a high-affinity H-box site but do not give a signal in this assay, steric factors such as nucleosome arrangement might hinder transcription factor binding or reporter gene expression. Taken together, these Y1H assays suggest that most of the binding sites we identified in vitro are bound by Her proteins in the context of a eukaryotic nucleus.
Next, to confirm our finding that Her7 and Hes6 gain DNA binding activity through heterodimerization, we used a novel Gateway-compatible vector called pDESTAD-DIMER (Hens et al., in preparation) that allows expression of two AD-fused proteins from the same vector. This vector system has lower sensitivity than the pDESTAD-system—when we expressed two copies of Her1 from the pDESTAD-DIMER vector we detected only three interacting fragments, those that gave strong signals with the pDESTAD-system (compare Figure 3E to Figure 3A–C). Expression of two copies of Her7 or Hes6 gave no signal, but co-expression of Her7 and Hes6 resulted in interaction with the same fragments that were targeted by Her1 (Figure 3E). This finding is consistent with our in vitro measurements and suggests that, in the context of a eukaryotic cell, Her7 and Hes6 need to heterodimerize to gain DNA binding activity, and that Her1 homodimers and Her7:Hes6 heterodimers target the same sites in the promoters of the cyclic genes her1, her7, and dlc. Together, the protein–DNA binding assays described here suggest a network topology for transcriptional regulation in the segmentation clock where Her1 homodimers and Her7:Hes6 heterodimers form two parallel redundant feedback loops that converge on the same regulatory sites in the her1, her7, and dlc promoters.
To assess the relevance of the two-loop network described above for oscillatory gene expression and segmentation in the embryo, we decided to examine the phenotypes of embryos with mutations in the network's components. Because each of the three genes is involved in only one of the two feedback loops, the two-loop model predicts that her1, her7, and hes6 single gene mutants should all be competent to oscillate and to support somite segmentation.
The hes6 mutant has previously been shown to segment normally along its entire axis [17], and accordingly we found clear evidence for transcriptional oscillations of her1, her7, and dlc at the 10-somite stage (Figure S3A–C). The hes6 mutant phenotype is therefore in agreement with the predictions of the network topology. Previously reported defects in cyclic gene oscillations after hes6 morpholino knockdown may have resulted from off-target effects, or from the temperature at which the embryos were incubated in these experiments [18],[26].
The consequences of loss of her1 and her7 function in somitogenesis have also previously been addressed using morpholino antisense oligonucleotides [11],[12],[27]. To overcome possible off-target effects or incomplete gene knockdown associated with this approach, we decided to re-examine her1 and her7 loss-of-function phenotypes in genetic mutants. These mutants were generated by ENU mutagenesis [28] and carry premature stop codons in the her1 and her7 genes, respectively (Figure S4A,B,E,F). These genetic lesions lead to full loss of function of the respective gene (Figure S4C,D,G,H).
Most her1 mutants segmented normally along most of the trunk and tail (Figure 4A). Segmentation defects occurred only in a subset of her1 mutant embryos and were preferentially localized in the anterior trunk (see Figure S4C and D for a quantitative analysis of segmentation defects in her1 mutants). We observed evidence of tissue-level transcriptional oscillations in her1 mutants already at the bud stage (Figure S5A,B), suggesting that the partially penetrant anterior defects are not caused by a failure in establishing transcriptional oscillations at early somitogenesis stages. Tissue-level transcriptional oscillations of her7 mRNA were also evident at the 10-somite stage (Figure 4B). Because the wave pattern of transcriptional oscillations is altered in her1 mutants compared to wildtype embryos, it is difficult to visualize oscillating expression of her1 and dlc mRNA on the tissue level using standard chromogenic reagents. Subcellularly, transcriptional oscillations manifest in a succession of distinct localizations of the mRNA of oscillating genes—in early phases of the oscillatory cycle, cyclically expressed mRNAs are found in distinct nuclear spots at the sites of transcription, whereas later in the cycle they localize to the cytoplasm and give a diffuse staining [29]. We therefore used tyramide chemistry to detect changes in the subcellular localization of mRNA as a proxy for transcriptional oscillations. With this method, we were able to detect both her1 and dlc mRNA in distinct subcellular localizations in different her1 mutant embryos (Figure 4C,D), indicating that oscillations of dlc mRNA and the mutant her1 mRNA continue in the her1 mutant PSM. The her1 mutant phenotype is therefore in agreement with the predictions from the two-loop network topology—in the absence of Her1, the function of the Her7:Hes6 heterodimer is sufficient to drive oscillatory expression of her7, dlc, and the mutant her1 mRNA. The genetic her1 loss-of-function phenotype described here is consistent with reports from previous studies using MO-mediated gene knockdown [11],[12],[30] and suggests that the more widespread segmentation defects upon her1 knockdown that were reported by one other study [27] reflect off-target effects.
her7 mutants, in contrast to her1 and hes6 mutants, patterned only the anterior trunk correctly, and segmentation became defective in all her7 mutant embryos at the level of the 10th or 11th segment (Figure 4E, red arrowhead; Figure S4G,H). These posterior segmentation defects are accompanied by the decay of tissue-level transcriptional oscillations of her1, her7, and dlc between the bud- and the 10-somite stage (Figure 4F–H, Figure S5C–E). This her7 mutant phenotype is consistent with previous studies using MO-mediated her7 knockdown, where tissue-level oscillations have been shown to decay gradually [11],[12],[30]. The oscillatory behavior in the anterior trunk of her7 mutants is consistent with our expectations, but the highly penetrant posterior failure of cyclic gene expression and segmentation is not predicted by the simple two-loop network topology. The finding that posterior segmentation defects are observed in her7 but not hes6 mutants suggests that the Her7:Hes6 heterodimer is not the only functional species involving these proteins. We will return to this issue below.
The observation that transcriptional oscillations occur in the absence of her1, her7, or hes6 alone supports the two-loop topology determined in our protein-DNA binding studies. A further prediction from this topology is that lesions to both loops should cripple the oscillator. To test this idea, we analyzed the phenotypes resulting from combined loss of her1 and her7 or her1 and hes6 function. In agreement with previous observations using only MO-mediated gene knock-down [11],[12],[31] we found that her1;hes6 double mutants and her1 mutants injected with her7 targeted MOs displayed a failure of segmentation along the entire axis (Figure S6A,C) and show no sign of oscillatory gene expression at the 10-somite or bud stage, respectively (Figure S6B,D). These findings indicate that there are no feedback loops in the zebrafish segmentation clock that operate independently from her1 or her7 and hes6 and that are sufficient to drive tissue-level transcriptional oscillations of her1, her7, or dlc, or segmentation of the embryo's body axis. Importantly, since lesions in both of the predicted loops prevent cyclic gene oscillations and embryonic segmentation, these phenotypes are again in agreement with the proposed network topology.
As a last test of the network's two-loop topology we examined cyclic gene expression and segmentation in her7;hes6 double mutants. Since Her1 homodimers with DNA binding activity are available in this condition, the network structure predicts that oscillations will at least initiate in this double mutant. Yet because of the role of Her7 in maintaining tissue-level oscillations throughout segmentation stages that became apparent in the her7 single mutant, we expected the oscillations to likewise decay in her7;hes6 mutants, and segmentation to become strongly defective in the posterior trunk and tail. However, we found that most her7;hes6 double mutants segmented normally and posterior segmentation defects occurred with a low severity and penetrance comparable to hes6 single mutants (Figure 5A). Furthermore, the expression patterns of the mutant her7 mRNA in her7;hes6 double mutants at the 10-somite stage showed clear indications of tissue-level oscillations, similar to the situation in hes6 single mutants (Figure 5B). These results indicate that hes6 is fully epistatic over her7: Since identical phenotypes are observed in hes6 single and her7;hes6 double mutants, her7 function in somitogenesis is entirely dependent on the presence of hes6. On the other hand, the fact that the decay of oscillatory gene expression and defective segmentation in a her7 mutant background can be rescued by mutating hes6 indicates that hes6 is not neutral in the absence of her7 function, but dominantly interferes with clock function in this condition. This suggests that segmentation defects in her7 mutants are caused by Hes6 protein, which would contribute to Her7:Hes6 heterodimers in wildtype embryos, but in the absence of Her7 interferes with other critical components of the segmentation clock, such as Her1 homodimers. Importantly, the normal segmentation and cyclic gene expression patterns in the her7;hes6 double mutant combination is in agreement with the redundant two-loop negative feedback network topology predicted by our in vitro studies.
We conclude that the single and double mutant phenotypes support the expectations from the biochemically determined two-loop topology of the segmentation clock's core circuit, but also reveal additional functions of Her7 and Hes6 in the segmentation clock.
While the phenotypic assays described above provide a straightforward test of the circuit's basic topology by asking whether the oscillatory state of the circuit can be initiated and maintained in mutant conditions, they do not probe more subtle aspects of the circuit's dynamics. In particular, neither the static phenotypes nor the simple topology of the network allow inference of how the period of oscillations is regulated. A previous mathematical model of the segmentation clock has emphasized the role of transcriptional and translational delays in a hes/her feedback system in setting the period of oscillations [10]. Specifically, it would be predicted that oscillations relying exclusively on a her7-based feedback loop might be faster than oscillations that are exclusively based on her1, because of shorter production delays in the her7-based loop. We have previously shown that mutating hes6 slows segmentation clock period [17], and our protein-DNA binding data indicate that Her7 requires Hes6 to gain DNA binding activity. Therefore, it seemed possible that the period slowing in hes6 mutants was due to the loss of the fast Her7:Hes6 loop and reflected a slower Her1-based loop operating in isolation. If this were the case, then we would expect that the formation of anterior segments in her7 single and her7;hes6 double mutants would be slowed as in hes6 mutants. Furthermore, if the slower her1 loop modulated the period in the wildtype, her1 mutants might segment faster than wildtype embryos.
We used multiple-embryo time-lapse imaging to test this idea and recorded the periodicity of somite formation as a morphological proxy for the oscillations of the segmentation clock [32],[33]. We found that somites form with similar dynamics in wildtype, her1 mutant, and her7 mutant embryos (Figure 6). Somitogenesis period in her7;hes6 double mutants was slowed compared to trans-heterozygous controls, and comparable to hes6 homozgyous;her7 heterozygous mutants (Figure 6). These findings indicate that neither her1 nor her7 have a crucial role in setting the tempo of segmentation clock oscillations, and that hes6 regulates segmentation clock period independently from simply providing a heterodimerization partner for Her7. The period control function of hes6 can therefore not be mapped to the simple two-loop topology determined by our protein-DNA interaction studies, and this motivated us to consider a possible role for non-DNA binding dimers in regulating the dynamics of the segmentation clock.
To explore the complex regulatory possibilities of the circuit we have described and determine whether wildtype and mutant phenotypes can be explained in a rigorous, internally consistent manner, we decided to investigate the dynamics of a mathematical model of the network describing the behavior of Her1, Hes6, and Her7 proteins (see Text S1, Tables S2, S3, and Figures S7, S8, S9, S10, S11 for details). Following our protein-protein and protein-DNA binding results, our model allows for formation of all possible dimers between Her1, Hes6, and Her7, but negative feedback regulation of her1 and her7 occurs exclusively through Her1:Her1 homodimers and Her7:Hes6 heterodimers (schematically depicted in Figure 7A). We use a Hill function to describe repression through these dimers in our model. This is motivated by our finding that there are multiple binding sites for each of the dimers in every cyclic gene (Figure 3D) [34],[35] and does not reflect any assumptions about the binding mechanism of Hes/Her dimers to individual target sites. For simplicity, we ignore potential transcriptional regulation via dimers with weak DNA binding activity. Nevertheless, the dimers that do not bind DNA in our model still perform an important post-translational regulatory function, since they sequester monomers and thereby affect the availability of dimers with DNA binding activity. The period measurements in Figure 6 provide strong constraints on the network's dynamics, and can be used to guide the choice of parameter values.
To understand the consequences of the basic network topology, and because reliable measurements of the rate constants of the processes in this network are not available, we intentionally avoid automatic methods to optimize model parameters to obtain the best possible quantitative fit to experimental results. Instead, when possible we keep rate constants associated with production, dimerization, and repression equal between the Hes/Her proteins (Table S3), and introduce only the minimum differences between species necessary to qualitatively reproduce experimental trends. For this reason, the parameter values we report below should not be understood as exact quantitative predictions. To simplify the model as far as possible, we focus only on the generation of oscillations within single cells and do not describe cell-cell coupling.
To analyze the dynamics of this network, we obtained numerical solutions of the minimal version of the model. The different mutant conditions were simulated by setting the production rate of the corresponding component(s) to zero. We started by setting the Hes6 production rate to zero, κ6 = 0, to describe a hes6 mutant; in this situation, the network can support high-amplitude oscillations (blue triangle in Figure 7B and second panel in Figure 7C). As the value of κ6 is increased, the period of oscillations decreases (blue line in Figure 7B, bottom; Figure S7). At κ6 = 90, the period of the simulated oscillations is approximately 6% faster than in a situation without hes6 (black point in Figure 7B), matching the experimentally observed difference in somitogenesis period between hes6 mutant and wildtype embryos (Figure 6C) [17]. We therefore chose κ6 = 90 as the wildtype value for Hes6 production. This value for Hes6 production is higher than the corresponding values κ1 = κ7 = 10 for Her1 and Her7 production, thereby distinguishing Hes6 from Her1 and Her7 at the parameter level. A parameter sensitivity analysis shows that the period and amplitude of oscillations are robust to changes in most of the parameters of the model (Figure S8).
This dependence of oscillation period on Hes6 production rate is caused by heterodimerization of Hes6 with Her1 and Her7, allowing Her1 and Her7 proteins to be degraded also as components of these heterodimers, which alters the effective stability of Her1 and Her7 (Figure S9). A striking consequence of the regulation of the effective half-lives of Her1 and Her7 by dimerization with Hes6 in the model is that Hes6 protein levels also oscillate, albeit with relatively low amplitude, despite a constant production (Figure 7D). We return to this distinctive prediction below.
Next, we simulated the her7 mutant by setting the Her7 production rate to zero, κ7 = 0. We find that the period of these simulated oscillations is shorter for a her7 mutant than the wildtype over a range of values for κ6 (Figure S7), in contrast to our experimental measurements (Figure 6). To fit our experimental data with our model as simply as possible, we introduce one more parameter asymmetry, choosing a slightly longer production delay τ for Her1 compared to Her7. Qualitatively, this choice appears to be justified by the physical properties of the two genes, because the her1 gene is longer and contains more introns than the her7 gene. Quantitatively, a relatively small difference is motivated from independent experimental data, because transcription is fast (∼4 Kb/min), intron splicing is co-transcriptional, and splicing times are short (∼5 min) and independent of length [36]. For a difference of 2% between the her1 and her7 delays, the oscillations in the simulated her7 mutant have a period similar to the wildtype situation, almost independently of our choice for κ6 (green line in Figure 7B). At κ6 = 90, oscillations initiate in the simulated her7 mutant, but the amplitude of these initial oscillations decreases over time, eventually falling to zero (green dot in Figure 7B, Figure 7C). This suggests that in the embryo, mutation of her7 results in lower amplitude or even fully damped oscillations on the single cell level, and this gradual damping of oscillations provides an explanation for the posterior segmentation defects in her7 mutants.
While the amplitude of oscillations in the simulated her7 mutant decays instantaneously and rapidly and Her7 trough levels are constantly elevated in this situation (Figure 7C), the amplitude of the early oscillations in bud stage embryos appears to be similar to the wildtype (Figure S5C–E). Therefore, although our model is qualitatively successful in explaining the her7 mutant phenotype, there remains a quantitative difference between the model and the data. We speculate that this may be due to the particular choice of parameters in our model, which have not been optimized to capture the amplitude of oscillations in early her7 mutant embryos, or due to effects of coupling between cells on the tissue level, which are not represented in our model.
In our model, the reduction of amplitude in the her7 mutant simulation arises because the resulting level of Her1 homodimer is insufficient to sustain oscillations (Figure S10). Although Her1 monomer production increases due to loss of repression via Her7:Hes6 heterodimers, most of it is sequestered as non-DNA binding Her1:Hes6 heterodimers. Consequently, decreasing κ6 in a simulated her7 mutant situation leads to a recovery of the amplitude of oscillations (green line in Figure 7B). Our model thereby also provides an explanation for the striking rescue of segmentation defects in the her7;hes6 double mutant. The period of oscillations in her7;hes6 double mutant simulations is slowed compared to the wildtype situation, in qualitative agreement with our experimental findings. We note, however, that our model predicts a period difference between the hes6 single and her7;hes6 double mutant situation (cyan square in Figure 7B and Figure 7C, bottom), which is not observed experimentally (Figure 6). In the model, this period difference is caused by heterodimerization and effective destabilization of Her1 by Her7 in the hes6 single, but not in the her7;hes6 double mutant. The difference between model and experiment suggests that we either overestimate this destabilization effect in our minimal parameterization of the model by setting all rate constants equal or that other proteins in the PSM that are not considered in this model can compensate for Her7 function in regulating Her1 stability.
In summary, our mathematical model indicates that the two-loop negative feedback topology of the her gene network as determined by our protein-protein and protein-DNA interaction studies is sufficient to explain the qualitative dynamic characteristics of the different experimentally observed single and double mutant phenotypes when the effects of dimers with and without DNA binding activity are considered. The success of this model supports an important role for both types of dimers in controlling the dynamics of the circuit, since together they determine the availability of DNA binding dimers. Below, we report on three experimental tests of the formation of Hes/Her dimers described in the model.
Our mathematical model makes predictions about the role of Hes6 production rate in controlling the period and the amplitude of oscillations. We reasoned that the expected roles of Hes6 production rate should be testable using the hes6 heterozygote mutant embryo. As predicted by our model, we observed that the periodicity of somite formation in hes6 heterozygous mutant embryos was slowed compared to their wildtype siblings (Figure S12A) and leads to a reduction in segment number that is in agreement with the period measurements (Figure S12B,C). Furthermore, we found a shift of the onset of defects towards the posterior when the hes6 locus was heterozygous mutant in a her7 mutant background, compared to her7 mutants with two wildtype hes6 loci (Figure S12D,E). This is consistent with the predicted tuning of the amplitude of Her1 oscillations by Hes6 levels. Together, these findings from hes6 heterozygous mutants are in agreement with the predictions of our mathematical model, providing additional support for the idea that Hes6 quantitatively affects segmentation clock functions by titrating critical oscillatory monomers and affecting their effective stability.
Finally, we sought to directly investigate the predicted effects of dimerization on the effective protein stability of Hes6. Our model suggests that dimerization of Hes6 with the cyclically expressed proteins Her1 and Her7 and subsequent degradation of the Her1:Hes6 and Her7:Hes6 heterodimers will manifest in post-transcriptional oscillation of Hes6 protein levels (Figure 7D), and we decided to test this prediction directly in the embryo (Figure 8). To first rule out that any potential oscillatory protein pattern could be caused by transcriptional oscillations of hes6 mRNA that had previously been overlooked [5],[17],[18], we searched 5 Kb upstream of the hes6 start codon for H-box sites. The only H-box-containing 12-mer within this stretch of DNA has the sequence ACTCACGTGAGA (unpublished data). Because the central CACGTG consensus is flanked by T and A, respectively, this H-box is not expected to be strongly bound by Her1 homodimers or Her7:Hes6 heterodimers (Table S1). Furthermore, when we examined the spatial pattern of hes6 mRNA expression in wildtype embryos at the 10-somite stage, we observed a smooth decay of the staining intensity from posterior to anterior; there was no evidence for spatial mRNA waves, which would be expected if Her1 or Her7 controlled hes6 expression (Figure 8A,B, n = 20). Finally, the pattern of hes6 mRNA expression is not altered in her1 and her7 mutants (Figure 8A,B, n≥20 for each genotype). This indicates that hes6 mRNA expression is not subject to transcriptional control by Her1 or Her7. Taken together, these data indicate that hes6 mRNA levels do not oscillate, in line with reports from several previous studies [5],[17],[18].
To test the prediction that Hes6 protein levels oscillate, we raised monoclonal antibodies against Hes6. Whole-mount immunostaining revealed a nuclear signal in the tailbud and posterior PSM of wildtype embryos (Figure 8C, left). No nuclear signal was obtained in hes6 mutant embryos, indicating the specific detection of Hes6 protein (Figure 8C, right). When we examined Hes6 immunoreactivity in wildtype embryos, a subset (10/26) showed striped staining patterns indicative of Hes6 protein oscillations (arrowheads in Figure 8D). These patterns can be compared by plotting the intensity profile of Hes6 immunoreactivity in the intermediate PSM (Figure 8E). Their shape and position recapitulates the wave pattern of her1 and her7 mRNA expression in this region of the PSM. Although we cannot formally rule out the possibility that these Hes6 protein oscillations might be produced by hes6 mRNA oscillations below our detection limit or by an influence of cyclically expressed proteins on Hes6 translation, this finding is consistent with our prediction that the cyclically expressed proteins Her1 and Her7 regulate Hes6 protein stability in the PSM.
To test the influence of Her1 and Her7 proteins on Hes6 protein oscillations, we examined Hes6 immunoreactivity in her1 and her7 mutants. If Her1 and Her7 oscillations were responsible for generating the tissue-level Hes6 protein oscillations, they should be lost in her7 mutants, where tissue-level oscillations of all cyclic genes examined decay by the 10-somite stage (Figure 4F–H). In contrast, Hes6 protein oscillations should remain in her1 mutants, where her7 is still expressed in a wave pattern (Figure 4B). In her1 mutants, we detected anterior waves of Hes6 immunoreactivity (n = 4/16, red arrowheads in Figure 8F,G), in line with our expectation of ongoing Hes6 protein oscillations in her1 mutants. These patterns were less pronounced than in wildtype embryos, consistent with the altered wave-like expression patterns of her7 mRNA in the her1 mutant (Figure 4B,D). In contrast to wildtype and her1 mutant embryos, we found no evidence for tissue-level protein oscillations in her7 mutants (Figure 8F,G, n = 16). Hes6 protein levels decayed smoothly from posterior to anterior, in parallel with the shape of the hes6 mRNA profile (Figure 8A,B,F,G), in line with our expectation. Taken together, our finding of characteristic cyclic wave patterns of Hes6 protein in wildtype embryos (despite smooth hes6 mRNA patterns) and the differentially perturbed patterns of Hes6 protein oscillation in the her1 and her7 mutants is consistent with one of the key features of our model, namely that heterodimerization regulates effective protein stability in the segmentation clock.
In this work we describe a new dynamic model for the core pace-making circuit of the segmentation clock, which is based on a systematic biochemical investigation of the network's topology and precise measurements of somitogenesis dynamics in novel genetic mutants. Our key finding is that the core pace-making circuit consists of two distinct negative feedback loops, one involving Her1 homodimers and the other involving Her7:Hes6 heterodimers, operating in parallel. We can account for the single and double mutant phenotypes of her1, her7, and hes6 mutants in a mathematical model of this core circuit, wherein protein-protein interactions between oscillating and non-oscillating Hes/Her genes control the availability of dimers with DNA binding activity. This is a new principle for the regulation of oscillatory dynamics that may be relevant beyond the zebrafish segmentation clock.
The protein-DNA interaction assays in this article extend previous efforts to biochemically map regulatory interactions in the segmentation clock. The interaction of Her1 homodimers with the her1 promoter using electrophoretic mobility shift assays was previously observed [14]. While this article was under review, evidence for the existence of binding sites for Her1 homodimers and Her7:Hes6 heterodimers in the her7 promoter was published [26]. Here we show that, in addition, Her7:Hes6 heterodimers also bind to the her1 promoter, and both Her1 homodimers and Her7:Hes6 heterodimers target sequences in the dlc promoter. Using MITOMI technology, we were able to determine and compare the DNA binding specificity of both types of dimers in an exhaustive and unbiased way. We find that both dimers bind to the core consensus sequence CACGNG. This sequence differs from the N-box sequence CAC[G/A]AG commonly reported for mouse Hes proteins [37], which had been determined via a candidate approach, but is similar to the consensus binding site CACG[T/C]G established for the Drosophila Hairy protein [38]. Therefore, it appears likely that binding to the CACGNG consensus sequence is a common feature of Hairy-related bHLH transcription factors in different species, and we term this consensus binding site the H-box. When we compare the relative affinities of Her1 homodimers and Her7:Hes6 heterodimers to H-box sites in the three cyclic gene promoters, we detect differences. Although this observation is in line with recent results [26], our data do not support the proposed hypothesis [26] that these differences might have functional consequences to the dimers in the context of the segmentation clock, since sites with relatively higher affinity to either of the two types of dimer do not appear to be enriched in any of the three promoters examined.
Our protein-DNA interaction experiments suggest a regulatory network architecture for the segmentation clock where her1 and her7 engage in two parallel, redundant negative feedback loops, which converge on the same DNA regulatory elements (Figure 7A). The concept of redundant her1- and her7-based feedback loops has been introduced before in dynamic models of the segmentation clock, following genetic evidence [8],[10],[11]. However, the biochemical realization of these redundant feedback loops that we determine here differs from these earlier models. Previously, all possible dimers between Hes/Her proteins were assumed to have DNA binding activity and participate in feedback regulation. We find here that although Hes/Her proteins indeed dimerize promiscuously, only Her1 homodimers and Her7:Hes6 heterodimers have strong DNA binding activity. These findings are again in general agreement with the recent study mentioned above [26]. An equilibrium description of the Hes/Her dimerization and DNA binding interactions presented in this recent work does not consider any dynamics [26] and thus does not model temporal behavior of the segmentation clock. In summary, Hes/Her dimers with low DNA binding activity have not been considered in previous dynamic models of the segmentation clock, but in our new model they have important functions in determining the dynamics of the circuit, as we will explain below.
Our finding that Her1 homodimers and Her7:Hes6 heterodimers target the same DNA-sites in the regulatory regions of her1 and her7 and within the promoter of the cyclic dlc gene thought to mediate coupling between single cell oscillators [10],[11],[13],[29] appears at first difficult to reconcile with the distinct phenotypes of her1, her7, and hes6 single mutants. It is of course possible that Her1 homodimers and Her7:Hes6 heterodimers recruit distinct accessory machineries to regulate transcription at cyclic gene promoters. Alternatively, Her1, Her7, or Hes6 might gain distinct DNA binding activity by selective dimerization with bHLH factors not investigated here, and either of these scenarios might contribute to the observed phenotypic differences. However, our analysis of a mathematical model that is solely based on the experimentally determined interactions between Her1, Her7, and Hes6 suggests that this small network alone is sufficient to account for the distinct mutant phenotypes. Our finding that Hes/Her proteins dimerize promiscuously and form a “dimer cloud” that contains complexes with strong and weak DNA binding activity provides the key ingredient required to generate these phenotypic differences.
In our model, period slowing upon loss of hes6 is caused by the influence of Hes6 on effective protein stability. Although we assume that degradation acts equally on Hes/Her monomers and dimers, the sequestration of monomers into dimers and degradation of these dimers introduces an effective monomer degradation rate that is a function of dimer concentrations, which are nonlinear functions (products) of the monomer concentrations. Experimental support for this feature of our model comes from the expression patterns of Hes6 protein. These show evidence for Hes6 protein oscillations, despite non-oscillatory hes6 mRNA expression. Our model suggests how these protein patterns might arise—the cyclically expressed proteins Her1 and Her7 dimerize with the constantly expressed Hes6 protein and thereby cyclically modulate its effective stability. Conversely, in our model dimerization of Hes6 with Her1 and Her7 results in a shorter effective half-life of the two cyclic proteins. Consequently, in the hes6 mutant situation, the effective stability of the oscillating proteins Her1 and Her7 is increased, and this increases the period. Previous theoretical work also shows that increased stability leads to longer oscillation period in Hes/Her feedback systems [10],[39],[40]. It will be interesting to directly investigate this hypothesis arising from our model by measuring Her1 and Her7 half-lives in wildtype and hes6 mutant embryos. Furthermore, according to our model, Hes6 overexpression should further reduce the effective stability of Her1 and Her7 and might therefore reduce the period of the clock. This will have to be tested in transgenic fish carrying additional copies of the hes6 gene in order to allow for controlled Hes6 overexpression and to circumvent gastrulation defects induced by hes6 mRNA injection [18].
Dimerization-induced changes in stability are known for several bHLH proteins, where complex formation usually results in an increased half-life [41],[42]. This effect is termed cooperative stability, and theoretical studies have established that this phenomenon can affect the dynamic behavior of genetic networks [43]–[45]. In the context of the segmentation clock, it has been suggested that cooperative stability could increase the region in parameter space where sustained oscillations are possible, rendering the oscillator more robust [46]. Although different monomer and dimer stabilities can be accounted for in our model, they are not necessary to recapitulate the observed embryonic phenotypes, and we do not explore their effects here.
While the effects of Hes6 on oscillation period in our model arise from its equally destabilizing effect on both Her1 and Her7, the fact that Hes6 forms heterodimers with strong DNA binding activity only with Her7 provides an explanation for posterior segmentation defects in her7 mutants. In the absence of Her7, more Hes6 engages in Her1:Hes6 heterodimers with low DNA binding activity, thereby reducing the pool of Her1 homodimers available for transcriptional regulation. In our model, this results in a decrease in the amplitude of oscillations, which is a plausible explanation for the posterior segmentation defects observed in her7 mutants. To directly test this idea experimentally, it will be necessary to follow oscillations in her7 mutants with single cell resolution.
The inactivating function of Hes6 towards Her1 in our model of the zebrafish segmentation clock contrasts with a previous report, where Hes6 was shown to increase the repressive activity of Her1 towards the her1 promoter [18], although the duration (48 h) and biological host (293T cells) used in this experiment makes direct comparison to the segmentation clock difficult. In the developing mouse nervous system, Hes6 has been shown to inactivate the Her1 homolog Hes1 [47], analogous to its function in our model. Negative regulation of bHLH factor activity by dimerization is a well-established concept: Id factors, which lack a basic domain, can inhibit DNA binding of tissue specific bHLH factors by forming inactive heterodimers [48], and Hairy-related transcription factors can inhibit the function of lineage-specific bHLH proteins by heterodimerization [49]. However, while these well-known examples of negative regulation by dimerization occur between proteins that belong to different classes of the bHLH family, the inactivation of Hes1 by Hes6 in the mouse nervous system and the formation of dimers with lowered DNA binding activity described in this work occur between proteins of the same subclass. Furthermore, our results suggest that Hes6 has opposite effects on the closely related proteins Her7 and Her1, being a necessary dimerization partner for Her7 to gain DNA binding activity, while inhibiting DNA binding of Her1. This is a new observation for bHLH proteins, and it will be interesting to investigate which structural features determine Hes6's mode of action towards different Hes/Her factors.
The role of Hes/Her dimerization in the zebrafish segmentation clock may be relevant for understanding the mouse segmentation clock. While the overall topology of the mouse and zebrafish segmentation clock networks, including intercellular signaling pathways, has clear differences, the presence of multiple oscillating Hes/Her genes is conserved [5]. The Hes1, Hes5, and Hes7 genes oscillate in mouse PSM, and Hes7 is required for oscillations via a transcriptional auto-repression negative feedback loop [50]. Hes7 mutant embryos have profoundly disrupted segmentation [51], but in contrast, Hes1 and Hes5 single and double mutants segment overtly normally [52]. This suggests that the Hes-based negative feedback loop topology differs between mouse and zebrafish. Based on our observations, we would expect that the Hes1 and Hes5 proteins, as well as non-oscillating bHLH proteins with ubiquitous expression [53], could participate in a dimer sequestration mechanism with Hes7 similar to the zebrafish clock, thereby potentially regulating the dynamics. Although there is currently no evidence that Hes genes regulate the period in mouse, the development of techniques to measure the dynamics of mouse segmentation with sufficient precision and temporal resolution may allow such effects to be detected.
Because of the general role of Hes6 in controlling protein stability and its differential effects on the DNA binding properties of Her1 and Her7, the activities of these proteins need to be appropriately balanced to ensure the reliable function of the zebrafish segmentation clock in the wildtype. This need to balance competing activities appears to be a common theme in different dynamic biological systems. In the developing Drosophila eye, the activity of a network of bHLH factors has recently been shown to direct the timing and spacing of cellular differentiation [54]. In the genetic network of the mouse circadian clock, loss of mPer2 disrupts circadian rhythmicity, and this phenotype can be rescued by disruption of mCry2 in mPer2 mutant mice [55]. This rescue phenotype is similar to the restoration of segmentation clock oscillations by mutating hes6 in a her7 mutant background that we describe here, and by joint morpholino knockdown described recently elsewhere [26]. This suggests that, similar to control of Hes6 levels by Her7 in the segmentation clock, mPer2 levels in the circadian clock need to be held in check by mCry2.
In this work, we take a reductionist approach to understand the identity and function of the core oscillator of the zebrafish segmentation clock: We investigate regulatory interactions between only three genes and try to understand their system-level loss-of-function phenotypes by modeling the dynamics of a small single-cell network. Our success in modeling a range of phenotypes with this small set of components, as well as the finding that joint loss of her1 and her7 or her1 and hes6 function completely abrogates oscillatory cyclic gene expression and segmentation (Figure S6) [11],[12],[31], indicates that the two-loop system described here forms the core of the zebrafish segmentation clock. Nevertheless, the dynamics of this core network may be influenced by its interaction with factors not considered here. For example, several other hes/her genes as well as tbx16 display transcriptional oscillations in the PSM [5],[56],[57]. Specifically, it has been shown that the gene products of the oscillating genes her12 and her15 bind targets sites from the her7 promoter in vitro both as homodimers and as heterodimers with Hes6 [26]. In addition, we show here that Hes/Her proteins can interact with non-hairy bHLH proteins such as MyoD with considerable affinity, which presumably impacts on their activity. How cyclic and non-cyclic bHLH proteins integrate into the core network formed by Her1, Her7, and Hes6 and modulate its dynamics will depend on whether they form dimers with or without DNA binding activity amongst each other and with the three proteins investigated here.
Because we focus on the regulation of oscillatory dynamics at the single cell level, we have not explicitly addressed tissue-level aspects of oscillatory gene expression in the PSM. We have previously shown that coupling of single cell oscillators through Delta-Notch signaling can modulate the period of tissue-level oscillations [58],[59]. Therefore, although our single cell model allows us to fit a range of dynamic tissue-level phenotypes, it is possible that these phenotypes do not solely arise from changes in the regulatory interactions within single cells, but might also depend on the modulation of altered single cell dynamics by cell-cell communication.
Furthermore, on the tissue level the transcriptional oscillations of the segmentation clock manifest as moving stripes of gene expression, and theoretical studies have established that this spatial aspect of the clock's oscillations can be reproduced by a gradual slowing of cell-autonomous oscillations in more anterior regions of the PSM [6],[13],[59]–[65]. While our model describes the situation in the posterior PSM, this gradual slowing could be achieved by a changing configuration of the hes/her gene regulatory network in more anterior parts of the PSM. The integration of oscillatory her genes with region-specific expression in the PSM [5],[56],[57] into the core network might provide one mechanism for the position-specific control of single-cell oscillatory period.
The graded expression of hes6 across the tailbud and PSM (Figure 8, [18]) has been suggested to directly control the slowing of oscillations [8], but the evidence here does not obviously support this scenario. Mutating hes6 changes the period of oscillations by only 6% [17], and theoretical work shows that such a small period change is insufficient to generate the observed wildtype wave pattern [46]. Furthermore, we found that mutating hes6 does not grossly affect the wave patterns of cyclic gene expression (Figure 5B). In contrast, we note that the wave patterns are strongly altered in her1 mutants, although the overall pace of segmentation is not (Figures 4 and 6). This suggests that her1 has a primary role in period control in the anterior PSM where oscillations slow down. Conversely, in the tailbud where the pace of segmentation is determined, elevated levels of hes6 expression may allow the Her7:Hes6 heterodimer to exert a dominant control over the circuit's period. Thus it is tempting to speculate that the wildtype zebrafish segmentation clock may use changing Hes6 expression levels to switch between the two core oscillatory negative feedback loops in a position-dependent manner across the PSM.
In summary, combining biochemical and genetic data with mathematical modeling, we have developed a model for the zebrafish segmentation clock's core circuit with a novel regulatory topology and an unexpectedly prominent role for a “dimer cloud” of Hes/Her complexes in regulating the formation and availability of DNA binding dimers. Oscillating systems generally must display (i) negative feedback, (ii) delays in this feedback, (iii) sufficient nonlinearity, and (iv) a balance of timescales [66]. Our biochemical measurements of DNA-binding Her1 homodimers and Her7:Hes6 heterodimers onto target sites from oscillating gene promoters reveal the redundant two-loop topology of the negative feedback in the core circuit, as well as a likely source of strong non-linearity via the existence of multiple binding sites [67]. This redundant two-loop topology may provide the circuit with robustness to genetic and environmental perturbation, while the distinct components of each loop could provide the core circuit with independent input and/or output regulatory linkage that might vary with the position of the cell in the PSM, or with developmental stage, or through evolutionary transitions. The balance of timescales in the system is controlled in a surprising way: Hes6 acts to tune this balance by regulating the effective degradation of the oscillating DNA binding dimers, thereby changing the period of the clock. This global regulation of the stability of the oscillating components, and hence the period, through protein-protein interaction with a component whose levels can be smoothly and gradually tuned by external signals is a new control mechanism for the segmentation clock. How widely this principle is employed in other genetic regulatory systems remains to be explored.
her1, hes6, her7, and myoD coding sequences were PCR-amplified from wildtype zebrafish cDNA, cloned into pDON221 entry vectors using Gateway technology, and verified by sequencing. pMARE vectors for in vitro expression of C-terminally GFP-tagged proteins have been described [68]. To generate pMARE vectors for in vitro expression of mCherry- and GST-tagged bHLH proteins, the GFP-coding sequence of pMARE was replaced by the respective mCherry or GST coding sequence. her1, hes6, her7, and myoD coding sequences were subcloned into pMARE via a Gateway LR-reaction. Expression constructs for yeast one-hybrid experiments were generated by LR-subcloning of her1, her7, and hes6 cDNAs into pDESTAD [25] or pDESTAD-DIMER. The generation of pDESTAD-DIMER will be described elsewhere (Hens et al., in preparation). Target DNA sequences for MITOMI experiments were Cy5-labeled as previously described [19].
pMARE expression vectors or labeled target DNAs in a carrier solution containing 1% BSA were deposited onto epoxy coated glass sub-strates (CEL Associates) using a QArray Microarray spotter. MITOMI experiments were performed using microfluidic chips with 768 reaction chambers. Flow and control molds and microfluidic devices were fabricated, aligned, and bonded to spotted slides as previously described [19],[69], and initial surface derivatization was performed according to published protocols [19]. Antibodies to immobilize tagged Her-proteins on the chip surface were anti-GST and anti-GFP (abcam). For protein-protein interaction experiments, 12.5 µL of a TnT SP6 high-yield wheat germ protein expression extract (Promega) were programmed with 500 ng of expression vector encoding C-terminally GFP-tagged Her-proteins, incubated for 3 h to express protein and loaded onto the chips with buttons closed. Flow was then stopped and buttons opened to allow pulldown of GFP-tagged Her protein to the button area from the volume corresponding to one unit cell. Excess GFP-tagged protein was removed by extensive washing with PBS. Next, unprogrammed wheat germ extract containing 250 nM of a Cy5-labeled DNA-oligomer containing the CGACACGTGCTC sequence was loaded onto the device and flushed for 5 min, after which the chamber valves were opened allowing for dead end filling of the chambers with the expression extract [20]. For experiments testing the interaction of one dimer combination with multiple DNA-sequences, chambers were filled directly after depositing the antibody to the button surface with a wheat germ extract programmed with the respective expression constructs. Devices were incubated for 3 h at room temperature to allow for protein expression and equilibrium binding. Device imaging and image and data analysis were performed as previously described [19]. Data fitting and statistical analysis of the fits was performed in GraphPad Prism. Error bars state standard error of the linear fits.
Eighteen fragments between approximately 100 bp and 1 kb in length upstream of the her1, her7, and dlc start codon were PCR-amplified from genomic zebrafish DNA, introduced into pDONR-P1P4R using Gateway technology, sequence verified, and subcloned into pDB-DEST-His and pDB-DEST-LacZ [25]. Y1H bait strains were generated and tested for self-activation as described [25]. Reporter strains were transformed with pDESTAD or pDESTAD-DIMER vectors encoding N-terminal Gal4-AD fusions of different Her-proteins according to standard procedures, and transformants selected by growth on media lacking tryptophan. Yeast cells were arrayed using a Singer Rotor pinning robot, and interactions determined by assaying growth on 3-aminotriazole containing minimal media or β-gal expression according to standard procedures.
Wildtype and mutant zebrafish were maintained according to standard procedures and embryos obtained by natural spawning. hes6 mutants have previously been described [17]. The her1hu2124 and the her7hu2526 alleles were generated by ENU mutagenesis [28] and distributed in the framework of the ZF-MODELS project. Carriers of the her1 and her7 mutant alleles were continuously backcrossed to the AB wt strain to reduce background mutations. Backcrossing improved longevity of adult fish but did not impact on selected embryonic phenotypes such as anterior or posterior segmentation defects in the her1 or her7 mutant, respectively. Homozygous her7 mutants for raising and double homozygous her1;hes6 mutants were identified visually by scoring for segmentation defects around the 18-somite stage. her7 mutants in period measurement experiments and all other mutants were identified by PCR-based genotyping protocols. Genomic DNA was isolated from tissue samples (adult fish) or whole embryos using standard procedures [28]. her1 mutant fish were identified by digesting a PCR amplicon covering the hu2124 lesion with TfiI. This restriction site is present in the wildtype, but not the mutant allele. her7;hes6 double mutants were identified by sequencing of a PCR amplicon of the her7 locus covering the hu2526 lesion and PCR genotyping of the hes6 lesion as previously described [17]. Primer sequences and reaction conditions for all PCR-based genotyping protocols are available from the authors upon request. All single and double mutants described in this work were homozygous viable and fertile.
Embryos for period measurements in single mutants and myotome counts in hes6 heterozygous mutants were obtained from incrosses of heterozygous carriers of the respective lesion. Embryos for analyzing somitogenesis period in her7;hes6 double mutants were obtained by crossing trans-heterozygous and double homozygous carriers. Experiments addressing how the onset of segmentation defects in her7 mutants depends on hes6 zygosity were performed on embryos from incrosses of her7 homozygous;hes6 heterozygous adults. Embryos to determine somitogenesis period, myotome number, and position of segmentation defects were individually genotyped after analysis to eliminate potential analyzer bias. Embryos for all other experiments were obtained from incrossing of homozygous carries of the desired genotype.
Antisense probes to her1 [70], her7, dlc [11], cb1045 [71], hes6 [17], and myoD [72] have been described. In situ hybridization using NBT/BCIP and FastRed chemistry was performed as described [11],[73]. For fluorescent in situ staining using tyramide signal amplification (TSA), riboprobe hybridization, washes, and antibody incubation were carried out as described above, except that peroxidase-coupled anti-digoxigenin antibody (Roche) was used. Color development was performed with Cy3-coupled tyramide according to the manufacturer's recommendations (Perkin Elmer). Cy-3 stained embryos were cleared with methanol, deyolked, and imaged on a Zeiss Axioskop 200 M equipped with a Photometrics Coolsnap HQ Camera and a motorized stage (Zeiss MCU 28) driven by MetaMorph software (version 6.2r4, Universal Imaging Corp.). Z-stacks were deconvolved with Huygens software (Scientific Volume Imaging), and maximum intensity projections generated from deconvolved stacks. All other in situ stained embryos were photographed either in whole mount on an Olympus SZX12 stereomicroscope equipped with a QImaging Micropublisher 5.0 RTV camera or flat mounted and photographed on a Zeiss Axioskop 2 equipped with a Retiga SRV camera (QImaging). Images were processed in Photoshop and ImageJ. Intensity profile plots were measured in ImageJ in two rectangular boxes on either side of the notochord that were 50 µm wide and extended 315 µm from the posterior end of the notochord. Segmentation defects and myotome number were scored in embryos stained with a cb1045 riboprobe as described in [74] and [17], respectively.
Somitogenesis period was determined by multiple-embryo time-lapse imaging as previously described [32],[33]. Briefly, somitogenesis movies were analyzed visually by annotating the time of boundary formation for somites 2–17 in her1 and hes6 mutants or somites 2–10 in her7 mutants. Somitogenesis period in minutes was calculated in Microsoft Excel for each embryo individually from the slope of the linear fit to the data points. Period measurements were subsequently normalized by dividing through the mean of the period measurements of the wildtype control population in each experiment, yielding the non-dimensional normalized somitogenesis period. The 95% confidence intervals were calculated in Microsoft Excel, and Mann-Whitney test was used to assess significance.
Our mathematical model is based on delay differential equations describing the dynamics of Her1, Hes6, and Her7 proteins and all their possible dimers. Using an equilibrium approximation that assumes that Hes/Her dimerization is faster than the other processes in the system [22], we have reduced this model to only three equations describing the change of concentrations with time s, of monomers h1, h6, and h7:where the production delays associated with synthesis of Her1 and Her7 are τ1 and τ7, respectively; the production rates of Her1, Her7, and Hes6 are κ1, κ2, and κ6, respectively; n is a phenomenological Hill coefficient describing effective cooperativity arising from multiple binding sites for Hes/Her dimers in each of the promoters; and δ is an effective dimer-mediated degradation rate. Details of the derivation, parameterization, and simulation of the model are given in the Text S1.
Recombinant full-length Hes6 protein was expressed in bacteria, purified, and used to immunize mice according to standard procedures. Complete antisera were screened for Hes6 binding, and one mouse was selected for boosting and production of hybridoma cell lines. For our experiments we used a mixture of the purified supernatants from two hybridoma cell lines that both gave specific immunostaining in the PSM of 10-somite stage embryos. Immunostaining was performed according to standard procedures. Briefly, embryos were fixed for 2 h in 2% paraformaldehyde, dechorionated, and permeabilized for several hours in methanol. Following rehydration, embryos were blocked in 10% BSA/1% DMSO for 4 h and incubated with primary antibodies at a concentration of 0.8 µg/ml at 4°C for several hours. Secondary antibodies were peroxidase-coupled anti-mouse IgG at 1∶1,000 dilution. Color development was performed with Cy3-coupled tyramide according to the manufacturer's recommendations (Perkin Elmer). Embryos were cleared with methanol, flat mounted, and imaged on a Zeiss Axioskop 2 equipped with a Retiga SRV camera (QImaging) or on a Zeiss LSM510 confocal system. Images were processed in Photoshop and ImageJ.
|
10.1371/journal.pntd.0004428 | Modeling Relapsing Disease Dynamics in a Host-Vector Community | Vector-borne diseases represent a threat to human and wildlife populations and mathematical models provide a means to understand and control epidemics involved in complex host-vector systems. The disease model studied here is a host-vector system with a relapsing class of host individuals, used to investigate tick-borne relapsing fever (TBRF). Equilibrium analysis is performed for models with increasing numbers of relapses and multiple hosts and the disease reproduction number, R0, is generalized to establish relationships with parameters that would result in the elimination of the disease. We show that host relapses in a single competent host-vector system is needed to maintain an endemic state. We show that the addition of an incompetent second host with no relapses increases the number of relapses needed for maintaining the pathogen in the first competent host system. Further, coupling of the system with hosts of differing competencies will always reduce R0, making it more difficult for the system to reach an endemic state.
| An important development in the study of infectious diseases is the application of mathematical models to understand the interplay between various factors that determine epidemiological processes. Vector-borne diseases are additionally complex with interactions between multiple host and vector species. Understanding the transmission dynamics of vector-borne diseases is an important step towards controlling outbreaks and mitigating human infection risk. Identifying the biotic and abiotic interactions and mechanisms that may contribute to disease emergence, establishment and persistence is necessary for assessing current and future disease risk, as well as developing effective control strategies. Tick-borne relapsing fever (TBRF) is found around the world and is caused by several species of Borrelia spirochetes, which are vectored by soft ticks of the genus Ornithodoros. TBRF is a cryptic disease that still causes significant morbidity and mortality, especially in some African countries. Here, we develop and adapt a compartmentalized mathematical model (SIR) with a relapsing component to investigate the dynamics of TBRF.
| An important development in the study of infectious diseases is the application of mathematical models to understand the interplay between various factors that determine epidemiological processes. Many systems show a rich variety of dynamics that arise from nonlinear interactions (due to the mixing of different infectious populations) or temporal forcing (caused by changes in the average contact rate) [1]. Vector-borne diseases are additionally complex with interactions between multiple host and vector species [2–4]. Compartmental models, such as susceptible, infectious, and removed models (SIR) [5], have been applied to many disease systems in an effort to examine system dynamics. In these epidemic models, susceptible individuals pass into the infective class, from which they transition to the removed class. For some diseases, recovered individuals may relapse with a reactivation of infection and revert back to an infective class. An example of such a system is found in van den Driessche et al. [6], which included a relapsing rate between the susceptible and the same infected compartment. Adding additional infected compartments simulates disease systems in which there is a relapsing component, leading to a prolonged infectious period, presumed to be important to disease persistence. To our knowledge, the addition of a relapsing component has not been applied to a host-vector system. Noteworthy vector-borne relapsing diseases include tick-borne relapsing fever (TBRF) and malaria.
An advantage of these types of models is the ability to vary parameters, while monitoring the overall effect on the disease system, allowing for the exploration of characteristics of the system that may not be well understood. Tick-borne relapsing fever (TBRF) is a cryptic disease that still causes significant morbidity and mortality worldwide, especially in African countries [7–10]. TBRF is a vector-borne zoonotic disease endemic to central Asia, Africa, and the Americas [11], and is caused by infection with Borrelia spirochetes. All but one species of relapsing fever spirochetes are vectored by soft ticks (Ornithodoros spp.) [12]. Relapsing fever is characterized by recurring febrile episodes and generalized symptoms including headache, chills, myalgia, nausea, and vomiting[13]. There is a rapid onset of disease symptoms, with a febrile episode lasting 3–6 days, after which symptoms subside, only to return in 7–10 days. Symptoms are associated with large number of spirochetes present in the bloodstream (spirochetemia), and subside when the host generates an antibody response against the variable major proteins (Vmps). The Vmps are involved with antigenic variation, and relapsing fever Borrelia produce a new variant during infection, subsequently attaining high densities [14, 15]. Little is known regarding the number of relapses in natural hosts, but studies have shown a range from 1 to 5 in experimentally infected animals [16]. In humans, there is an average relapse rate of three febrile episodes without treatment, but up to 13 relapses have been observed [17].
Ornithodoros spp. ticks are long-lived, fast feeding vectors that are known to live > 10 years, and have been shown to survive for up to five years without feeding [18]. Ornithodoros ticks are nidicolous ticks that rarely leave the confines of the host nest or burrow and are able to obtain a blood meal and detach from the host in < 90 min. Additionally, soft ticks only obtain a blood meal about once every 3 months; even when presented with the opportunity to feed daily. Ornithodoros ticks require several months between feedings and can survive years between feeding. The longevity of these ticks means that they outlive their rodent hosts, affording the potential to infect several cohorts of rodents over the course of the tick lifespan. Once infected ticks remain infected and infectious for the duration of their lifespan.
Here, we model TBRF caused by infection with B. hermsii and vectored by O. hermsi. We parameterize the model with field-derived values from hosts on Wild Horse Island in Montana and a single genomic group I (GGI) strain of B. hermsii. The overall goal of this study was to develop a SIR model using TBRF dynamics to describe a host-vector system with a relapsing class of host individuals. First, using specific information from a TBRF system located on Wild Horse Island, MT, a model for the dynamics of a single host-vector interaction was developed. For models with increasing numbers of relapses and multiple hosts, equilibrium analysis was performed and R0 was generalized. Parameter values were considered in the model to provide theoretical criteria for population stability and to determine the parameters that would result in elimination of the disease. Finally, single and coupled host-vector systems were explored, focusing on the addition of less competent hosts and the number of relapses needed in order to maintain an endemic equilibrium. We use the model to ask several important biological questions pertaining to the TBRF system determining effect adding relapsing classes has on pathogen persistence and the effect of multiple host species with varying competency for acquiring and transmitting B. hermsii.
We sought to develop a model based on disease dynamics on Wild Horse Island (WHI), Flathead Lake, Lake County, MT. WHI is the largest island (~2100 acres) on Flathead Lake and like other islands on the lake has a limited diversity of rodent host species. WHI is almost exclusively inhabited by deer mice (Peromyscus maniculatus) and pine squirrels (Tamiasciurus hudsonicus) as the terrestrial rodents and provided an important opportunity to develop and parameterize a model including only two hosts. Although there are two genomic groups (GGI and GGII) of B. hermsii present on WHI, we parameterize the model using estimates for only GGI B. hermsii, as host competency experiments have primarily been performed with GGI B. hermsii [16].
A key assumption for host-vector disease modeling is the definition of the transmission term, which represents the contact between hosts and vectors. The formulation of the transmission term affects the reproduction number, R0, which is a central predictor of disease systems [19]. For host-vector disease models, the transmission term includes the vector biting rate. This rate controls the pathogen transmission both from the vector-to-host and from the host-to-vector. The TBRF model follows frequency-dependent transmission assumptions through the biting rate, since a blood meal is only required approximately once every three months regardless of the host population density. Following this framework, hosts would likely experience an increasing number of bites as the vector population increased.
Given a mathematical model for disease spread, R0 is an essential summary parameter. It is defined as the average number of secondary infections produced when one infected individual is introduced into a completely susceptible host population [20]. When R0 < 1, the disease free equilibrium (DFE) at which the population remains in the absence of disease is locally asymptotically stable. However, if R0 > 1, then the DFE is unstable and invasion is always possible (see [21]) and a new endemic equilibrium (EE) exists. For this study, R0 was extracted following the methodology developed in van den Driessche et al. [22] (see also [23, 24]) for general compartmental disease models, which can be extended to more complicated host-vector disease systems [25, 26].
Specific parameter values for this system have not yet been determined, but can be estimated from similar studies and from data collected on O. hermsi from laboratory experiments. The units of the rates are individuals per month. Table 1 summarizes the notation for all system parameters and variables. See Table 2 for specific model values used in all of the host-vector models. Note that parameters denoted with additional subscripts of ps and dm refers to values specific to the pine squirrel and deer mouse host-vector systems, respectively.
The birth rates for host and vector are each set to a constant value (β and βv, respectively) and the compartmental death rates (for host and vector) are identical and set equal to birth rate. Then the death rates must be
μs=μi1=⋯=μij=μr=βj+2
(1)
and
μsv=μiv=βv2.
(2)
The growth rate of pine squirrels (βps = 0.33 individuals per month) is an average of the rates found in the literature, i.e., four individuals per litter at 1 litter per year [27]. The growth rate of deer mice is also taken from average estimates from the literature; we estimate growth rate based on an average of three litters per year and four young per litter, (βdm = 1 individual per month) [28]. The death rates are determined from Eq (1), which depends on the number of relapses in the system. For example, for a pine squirrel host-vector system with one relapse, all death rates would be 0.0825. Life history dynamics of O. hermsi are not well documented and virtually nothing is known about the reproductive behavior and survival of these ticks in nature. Conservative estimates from the laboratory show that soft-bodied ticks lay on average five clutches over their approximately 10 year lifespan with roughly 50 eggs per clutch [29] (T. Schwan personal communication). Thus, the vector birth rate is βv = 2.08 individuals per month. Following Eq (2), we get death rates of μsv = μiv = 1.04 for the vector compartments.
The rate at which an individual transitions among infected compartments and to the removed compartment is fixed and is assumed to be the same for all compartments. As more infected compartments are added to the system, the corresponding constant rates are γ = α = α1 = … = αj-1, for j infected compartments. Field parameter estimates have not yet been made for these transition rates (i.e., relapse and recovery rates). Laboratory results from three pine squirrels indicate a transition rate of approximately 4.35 individuals per month for a single compartment (Burgdorfer and Mavros 1970). Then γ = α = α1 = … = αj-1 = 4.35.
Ticks are assumed to bite a host once every three months (i.e., f = 0.33). Competency values are between 0 and 1 and thus modify the transmission rate of the infection by multiplying the biting rate. Burgdorfer and Mavros [16] observed a high competency in pine squirrels successfully infecting 3/3 animals by tick bite or injecting them with triturated ticks. Using the same methods, they challenged deer mice with B. hermsii and were unsuccessful in establishing infection. Thus, we used competency values cv = 0.95 for the probability of transmission for vectors, cps = 0.90 for pine squirrels, and cdm = 0.10 for deer mice.
The carrying capacity for the pine squirrel and deer mouse system is determined specifically for WHI. On WHI there are approximately 425 ha of suitable habitat for pine squirrels with up to a maximum of 2 individuals per suitable habitat patch and approximately 850 ha of suitable deer mouse habitat with a conservative estimate of just less than 12 mice per ha [28]. Thus, the total number of pine squirrels (Nps) is estimated at 850 and total number of deer mice (Ndm) is estimated at 10,000. The soft bodied tick population (Nv) is virtually unknown, however, we assume that they are limited to the nests of their hosts. Initial field collections have found as many as 14 ticks in one nest on the island [30]; other collection efforts show > 300 ticks can be collected from a single nest or snag [31]. Because the estimates of ticks per nest vary largely between our limited collection on WHI and the literature we chose a conservative number of ticks. We estimate that each squirrel has less than one nest (because of juveniles in the system), and each nest is inhabited by 14 ticks. We found no ticks in nest material collected from deer mice, however, nest material collected during the human outbreak in 2002 yielded 14 O. hermsi; the carcasses of two deer mice were found nearby and American Robins (Turdus migratorius) had been nesting there [30]. Thus it is nearly impossible to estimate the average number of ticks in a deer mouse nest, or if in fact they are coming in contact with ticks while visiting other nests. We used an estimate of 20,000 total ticks on the island split equally among host systems. We chose a conservative estimate of 1% of all ticks are infected, as none of 12 of 14 field collected ticks were found to be infected [30]. Thus, we used Sv(0) = 9,900 ticks for the single host-vector system and Sv(0) = 19,800 ticks for the coupled host-vector system.
A model for the dynamics of TBRF in a single host-vector system is considered (see Fig 1A). The following assumptions are used to establish a model that is appropriate for the WHI TBRF system for the host pine squirrel and soft tick vector, O. hermsi. (1) The only sources of infection occur between the bite of an infective vector and susceptible host and between a bite of a susceptible vector and infective host (i.e. there are no horizontal or vertical transmission events). (2) The vector becomes infected and infectious for life immediately upon biting an infectious host. (3) The transmission terms are frequency-dependent through the biting rate, f. (4) The hosts relapse to different infected compartments (i.e. different serotypes within the hosts caused by antigenic variation) at rate α and recover from the disease at rate γ. (5) Though mortality rates are noted to differ for each compartment, we assume a constant total population for both hosts and vectors (N and Nv, respectively). Thus, recruitment (or birth) and the sum of the removal (or death) rates from each compartment must be equal (Eqs 1 and 2).
The generalized system for the infection dynamics in a single host-vector system with j—1 relapsing rates for j = 1 infected compartments describes the number of susceptible hosts S(t), infectious hosts Ik(t), removed hosts R(t), susceptible vectors Sv(t), and infected vectors Iv(t), where the total host population is
N=S+∑k=1jIk+R
and the total vector population is Nv = Sv + Iv (see Fig 1A for a compartmental diagram and Table 1 for parameter definitions). The equations are
Host equations:
S•=βS−fcvIvSN−μsSI•1=fcvIvSN−α1I1−μi1I1I•2=α1I1−α2I2−μi2I2...I•j−1=αj−2Ij−2−αj−1Ij−1−μi(j−1)Ij−1I•j=αj−1Ij−1−γjIj−μijIjR•j=γjIj−μrR.
(3)
Vector equations:
S•v=βvSv−fcSvN∑i=1jIi−μsvSvI•v=fcSvN∑i=1jIi−μivIv.
(4)
To evaluate the invasiveness of the disease in this system, we extracted R0 following the techniques developed by van den Driessche and Watmough [22] by sequentially adding infected compartments (see S1 for equilibrium analysis and derivations). The form of R0 was then inferred for j—1 relapsing rates between j infected compartments as
R0=fccvμivSv(0)N(0)[1α1+μi1[1+α1α2+μi2[⋯[1+αj−1γ+μij]]]].
(5)
R0 is directly proportional to the biting rate (f), competency values (c and cv), and the ratio of initial vectors to initial hosts
(Sv(0)N(0))
and inversely proportional to the vector death rate (μiv) and the rate that moves individuals out of the infected compartments (α α1,…., αj-1, μi1, …, μij, and γ). In addition, a pattern emerges as more infected compartments are added: a nesting sequence of terms that increase the value of R0 and potentially contribute to a change in stability of the DFE. To illustrate this concept, we used the pine squirrel host parameters (Table 2) for increasing number of infected compartments and plotted R0. R0 crosses 1 at between j = 4 and j = 5 infected compartments (i.e., four relapses; Fig 2).
Here, the single host-vector model is expanded to include two hosts, namely pine squirrels and deer mice. Fig 1B is a compartmental diagram for the two systems with no relapses. The first host-vector system (Sps, I1,ps, Rrs, Sv,ps, Iv,ps) is coupled with the second system (Sdm, I1,dm, Rdm, Sv,dm, Iv,dm) through ticks biting either host species, with parameter f, and is further controlled by competency values of either the ticks (cv) or hosts (cps or cdm for pine squirrel and deer mice, respectively). Transmission occurs through three mechanisms: 1) fcv, which is the biting rate modified by the tick competency through which an infected tick bites a host from each system, 2) fcps, which is the biting rate modified by the pine squirrel competency in that a susceptible tick bites an infected pine squirrel, and 3) fcdm, which is the biting rate modified by the deer mouse competency, such that a susceptible tick bites an infected deer mouse. The parameters remain as in the single host vector system, denoted with additional subscripts to represent the respective host-vector system (either ps or dm), and are explained in Tables 1–2.
The generalized system for the infection dynamics in a coupled host-vector system with j—1 relapsing rates for j = 1 infected compartments describes the pine squirrel system with the number of susceptible hosts Sps(t), infectious hosts Ik,ps(t), and removed hosts Rps(t). The total pine squirrel host population is
Nps=Sps+∑k=1jIk,ps+Rps.
Likewise, the deer mouse host system consists of susceptible hosts Sdm(t), infectious hosts Ik,dm(t), and removed hosts Rdm(t) with a total deer mouse host population of
Ndm=Sdm+∑k=1jIk,dm+Rdm.
The vector compartments are susceptible vectors Sv(t), infected vectors Iv(t) and a total vector population of Nv = Sv + Iv. The equations are
Pine squirrel host system:
S•ps=βSps−fcvIvSpsNps−μs,psSpsI•1,ps=fcvIvSpsNps−α1,psI1,ps−μi1,psI1,psI•2,ps=α1,psI1,ps−α2,psI2,ps−μi2,psI2,ps⋮I•j−1,ps=αj−2,psIj−2,ps−αj−1,psIj−1,ps−μi(j−1),psIj−1,psI•j,ps=αj−1,psIj−1,ps−γj,psIj,ps−μij,psIj,psR•j,ps=γj,psIj,ps−μr,psRps.
(6)
Deer mouse host system:
S•dm=βSdm−fcvIvSdmNdm−μs,dmSdmI•1,dm=fcvIvSdmNdm−α1,dmI1,dm−μi1,dmI1,dmI•2,dm=α1,dmI1,dm−α2,dmI2,dm−μi2,dmI2,dm⋮I•j−1,dm=αj−2,dmIj−2,dm−αj−1,dmIj−1,dm−μi(j−1),dmIj−1,dmI•j,dm=αj−1,dmIj−1,dm−γj,dmIj,dm−μij,dmIj,dmR•j,dm=γj,dmIj,dm−μr,dmRdm.
(7)
Coupled vector system:
S•v=βvSv−fcpsSvNps∑i=1jIi,ps−fcdmSvNdm∑i=1jIi,dm−μsvSvI•v=fcpsSvNps∑i=1jIi,ps+fcdmSvNdm∑i=1jIi,dm−μivIv.
(8)
As with the single host-vector system, we performed equilibrium analysis (S2) and the form of R0 was inferred for j—1 relapsing rates between j infected compartments.
Where
PS=cpsNps(0)[1(∝1,ps+μi1,ps)[1+∝1,psα2,ps+μi2,ps[⋯[1+∝j−1,psγ+μij,ps]⋯]]]andDM=cdmNdm(0)[1(∝1,dm+μi1,dm)[1+∝1,dmα2,dm+μi2,dm[⋯[1+∝j−1,dmγ+μij,dm]⋯]]].
(10)
From the coupled host-vector system it is apparent that R0 has the additional dependency for both the host competency values (cps and cdm). Since competency values are probabilities between 0 and 1, then they will always decrease the value of R0 as they decrease. Like the single host-vector system, a pattern emerges as more infected compartments are added to each host system (Eqs 9 and 10): a nested sequence of terms that increase the value of R0 and potentially contribute to a change in stability of the DFE. To compare the results of the number of relapses needed for R0 > 1 in the coupled host-vector system with the single host-vector, we added an incompetent deer mouse host system (cdm = 0.2) and increased the number of relapses in a pine squirrel host system until R0 reached 1. R0 crosses 1 at between j = 7 and j = 8 infected compartments (seven relapses; Fig 3).
Incorporating a relapsing component into a host-vector SIR modeling framework represents a step towards a better understanding and representation of complex disease systems. We investigated the disease dynamics of TBRF and used the model to better understand the underlying dynamics and interactions among spirochetes, rodent hosts, and tick vectors that contribute to pathogen persistence. Disease models were presented that describes (1) a single host-vector system with a single relapsing class of host individuals, and generalized to j-1 relapsing host classes and (2) a coupled host-vector model generalized as above to j -1 relapsing host classes. Analytical techniques allowed for the generalization of R0 with increasing numbers of relapses, and parameters were identified that affect the elimination or persistence of the pathogen (e.g., biting rates, competency values, and population numbers).
In the single host-vector system, R0 is directly proportional to the biting rate (f), competency values (c and cv), and the ratio of initial vectors to initial hosts
(Sv(0)N(0)).
An inverse relationship exists between R0 and the vector death rate (μiv) and the rate that moves individuals out of the infected compartments (α1,…., αj-1, μi1, …, μij, and γ). When additional relapsing classes are added to the system, R0 always increases because of the addition of a nested sequence of terms that is always > 1 (Eq 5). The coupled host-vector system has similar dependencies with additional interesting dynamics that may be very important to understanding pathogen persistence and host diversity. Coupling of the system with hosts of lower competencies will always reduce R0 (Eqs 9 and 10). As the number of incompetent hosts available as blood meals for infected ticks increases, an effect comparable to the dilution effect occurs and R0 always decreases, leading to DFE. The dilution effect states that in the presence of a second, less competent species, competent host-vector encounters leading to transmission events may be replaced by incompetent host-vector encounters that do not end in a pathogen transmission event, thus decreasing R0 [3, 4].
The model presented here addresses the presence of multiple hosts with varying competencies and a single pathogen, however, the model can be extended to address not only differences in host species diversity but also the presence of > 1 pathogen strain. The genetics of B. hermsii have been well characterized and isolates have been shown to fall into two distinct genomic groups, referred to as genomic group I and II (GGI and GGII) [32, 33]. The presence of both genomic groups of B. hermsii has been documented on WHI, while only GGII B. hermsii has been found to date on the mainland around Flathead Lake where host species diversity is greater than that of the WHI.
Field investigations of rodents on WHI confirmed infection in a single deer mouse (Peromyscus maniculatus) infected with GGII B. hermsii (Johnson et al. In. Prep.). This prompted a laboratory experiment in which we infected deer mice with both GGI and GGII B. hermsii and monitored them for infection. We challenged deer mice with infection via needle inoculation and infectious tick bite and observed that deer mice show no susceptibility to GGI but are highly susceptible to GGII spirochetes (Johnson et al. In. Prep.). These findings were in contrast with Burgdorfer and Mavros [16] who were unable to establish infection in deer mice, however, they used infected ticks from a TBRF outbreak near Spokane, WA, U.S.A., which resulted in isolation of GGI B. hermsii.
The coupled system presented here could be used to examine the effects of not only host species with varying competencies, but also diverse host communities in the presence of B. hermsii GGI and GGII. The presence of both genomic groups simultaneously may result in a dampening of the dilution effect if GGII is able to infect a diverse array of host species even though GGI is more species limited. Rodent trapping and tick collection on WHI showed one squirrel and one tick infected with GGI and three squirrels infected with GGII. On WHI, 95% of all pine squirrels captured were seropositive for relapsing fever spirochetes while only 4% of deer mice possessed antibodies (Johnson et al. In Prep.). All infected individuals at mainland sites with diverse host species were infected with GGII spirochetes (Johnson et al. In Prep.).
Although there are limitations to the model presented here, the model is an important first step in understanding a relapsing host-vector disease system. All known complexities of the system were not addressed at this time, including incorporation of GGII strains of B. hermsii which can infect deer mice and possibly a wide range of other potential hosts (Johnson et al. In Prep.). Although there is conflicting evidence at the rate which transovarial transmission of B. hermsii occurs in O. hermsi, we can see from the R0 calculation that Iv does not appear in the equation and therefore will have little impact on disease persistence in the presence of hosts. However, the existence of transovarial transmission may provide insight into the implication of O. hermsi serving as the reservoir for B. hermsii, i.e., the ability to maintain infectious ticks in a prolonged absence of competent hosts and/or hosts in general. Additionally, the model could be used to explore drivers in the host and vector communities and prevention/intervention strategies may be explored to identify the effectiveness of host control versus vector control. Further, this may provide insight into human protective measures and the effectiveness of control strategies such as host vaccination; simulations could be run to assess the efficacy of control programs such as vaccination regimes and vector control.
Ecological factors including biotic and abiotic interactions may play a primary role in the emergence and persistence of infectious diseases [34–39]. Understanding the complete epidemiology of a disease is crucial to advancing the ability to predict and control outbreaks in human and wildlife populations, however, this is rarely an attainable goal. Sonenshine [40] outlines the sequence of steps typically undertaken when attempting to understand the epidemiology of a given system. The pathway typically begins with the identification of a clinical syndrome, followed by discovery of the causative disease agent, and then the identification of the source of the agent in nature. The final step includes investigating the often complex biology and ecology of the hosts and/or vectors involved. Given the difficulty frequently encountered when attempting to study a disease in nature, the last step is often the most difficult. The application of advanced modeling techniques to poorly understood systems is often the only way to begin to understand the drivers of these systems.
The ecological dynamics of relapsing fever systems around the world are poorly understood. Here we use a North American system of relapsing fever caused by B. hermsii; however, information gathered from this modeling exercise can be applied to TBRF systems around the world. TBRF remains a major public health threat in Africa [41]. In addition to other TBRF systems, the ideas presented here may provide the groundwork for relapsing components to be included in other disease systems with greater public health implications such as malaria.
|
10.1371/journal.ppat.1005829 | The G1/S Specific Cyclin D2 Is a Regulator of HIV-1 Restriction in Non-proliferating Cells | Macrophages are a heterogeneous cell population strongly influenced by differentiation stimuli that become susceptible to HIV-1 infection after inactivation of the restriction factor SAMHD1 by cyclin-dependent kinases (CDK). Here, we have used primary human monocyte-derived macrophages differentiated through different stimuli to evaluate macrophage heterogeneity on cell activation and proliferation and susceptibility to HIV-1 infection. Stimulation of monocytes with GM-CSF induces a non-proliferating macrophage population highly restrictive to HIV-1 infection, characterized by the upregulation of the G1/S-specific cyclin D2, known to control early steps of cell cycle progression. Knockdown of cyclin D2, enhances HIV-1 replication in GM-CSF macrophages through inactivation of SAMHD1 restriction factor by phosphorylation. Co-immunoprecipitation experiments show that cyclin D2 forms a complex with CDK4 and p21, a factor known to restrict HIV-1 replication by affecting the function of the downstream cascade that leads to SAMHD1 deactivation. Thus, we demonstrate that cyclin D2 acts as regulator of cell cycle proteins affecting SAMHD1-mediated HIV-1 restriction in non-proliferating macrophages.
| Macrophages are a heterogeneous population of immune cells that provide crucial innate immune defense against pathogens, including HIV-1. The molecular biology of HIV-1 infection in macrophages is influenced by the presence of the host cell restriction factor SAMHD1, which is regulated by phosphorylation by cyclin dependent kinases, the catalytic proteins responsible for cell cycle progression. This study shows that differentiation stimuli strongly influence macrophage cell cycle and proliferation characteristics as well as susceptibility to HIV-1 infection through modulation of SAMHD1 activation. We have identified cyclin D2 as the key step controlling susceptibility to HIV-1 infection by modulation of the signaling pathway leading to SAMHD1 phosphorylation. We show that a complex formed by cyclin D2-CDK4-p21 in GM-CSF macrophages is responsible for the lack of the active CDK, which phosphorylates SAMHD1. This situation is reversed in the absence of cyclin D2, leading to the activation of CDKs and subsequent phosphorylation of its substrates, including SAMHD1. Thus, we propose that the differential expression of the G1/S-specific cyclin D2 controls the HIV-1 restriction pathway in primary macrophages.
| Macrophages are a highly heterogeneous cell population that plays a prominent role in innate immune system as key effector cells for the elimination of pathogens, infected cells and cancer cells [1, 2]. Macrophages also play an essential role in maintaining tissue homeostasis by supporting tissue development and repairing damaged tissue architecture [1, 3]. Macrophage differentiation from monocytes occurs in the tissue in concomitance with the acquisition of a functional phenotype that depends on microenvironmental signals, accounting for the wide and apparently opposed variety of macrophage functions [4, 5].
Macrophages, as well as other myeloid lineage cells, become susceptible to HIV-1 infection after degradation or inactivation of the restriction factor SAMHD1, a triphosphohydrolase enzyme that controls the intracellular level of dNTPs [6–9]. Phosphorylation of SAMHD1 by cyclin dependent kinases (CDK) has been strongly associated with inactivation of the virus restriction mechanism, providing an association between virus replication and cell proliferation [10–12]. The activity of CDK is regulated by the binding of cyclins, a family of proteins characterized by a periodic, cell-cycle dependent pattern of expression [13, 14]. Cyclin-CDK complexes govern cell cycle progression and proliferation of mammalian cells and thus, pinpoint the specific time in which an event occurs during the cell cycle [13, 14]. We and others have shown that the complex cyclin D3-CDK6 acting upstream of CDK2 controls SAMHD1 phosphorylation and function in primary lymphocytes and macrophages [11, 15–17]. Cyclin-CDK function is also controlled by cyclin dependent kinase inhibitors (CDKIs) that generally act as negative regulators of the cell cycle by binding to CDKs and inhibiting their kinase activity [18]. Of particular importance is p21/waf1, a G1/S phase CDKI, that may also control HIV-1 replication through SAMHD1 [19, 20].
D-type cyclins (cyclins D1, D2 and D3) are regarded as essential links between cell environment and the core cell cycle machinery. D-type cyclins drive cells through the G1 restriction point and into the S phase, after which growth factor stimulation is no longer essential to complete cell division [21]. D-type cyclins share the capacity to activate both CDK4 and CDK6 [14]. Studies on single, double and triple cyclin D knockout mice revealed that D-type cyclin complexes have redundant functions. However, different D-type cyclins exhibit distinct expression patterns depending on the cell type, indicating that each D-type cyclin has essential functions in particular settings, as suggested by the narrow and tissue-specific phenotypes of the knockout mice (reviewed in [21]).
Here, we have used primary human monocyte-derived macrophages (MDMs) differentiated through different stimuli to evaluate macrophage heterogeneity on cell activation and proliferation, characteristics that influence gene and protein expression patterns and determine susceptibility to HIV-1 infection. The comparative study has led to the identification and characterization of a cell cycle dependent pathway that restricts HIV-1 infection in primary macrophages. These non-proliferating macrophage population is characterized by a high expression of the G1/S-specific cyclin D2. Cyclin D2 acts through the binding to CDK4 and p21 in GM-CSF macrophages, a complex which is responsible for the lack of the active CDK that phosphorylates SAMHD1. Data from mouse peritoneal macrophages confirmed the existence of cyclin D2 expressing macrophages in vivo, further supporting the key role of cyclin D2.
Primary monocyte-derived macrophages were differentiated either with M-CSF or GM-CSF, with the aim to characterize differences in cell activation and proliferation patterns and susceptibility to HIV-1 infection. Differentiated macrophages displayed different morphological characteristics dependent on the differentiation stimuli, but no significant differences in cell surface antigen expression or HIV receptor and co-receptors were observed (S1A Fig). Interestingly, cell proliferation and cell cycle patterns were significantly different between macrophage types, i.e., M-CSF macrophages proliferated at higher rates than GM-CSF measured by intracellular Ki67 staining (Fig 1A, 7% vs. 0.5% of Ki67+ cells in M-CSF and GM-CSF macrophages respectively, p = 0.02). Similar results were obtained analyzing cell cycle profile by DNA and RNA content staining (Fig 1B), showing higher percentage of cells in S/G2M stage in M-CSF than GM-CSF macrophages.
Susceptibility to HIV-1 infection was also significantly different, being M-CSF macrophages in average roughly 10-fold more susceptible to HIV-1 infection (Fig 1C, left panels and 1D, p = 0.0072), which correlated with higher dNTP levels in cycling M-CSF macrophages compared to GM-CSF as previously reported (S1B Fig and [22, 23]). These results point towards the established link between cell cycle progression and susceptibility to infection as the determinant of the differences observed between macrophage types, a process where SAMHD1 restriction is central [11]. Indeed, degradation of SAMHD1 by HIV-2 Vpx increased HIV replication in both macrophage types, minimizing the initial differences in infection (Fig 1C, right panels, and 1D, p = 0.12). No differences were found in antiviral activity of drugs either targeting viral reverse transcription (AZT or nevirapine) or the cell cycle inhibitor palbociblib [24] (PD, specifically targeting CDK4/6 and inhibiting SAMHD1 phosphorylation) between M-CSF and GM-CSF differentiated macrophages (Fig 1E and 1F). Moreover, after SAMHD1 degradation by HIV-2 Vpx, decreased antiviral potency of AZT and a complete loss of PD antiviral activity were observed, suggesting that cellular events leading to SAMHD1-mediated viral restriction were similar in both types of macrophages [11, 15]. Similar results were obtained when culturing macrophages in the presence of human serum, indicating that the differentiation stimuli induced by the cytokine determines the macrophage phenotype (S2 Fig).
To further investigate the molecular determinants of the observed differences between macrophages types, expression of cell cycle genes implicated in SAMHD1 control were evaluated (Fig 2A). No major gene expression differences were observed, except for a significant upregulation of D-type cyclins (CCND1, 2-fold, p = 0.0006; CCND2, 40-fold, p = 0.0004; and CCND3, 3-fold p = 0.0002) and the CDK inhibitor p21 (CDKN1A, 4-fold, p = 0.038) in GM-CSF macrophages (Fig 2A). Analysis of protein expression confirmed the upregulation of cyclin D2, cyclin D3 and p21 in GM-CSF macrophages, and revealed a clear downregulation of CDK protein levels (CDK1, CDK2 and CDK6 but not of CDK4) and the negative cell cycle regulator p27 (Fig 2B). As expected, no differences were found in SAMHD1 expression but a change in SAMHD1 activation was observed, being only phosphorylated and partially inactivated in M-CSF macrophages (Fig 2B, upper panels). Importantly, mouse peritoneal macrophages showed a similar expression pattern than that observed in GM-CSF macrophages, suggesting the existence of cyclin D2 expressing macrophages in vivo (S3 Fig). From a functional point of view, stimulation with LPS resulted in the upregulation of IFNB1, IL-10 and CCL-2 production in both types of macrophages, albeit basal expression levels were different in M-CSF and GM-CSF as reported elsewhere [25, 26] (Fig 2C). These results demonstrate that differentiation stimuli strongly impact the cell cycle profile of primary macrophages, and consequently their capacity to support HIV-1 replication, that may be in part determined by differences in the restriction factor SAMHD1 activation. Differential expression of D-type cyclins, and specially cyclin D2 may represent key regulatory proteins that shape the distinct cell cycle profile and affect HIV-1 restriction.
The role of D-type cyclins in macrophage differentiation and HIV-1 susceptibility was further evaluated by RNA interference. Effective and specific downregulation of cyclin D2 and cyclin D3 was achieved at both mRNA and protein level in M-CSF and GM-CSF macrophages (Fig 3A and 3B). Higher expression of CCND2 in the GM-CSF macrophage population compared to M-CSF, was again clearly observed in all conditions tested (Fig 3A, left panel, and 3B). SAMHD1 expression was not affected by D-type cyclin inhibition, as measured by mRNA level (S4A Fig) or protein expression (Fig 3B). However, different effects on SAMHD1 phosphorylation were observed depending on the targeted cyclin and the differentiation stimuli. As previously reported [16], cyclin D3 knockdown resulted in the abolishment of SAMHD1 phosphorylation in M-CSF macrophages compared to a non-targeting siRNA (Fig 3B, lanes 1–3), an effect that could not be evaluated in GM-CSF macrophages due to low expression levels of phosphorylated SAMHD1. On the contrary, cyclin D2 knockdown resulted in increased SAMHD1 phosphorylation in the GM-CSF macrophage population (Fig 3B, lanes 4–6). Similarly, cyclin D3 but not cyclin D2 knockdown was associated with lower cell proliferation measured as percentage of Ki67 positive macrophages, an effect that was evident in proliferating M-CSF macrophages, but difficult to evaluate in non-proliferating GM-CSF macrophages (Fig 3C).
siRNA-treated macrophages were tested for their capacity to support HIV-1 replication. HIV-1 replication was inhibited after cyclin D3 knockdown in M-CSF macrophages (roughly 60% inhibition, p = 0.0095), whereas inhibition of cyclin D2 expression did not have any effect (Fig 3D, white bars). Conversely, in GM-CSF macrophages cyclin D2 downregulation led to a significant increase of HIV-1 replication (3-fold increase, p = 0.03), but no effect was observed in cyclin D3 knockdown macrophages (Fig 3D, black bars). These results demonstrated different roles for cyclin D2 and D3 at controlling cell proliferation and HIV-1 infection.
To determine the molecular bases of cyclin D2 effect on HIV-1 replication in GM-CSF macrophages, most common protein interactions for cyclin D2 were identified by database searches [27, 28]. p21, CDK4 and CDK6 were found as the most common proteins bound to cyclin D2. Cyclin D2 is known to form a complex with CDK4 and CDK6 where it functions as a regulatory subunit of both CDK at G1/S transition [18]. However, CDK6 expression could not be detected in GM-CSF macrophages (Fig 2B), therefore, further characterization was centered in D-type cyclins and their interactors p21 and CDK4.
RNA interference was used to effectively and specifically downregulate cyclin D2, cyclin D3, p21 and CDK4 expression in GM-CSF macrophages (Fig 4A and S4B Fig). Evaluation of gene expression in siRNA-treated macrophages suggested that CCND2 (cyclin D2) and CDKN1A (p21) expression is cross-regulated, as interference of CCND2 expression lead to a significant downregulation of CDKN1A (Fig 4A, third panel, 40% inhibition compared to mock macrophages, p = 0.007) and a similar trend was observed when CCND2 expression was evaluated in siCDKN1A macrophages, although it did not reach statistically significance (Fig 4A, first panel, 30% inhibition compared to mock, p = 0.06). Protein expression confirmed the effective downregulation of all target proteins (Fig 4B) as well as the mutual regulation of cyclin D2 and p21, as knockdown of cyclin D2 or p21 induced a downregulation of p21 or cyclin D2 expression, respectively (Fig 4B, lane 3 and lane 5). As above, SAMHD1 expression was not altered in siRNA-treated macrophages, but significant changes were observed in its phosphorylated form, showing an increased phophorylation in cyclin D2 and p21 knockdown macrophages (Fig 4B, two upper panels). Cell proliferation status showed a small percentage of proliferating cells in all cases, albeit a slight increase in Ki67 positive cells was observed in cyclin D2 and p21 knockdown GM-CSF macrophages (Fig 4C, upper panels). Cell cycle analysis did not show any significant differences between the different macrophages (Fig 4C, lower panels). Altogether, these results reinforce the idea of cyclin D2 and p21 sharing a common regulatory pathway in GM-CSF macrophages.
siRNA-treated GM-CSF macrophages were also tested for their capacity to support HIV-1 replication. Knockdown of cyclin D2 and p21 significantly increased HIV-1 replication in GM-CSF macrophages infected with a VSV-pseudotyped NL4-3 GFP expressing virus (roughly 4-fold increase, p = 0.0007 for siCCND2 and p = 0.0003 for siCDKN1A, respectively, Fig 4D). On the contrary, no effect was seen when cyclin D3 or CDK4 expression were inhibited. Proviral DNA formation was also enhanced in cyclin D2 and p21 knockdown GM-CSF macrophages in short-term infections with the fully replicative HIV-1 R5-tropic strain BaL (roughly 3-fold increase, p = 0.007 for siCCND2 and p = 0.03 for siCDKN1A respectively, Fig 4E). As expected, the HIV-1 reverse transcriptase inhibitor AZT completely blocked proviral DNA formation, while the HIV-1 integrase inhibitor Raltegravir (RAL) did not have an effect on viral DNA formation (Fig 4E). Importantly, confirmatory siRNA sequences targeting Cyclin D2 showed similar effects on infection and proviral DNA formation after HIV-1 BaL infection (S4C and S4D Fig), indicating that cyclin D2 acts as part of a viral restriction mechanism in GM-CSF macrophages. No significant differences in basal cytokine expression and the capacity to induce cytokine expression after LPS stimulation was also preserved in cyclin D2 or p21 knockdown macrophages, suggesting no major functional abnormalities as a result of inhibition of cyclin D2 or p21 (S5 Fig).
To investigate the interaction between cyclin D2 and p21, a plasmid expressing a fusion protein Flag-p21 [29] was transfected into HEK293T cells and p21 was immunoprecipitated using Flag-specific agarose beads (Fig 5A). Cyclin D2 co-immunoprecipitated with Flag-p21 (Fig 5A, last lane) and it was not identified when using lysates from mock-transfected cells (M), demonstrating the existence of a protein complex with cyclin D2 and p21. The presence of a CDK in the complex was also investigated and found that CDK1, but not CDK4 or CDK6, immunoprecipitated together with cyclin D2 and p21 in HEK293T cells. The interaction of cyclin D2 and p21 was confirmed by overexpression of a cyclin D2–HA fusion protein [30] in HEK293T cells followed by immunoprecipitation using HA-specific agarose beads. As expected, p21 co-immunoprecipitated with cyclin D2-HA and CDK1 but not CDK4 (Fig 5B), demonstrating the existence of a protein complex between cyclin D2, p21 and a relevant CDK associated to SAMHD1 function.
Deregulation of the cell cycle is a hallmark of most laboratory adapted cell lines in which CDK1 may play a preponderant role in controlling cell proliferation, while other CDKs may have a tissue specific role in driving the cell cycle and cell differentiation [13]. Taking this into account, co-immunoprecipitation experiments of endogenous cyclin D2 and p21 in GM-CSF macrophages were also performed. Importantly, immunoprecipitation of endogenous cyclin D2 or p21 resulted in the identification of p21 or cyclin D2, respectively, demonstrating the coexistence of both proteins in a complex also in GM-CSF primary macrophages (Fig 5C and 5D). The presence of a CDK in the same protein complex was also investigated and CDK4, but not CDK1, was found to co-immunoprecipitate with either cyclin D2 or p21 in primary macrophages (Fig 5C and 5D). When immunoprecipitating p21 higher amounts of cyclin D2 and CDK4 were found than when immunoprecipitating cyclin D2, which might indicate that most p21 is found complexed with cyclin D2 and CDK4, whereas in the case of cyclin D2, only a fraction of the total protein is bound to p21 and CDK4 (Fig 5D). These results further demonstrate the existence of a protein complex formed by cyclin D2, p21 and CDK4 that may govern cell cycle progression and HIV-1 susceptibility in GM-CSF primary macrophages.
CDK1 and CDK2 have been identified as the kinases responsible for SAMHD1 phosphorylation in cycling cells and macrophages, respectively [10–12]. Thus, to delineate the molecular pathway regulated by cyclin D2, expression and activation of CDK1 and CDK2 were analyzed in siRNA-treated GM-CSF macrophages (Fig 6). Knockdown of cyclin D2 enhanced significantly CDK1 mRNA (roughly 3-fold, p = 0.0021, Fig 6A, left panel) and protein expression (Fig 6B). Although no significant upregulation of CDK2 expression was observed (Fig 6A, right panel), CDK2 CDK2 regulatory phosphorylation at Thr130 (pCDK2, Fig 6B) was significantly increased suggesting a higher activity of CDK2.
According to the classical model of cell cycle control, CDK4 or CDK6 regulate events in early G0 to G1 phase, CDK2 triggers S phase, CDK2/CDK1 regulate the completion of the S phase and CDK1 is responsible for mitosis [13]. Thus, the complex formed by cyclin D2-CDK4-p21 might be responsible for the lack of active CDK2 and CDK1 expression in GM-CSF macrophages. This situation is reversed in the absence of cyclin D2, leading to the activation of CDK2 and CDK1 with the subsequent phosphorylation of its substrates, including SAMHD1. As a consequence, SAMHD1 phosphorylation results in inactivation of virus restriction and enhancement of HIV-1 replication (Fig 6C).
Macrophages are key components of the innate immune system that reside in tissues, where they function as immune sentinels. Although macrophage heterogeneity has classically been organized around two polarized endpoints known as M1 or classical and M2 or alternative activation [31], recent evidence have revealed an unrecognized greater diversity in the ontogeny and functional diversity of tissue-resident macrophages. It is now established that macrophages from embryonic progenitors can persist in tissues into adulthood and self-maintain by local proliferation (reviewed in [4, 5]). However, monocytes also contribute to the resident macrophage population, on which the local environment can impose tissue-specific macrophage functions (reviewed in [4, 5]). The myeloid colony-stimulating factors, M-CSF and GM-CSF are known to modulate macrophage phenotype and many studies illustrate their importance in the magnitude, duration and character of inflammatory responses [32, 33]. Thus, the description of the distinct molecular phenotypes consequence of M-CSF or GM-CSF stimulation may be important for both acute and chronic inflammatory pathology. Here, we show that differentiation stimuli determine distinct cell proliferation and cell cycle progression in primary macrophages, characteristics that are mostly dependent on the differential expression patterns of cell cycle proteins, especially D-type cyclins, their catalytic partners, CDKs, and the CDK inhibitor p21. Consistent with our observations, upregulation of cyclin D2 in response to GM-CSF treatment has already been reported in different hematopoietic cells [34–36], suggesting that cyclin D2 role might be relevant also for other cell types.
The molecular interplay between the cell cycle, cyclins, and cell function is far from being fully understood. Conceptual advances in the field continue to uncover novel and interesting roles for cyclins in cellular processes that contribute to diseases, such as cancer or pathogenic infections [37]. It is well accepted that cell cycle control plays a major role in determining susceptibility to HIV-1 infection [38, 39]. It has been previously shown that limiting dNTP synthesis, for example through ribonucleotide reductase inhibition by hydroxyurea, limits HIV-1 replication [40]. Thus, differences on cell cycle regulation between M-CSF and GM-CSF macrophages also determined the susceptibility to HIV-1 infection and this may be a direct consequence of the control of the restriction factor SAMHD1 by CDK-mediated phosphorylation. Previous observations in M-CSF macrophages, showed that SAMHD1 phosphorylation was directly phosphorylated by CDK2, whose kinase activity was upstream regulated by the cyclin D3-CDK6 complex [11, 16]. However, the molecular mechanism might be different in GM-CSF macrophages because CDK6 was barely expressed and D-type cyclins expression was significantly upregulated, especially that of cyclin D2. Accordingly, knockdown of cyclin D2 and cyclin D3 resulted in opposite effects depending on the macrophage type, in terms of susceptibility to HIV-1 infection: cyclin D2 restricts infection in non-cycling GM-CSF macrophages and cyclin D3 enables infection in proliferating M-CSF macrophages. All D-type cyclins (D1, D2 and D3) are closely associated to the G1 phase, whose expression is induced by mitogenic signals and therefore play a significant role in cell cycle entry, by assembling together with CDK4 or CDK6 [18, 21, 37]. However, although apparently redundant, single knockout mice exhibit several cell type specific abnormalities [21], suggestive of essential functions on particular settings for each cyclin, explaining, in part, the apparently opposed effects observed here for cyclin D2 and D3 when assessing susceptibility to HIV infection. On the other hand, D-type cyclins are linked to many human malignancies. Cyclin D1 is a well-established oncogene (review in [18]) but cyclin D2 or cyclin D3 overexpression are rarely reported [41, 42]. However, CCND2 is frequently methylated, with loss of cyclin D2 expression in pancreatic, breast and prostate cancer [43–45], pointing to a potential role as a tumor suppressor rather than an oncogene, indicating that overexpression of cyclin D2 might be limiting cell proliferation, as described here in GM-CSF macrophages.
To uncover the molecular mechanism underlying cyclin D2 function in GM-CSF macrophages and taking into account that cyclins alone do not have catalytic activity per se, the study of putative partners of cyclin D2 was addressed. In concordance with the differential gene expression patterns, the inhibition of CDK4 which is similarly expressed in both macrophage types did not change protein profile or HIV-1 susceptibility. Conversely, knockdown of the CDK inhibitor p21, whose expression was also upregulated in GM-CSF macrophages, showed the same effect as that of Cyclin D2, i.e., upregulation of SAMHD1 phosphorylation and enhancement of HIV-1 replication, similar to our previous observation in M-CSF macrophages [20]. p21 belongs to the Cip/Kip family of CDKIs that have historically been considered negative regulators of the cyclin-CDKs and therefore controlling cell cycle progression, especially when referred to cyclin-CDK1 or -CDK2 complexes [20, 46]. However, Cip/Kip family of CDKIs interaction with Cyclin D-CDK4/6 appear much more complex, being involved in both the stabilization of the cyclin D/CDK complex but also acting as inhibitors of its kinase activity [47, 48]. The observation that a complex formed by Cyclin D2/CDK4/p21 was identified in non-cycling GM-CSF macrophages argues in favor of the inhibitory hypothesis at least in this specific cell type. Moreover, the fact that p27, another Cip/Kip family member, is barely expressed in GM-CSF macrophages indicates the specificity to the effect of Cyclin D2/CDK4/p21 complex. A direct inhibition of CDK2 function, similar to that observed in M-CSF [20] cannot be completely rule out, but seems improbable due to the low expression of CDK2 and lack of cell proliferation markers seemed in GM-CSF macrophages.
The present works also highlights the interplay between cell cycle control and viral replication, with important implications that might be broader than simply affecting susceptibility to HIV-1 infection. Interestingly, both p21 and Cyclin D2 were identified as potential markers for viral latency, as both genes showed increased histone modification levels in HIV latently infected cells [49]. These observation suggests that the maintenance or not of HIV-1 latency may also be controlled by cell cycle related proteins such as Cyclin D2 and p21 and thus, open the opportunity for new therapeutic interventions. In addition, p21 and Cyclin D2 were also found upregulated after HTLV-I infection, a process that is mediated by the viral protein Tax, indicating that deregulation of G1/S checkpoint is also relevant for other retroviruses [48].
In summary, the identification of a novel cell cycle-mediated viral restriction pathway in primary non cycling macrophages have provided new evidences of the tight interplay between viral replication and cell cycle control, pointing towards the concerted action of Cyclin D2 and p21 in HIV-1 replication. The demonstration and characterization of specific D-type Cyclin roles in certain cell types, such as that reported for Cyclin D2 here, may also offer a window of opportunity for targeting D Cyclins in viral infections and in human cancers as well, where D-type Cyclin expression is frequently deregulated.
PBMC were obtained from buffy coats of blood of healthy donors using a Ficoll-Paque density gradient centrifugation and monocytes were purified using negative selection antibody cocktails (StemCell Technologies) as described before [22]. Monocytes were cultured in complete culture medium (RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum (FBS; Gibco) or human serum (HS; Sigma) and penicillin/streptomycin (Gibco) and differentiated to monocyte derived macrophages (MDM) for 4 days in the presence of monocyte-colony stimulating factor (M-CSF, Peprotech) or granulocyte-macrophage colony-stimulating factor (GM-CSF, Peprotech) both at 100 ng/ml. The protocol was approved by the scientific committee of Fundació IrsiCaixa. Buffy coats were purchased from the Catalan Banc de Sang i Teixits (http://www.bancsang.net/en/index.html). The buffy coats received were totally anonymous and untraceable and the only information given was whether or not they have been tested for disease. When appropriate, differentiated macrophages were incubated with 100 ng/ml of lipopolisaccaride (LPS, Sigma-Aldrich) overnight at 37°C.
TZM cells were received from the National Institutes of Health, AIDS Research and Reference Reagent Program. HEK293T cells were purchased from Dharmacon (Madrid, Spain).
Wild-type C57BL/6 inbred mice were purchased from Harlan Laboratories (Sant Feliu de Codines, Barcelona, Spain) and housed at the Animal House facility at the Research Institute Germans Trias i Pujol. Mice were housed under specific pathogen free conditions in a temperature and humidity-controlled room with 12-h light/12-h dark cycle. Only adult males were used in this study. In vivo experiments were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Generalitat de Catalunya, Catalan Government and the Principles of laboratory animal care (NIH pub.85–23 revised 1985; http://grants1.nih.gov/grants/olaw/references/phspol.htm). Murine peritoneal macrophages were isolated as described [50]. Briefly, C57Bl/6 mice were sacrificed by cervical dislocation. Immediately after, peritoneal wall was exposed and 10 ml of a cold solution of PBS with 3% FBS per mouse was injected into the peritoneal cavity. Using the same syringe and needle, the fluid from the peritoneum was aspirated, and 8 ml of fluid was recovered. Peritoneal fluid was centrifuged at 4°C and cell pellet was resuspended to adjust cell concentration at 1–3 x 106 cells / ml. To obtain monolayers of peritoneal macrophages, total cells were plated at 2–3 × 106 total nucleated cells/ml in DMEM/F12-10 medium. Cells were allowed to adhere for 1–2 hr at 37°C. Non-adherent cells were removed by gently washing three times with warm PBS. At this time, cells were greater than 90% macrophages and were processed.
Isolated monocytes were transfected as previously described [11, 16]. Briefly, 50 pmol of the corresponding siRNA (siGENOME SMARTpool from Dharmacon, Thermo-Scientific, Waltham, USA and ThermoFisher Scientific), were transfected using a Monocyte Amaxa Nucleofection kit (Lonza, Basel, Switzerland) following manufacturer instructions. Monocytes were left untreated overnight and then differentiated to macrophages as described above.
For intracellular Ki67 staining, cells were fixed for 3 min with Fixation Buffer (Fix & Perm, Life Technologies) before adding pre-cooled 50% methanol for 10 min at 4°C. Cells were then washed in PBS with 5% FBS and incubated for 30 min with the Ki-67 FITC antibody diluted in permeabilitzation buffer (1:10; clone B56, BD Biosciences). For cell cycle analysis, cell were suspended in 0.03% saponin (Sigma-Aldrich) in PBS and then incubated in 20 mM 7-aminoactinomycin D (7AAD; Sigma-Aldrich) for 30 min at room temperature in the dark, followed by 5 min at 4°C. Then, Pyronin Y (Sigma-Aldrich) was added at a final concentration of 1.5 μg/ml and cells were further incubated at 4°C for 15 min. Flow cytometry was performed in a LSRII flow cytometer (BD Biosciences). The data were analyzed using the FlowJo software (BD Biosciences). To correct the overestimation of G2/M population by miss discrimination of cellular doublets, FL2W versus FL2A of the 7AAD dye was plotted before gating for the distinct cell cycle phases [51].
3-Azido-3-deoxythymidine (zidovudine, AZT) was purchased from Sigma-Aldrich (Madrid, Spain). nevirapine (NVP) and raltegravir (RAL) were obtained from the NIH AIDS Research and Reference Reagent Program. PD-0332991 (palbociclib) was purchased from Selleckchem.
For relative mRNA quantification, RNA was extracted using the NucleoSpin RNA II kit (Magerey-Nagel), as recommended by the manufacturer, including the DNase I treatment step. Reverse transcriptase was performed using the High Capacity cDNA Reverse Transcription Kit (Life Technologies). mRNA relative levels of all genes were measured by two-step quantitative RT-PCR and normalized to GAPDH mRNA expression using the DDCt method. Primers and DNA probes were purchased from Life Technologies (TaqMan gene expression assays).
Cytokine expression was evaluated by using the commercial TaqMan Human Cytokine Network array (4414255, Life Technologies), which included primers and probes for 28 different cytokine genes. mRNA relative levels of all cytokine genes were measured by two-step quantitative RT-PCR and normalized to GAPDH mRNA expression using the DDCt method.
Intracellular dNTP content was determined using a polymerase-based method [52] as previously described [23].
Envelope-deficient HIV-1 NL4-3 clone encoding IRES-GFP (NL4-3-GFP) was pseudotyped with VSV-G by cotransfection of HEK293T cells using polyethylenimine (Polysciences) as previously described [11, 23]. For the production of viral-like particles carrying Vpx (VLPVpx), HEK293T cells were cotransfected with pSIV3+ and a VSV-G expressing plasmid. Three days after transfection, supernatants were harvested, filtered and stored at -80°C. Viral stocks were concentrated using Lenti-X concentrator (Clontech). Viruses were titrated by infection of TZM cells followed by GFP quantification by flow cytometry. R5-tropic HIV-1 strain BaL was grown in stimulated PBMC and specifically titrated for its use in assays of total viral DNA formation in MDM.
M-CSF or GM-CSF differentiated MDM were infected with VSV-pseudotyped NL4-3-GFP and antiviral drugs were added at the time of infection. When necessary, differentiated MDM were pretreated with VLPVpx for 4h before infection or left with fresh media as a control. Viral replication was measured in all cases two days later by flow cytometry (LSRII, BD Biosciences). Measurement of cell cytotoxicity was performed by flow cytometry, i.e., cells were gated as living or dead, according to flow cytometry FSC and SSC parameters. BaL infections were stopped at 16h to measure only early events of viral infection (reverse transcription). For quantification of proviral DNA, a primer and probe set that is able to amplify both unintegrated and integrated viral DNA was used as described before [11, 16]. DNA was extracted using a DNA extraction kit (Qiagen) and proviral DNA quantifications were performed. Ct values for proviral DNA were normalized using RNaseP as housekeeping gene by the ΔΔCt method. Infections were normalized to an untreated control. To ensure that measured proviral DNA was the product of infection and not result from DNA contamination of the viral stocks samples treated with RT inhibitor AZT (1 μM) were run in parallel. raltegravir (2 μM) was used to ensure that no post-RT steps were being quantified by the assay.
Cells were rinsed in ice-cold phosphate-buffered saline (PBS) and extracts prepared in lysis buffer (50 mM Tris HCl pH 7.5, 1 mM EDTA, 1 mM EGTA, 1 mM Na3VO4, 10 mM Na β-glycerophosphate, 50 mM NaF, 5 mM Na Pyrophosphate, 270 mM sucrose and 1% Triton X-100) supplemented with protease inhibitor (Roche) and 1 mM phenylmethylsulfonyl fluoride. Lysates were subjected to SDS-PAGE and transferred to a PVDF membrane (ImmunolonP, Thermo). The following antibodies were used for immunoblotting: anti-rabbit and anti-mouse horseradish peroxidase-conjugated secondary antibodies (1:5000; Pierce); anti-human Hsp90 (1:1000; 610418, BD Biosciences), anti-SAMHD1 (1:1000; ab67820, Abcam), anti-CDK1 (9116) anti-CDK2 (2546), anti-phosphoCDK2 (Thr160; 2561), anti-CDK4 (D9G3E), anti-CDK6 (3136), anti-cyclin A2 (BF683), anti-cyclin D2 (D52F9), anti-cyclin D3 (DCS22), anti-p21 (2947) and anti-p27 (2552) all 1:1000 from Cell Signaling. Anti-phospho-SAMHD1 Thr592 was obtained by immunization of rabbit using a phosphorylated peptide as described before [53].
HEK293T cells were transfected with Flag-tagged p21 (Addgene plasmid # 16240, gift from Mien-Chie Hung) [29] or HA-tagged Cyclin D2 (Addgene plasmid # 8950, gift from Philip Hinds) [30] expression vectors using lipofectamine 2000 (Invitrogen). 48 h later, cells were chilled to 4°C and cell extracts prepared with lysis buffer as described above. Lysates were cleared by centrifugation at 10500 rpm for 10 min and incubated with anti-FLAG (anti-Flag M2 Affinity Gel, Sigma) or anti-HA (monoclonal anti-HA-agarose, Sigma) antibodies covalently attached to agarose overnight at 4°C on a rocking platform. Beads were then collected by centrifugation at 3000 rpm for 5 min at 4°C, extensively washed in lysis buffer and resuspended in SDS gel loading buffer. The proteins were separated on a 10% SDS-polyacrylamide gel, transferred to a PVDF membrane, and analyzed by immunoblotting with the corresponding antibodies.
Co-immunoprecipitation of endogenously expressed proteins was performed using GM-CSF differentiated macrophages. Cell extracts were prepared as above and lysates were incubated with anti-Cyclin D2 antibody (D52F9, Cell Signaling), anti-p21 antibody (2947, Cell Signalling) or rabbit IgG overnight at 4°C and further incubated with Fast flow Sepharose (Sigma-Aldrich) for 1-2h. Beads were then collected by centrifugation at 3000 rpm for 5 min at 4°C, extensively washed in lysis buffer and resuspended in SDS gel loading buffer. The proteins were separated on a 10% SDS-polyacrylamide gel, transferred to a PVDF membrane, and analyzed by immunoblotting with the corresponding antibodies.
Data were analyzed with the PRISM statistical package. If not stated otherwise, all data were normally distributed and expressed as mean ± SD. p-values were calculated using an unpaired, two-tailed, t-student test.
|
10.1371/journal.pgen.1003679 | Identification of DSB-1, a Protein Required for Initiation of Meiotic Recombination in Caenorhabditis elegans, Illuminates a Crossover Assurance Checkpoint | Meiotic recombination, an essential aspect of sexual reproduction, is initiated by programmed DNA double-strand breaks (DSBs). DSBs are catalyzed by the widely-conserved Spo11 enzyme; however, the activity of Spo11 is regulated by additional factors that are poorly conserved through evolution. To expand our understanding of meiotic regulation, we have characterized a novel gene, dsb-1, that is specifically required for meiotic DSB formation in the nematode Caenorhabditis elegans. DSB-1 localizes to chromosomes during early meiotic prophase, coincident with the timing of DSB formation. DSB-1 also promotes normal protein levels and chromosome localization of DSB-2, a paralogous protein that plays a related role in initiating recombination. Mutations that disrupt crossover formation result in prolonged DSB-1 association with chromosomes, suggesting that nuclei may remain in a DSB-permissive state. Extended DSB-1 localization is seen even in mutants with defects in early recombination steps, including spo-11, suggesting that the absence of crossover precursors triggers the extension. Strikingly, failure to form a crossover precursor on a single chromosome pair is sufficient to extend the localization of DSB-1 on all chromosomes in the same nucleus. Based on these observations we propose a model for crossover assurance that acts through DSB-1 to maintain a DSB-permissive state until all chromosome pairs acquire crossover precursors. This work identifies a novel component of the DSB machinery in C. elegans, and sheds light on an important pathway that regulates DSB formation for crossover assurance.
| For most eukaryotes, recombination between homologous chromosomes during meiosis is an essential aspect of sexual reproduction. Meiotic recombination is initiated by programmed double-strand breaks in DNA, which have the potential to induce mutations if not efficiently repaired. To better understand the mechanisms that govern the initiation of recombination and regulate the formation of double-strand breaks, we use the nematode Caenorhabditis elegans as a model system. Here we describe a new gene, dsb-1, that is required for double-strand break formation in C. elegans. Through analysis of the encoded DSB-1 protein we illuminate an important regulatory pathway that promotes crossover recombination events on all chromosome pairs to ensure successful meiosis.
| Formation of crossovers between homologous chromosomes is essential for successful execution of the meiotic program in most sexually reproducing organisms. In addition to shuffling genetic information between parental chromosomes, crossovers, together with cohesion between sister chromatids, create physical links between homologous chromosomes that enable their segregation to daughter cells during the first meiotic division [1], [2]. Disruption of crossover formation leads to chromosome nondisjunction and the formation of aneuploid gametes, and thereby greatly reduces fertility.
Meiotic recombination is initiated by programmed DNA double-strand breaks (DSBs), a subset of which is repaired to form crossovers between homologous chromosomes (for a review, see [3]). While a minimum number of DSBs is needed to promote the necessary crossovers on all chromosome pairs, excessive DSBs could threaten genome integrity. The number of meiotic DSBs in various organisms appears to be highly regulated, presumably to balance the crossover requirement with the risk of mutation. The timing of programmed DSBs during early meiotic prophase is also tightly controlled to maximize the likelihood of productive repair [4], [5]. For example, in Saccharomyces cerevisiae, the activities of cell cycle-regulated kinases involved in DNA replication ensure that that DSBs occur only after DNA synthesis is complete [6]–[10]. DSB formation must also be inactivated during meiotic prophase to allow for repair prior to the meiotic divisions. Mechanisms that terminate DSB formation are not well understood, although recent studies have shown that the ATM/ATR family of DNA damage response kinases is involved in down-regulating the number of DSBs in mice, S. cerevisiae, and Drosophila melanogaster [11]–[14]. Further investigations are needed to better understand the mechanisms underlying these various aspects of DSB regulation.
Meiotic DSBs are catalyzed by the widely conserved, topoisomerase-related enzyme Spo11 [15], [16]. Although Spo11 is essential for DSB formation, it does not function alone. In various organisms – including fungi, plants, and animals – additional proteins required for meiotic DSBs have been identified (for a review, see [17]). Unlike Spo11, other known factors involved in DSB formation are poorly conserved. For example, of five meiosis-specific DSB proteins found in S. cerevisiae, only two (Rec114 and Mei4) have known orthologs in other phyla; and even these two proteins are absent in several species, including Caenorhabditis elegans, D. melanogaster, and Neurospora crassa [18]. Additional DSB proteins have also been identified in other organisms, but none are ubiquitous among eukaryotes [5], .
The nematode C. elegans has emerged as a valuable model system for molecular analysis of meiosis. As in other eukaryotes, SPO-11 catalyzes the formation of meiotic DSBs [23]. MRE-11 and RAD-50 are also required for DSB formation [24], [25] as in S. cerevisiae [17], but these proteins have other essential roles in DNA metabolism, including in the resection of meiotic DSBs [3], [26]. In C. elegans, as in other species, meiosis-specific chromosome architecture contributes to DSB proficiency. In particular, in the absence of HTP-3, an integral component of chromosome axes, DSBs are abolished or sharply reduced [27]. The related protein HTP-1, which is also associated with the axial elements, may also contribute to DSB formation, while other axial components appear to be dispensable for DSBs [28]–[30]. Roles for axis components homologous to HTP-3 and HTP-1 in promoting DSBs have also been demonstrated in other organisms [3], [31], [32]. Additionally, the meiotic kinase CHK-2, which regulates many key events during early meiotic prophase, is required for programmed DSBs in C. elegans [33]. Several other factors are known to influence meiotic DSB formation, but their effects may be indirect. These include the chromatin-associated proteins HIM-5, HIM-17, and XND-1, which promote normal levels of meiotic DSBs, but whose functions are pleiotropic and not well understood 34–36. Apart from SPO-11, no protein that specifically functions in initiating recombination has previously been reported. Some aspects of C. elegans meiosis are unusual among model organisms, including the fact that synapsis between homologous chromosomes is independent of recombination [23]. Thus, analysis of DSB regulation in C. elegans will likely reveal both conserved aspects of meiosis and how regulatory circuits are remodeled during evolution.
Here we identify a novel gene, dsb-1 (double-strand break factor 1), that is required for meiotic DSB formation in C. elegans. dsb-1 mutants lack meiotic DSBs, and show meiotic defects similar to spo-11 mutants. DSB-1 localizes to meiotic chromosomes coincident with the time of DSB formation, in a manner dependent on the CHK-2 kinase. We also find that a variety of mutations that disrupt crossover formation on one or more chromosomes extend the chromosomal localization of DSB-1, suggesting that the DSB-permissive state may be prolonged. Based on these observations, we infer the existence of a regulatory circuit in which meiotic nuclei monitor the recombination status of each chromosome pair and act through DSB-1 to maintain a DSB-permissive state until all chromosome pairs have attained crossover-competent recombination intermediates.
In C. elegans, mutations that impair meiotic chromosome segregation result in embryonic lethality and a high incidence of males (XO) among the surviving progeny [37]. The dsb-1(we11) mutant was isolated in a genetic screen for maternal-effect embryonic lethality, and was found to produce a high fraction of males among its few surviving self-progeny. A targeted deletion allele of the affected gene, dsb-1(tm5034), was generated independently (see below), and results in defects identical to dsb-1(we11) based on all assays described here. Whereas self-fertilizing wild-type hermaphrodites produce nearly 100% viable progeny and 0.2% males (Figure 1A, [37]), only 3% of progeny from self-fertilizing dsb-1 mutant hermaphrodites survived to adulthood (n>2000; 12 broods), (Figure 1A, Table 1). Among these survivors, 36–38% were male (Figure 1A, Table 1). The brood size (number of fertilized eggs) of self-fertilizing dsb-1 hermaphrodites was also reduced relative to wild-type animals (Table 1).
Chromosome segregation errors in meiosis often reflect defects in crossover formation between homologs. The levels of embryonic lethality and male progeny observed in dsb-1 mutants are quantitatively similar to several previously characterized mutants that fail to make any crossovers during meiotic prophase, such as spo-11 (Figure 1A, Table 1), msh-5, and cosa-1 [23], [38], [39], suggesting that dsb-1 mutants might also lack crossovers. Visualization of DAPI-stained oocytes at diakinesis provides a simple assay for crossover formation in C. elegans. In wild-type hermaphrodites, 6 DAPI-staining bodies are observed in each oocyte nucleus (average = 5.8, Figure 1B and 1C), corresponding to 6 pairs of homologous chromosomes, each held together by a chiasma [40]. In mutants that fail to make crossovers, oocytes typically display 12 DAPI-staining bodies. The number and morphology of DAPI-staining bodies observed in dsb-1 mutant oocytes was similar to spo-11 mutants (average = 11.6, Figure 1B and 1C), indicating an absence of chiasmata in dsb-1 animals.
We investigated whether the disruption of crossover formation in dsb-1 mutants might reflect a defect in homologous chromosome pairing or synapsis. Pairing was assessed using fluorescence in situ hybridization (Figure 1D). Early pachytene nuclei of both wild-type and dsb-1 animals contained a single focus or closely apposed pair of foci, indicating that homologous chromosomes were paired (Figure 1D). Further, co-staining of the axial element protein HTP-3 and the synaptonemal complex central region protein SYP-1 indicated that chromosomes were fully synapsed by early pachytene in dsb-1 animals (Figure 1E), as in wild-type animals. These results indicate that dsb-1 mutants are proficient for homologous chromosome pairing and synapsis.
To assess whether dsb-1 mutants initiate meiotic recombination, we used antibodies against the DNA strand-exchange protein RAD-51, which binds to single-stranded regions adjacent to resected DSBs [26], [41], as a cytological marker of recombination intermediates [42], [43]. Whereas wild-type oocytes in early pachytene showed abundant RAD-51 foci, dsb-1 gonads lacked RAD-51 staining (Figure 2A), indicating either failure to form DSBs or failure to load RAD-51. However, the lack of fragmented chromosomes at diakinesis seemed more consistent with an absence of DSBs.
To verify that dsb-1 mutants are defective in DSB formation, and to rule out the possibility of defects in the loading of RAD-51 or downstream steps of the recombination pathway, we tested whether exogenous DSBs could rescue the recombination defects observed in dsb-1 mutants. The same approach established a role for Spo11/SPO-11 in DSB formation [23], [44]. Young adult dsb-1 mutant hermaphrodites were exposed to 10 Gy of gamma rays, a dose that has previously been shown to efficiently rescue crossovers in spo-11 mutants with minimal associated lethality [25]. Wild-type and spo-11 controls were performed in parallel. At appropriate times after irradiation the animals were assessed for RAD-51 foci, chiasmata, and progeny viability. At 2 hours post irradiation, dsb-1 animals displayed abundant RAD-51 foci (Figure 2B), indicating that the mutants are proficient for resection and RAD-51 loading. At 18 hours post irradiation, both spo-11 and dsb-1 oocytes showed ∼6 DAPI-staining bodies (Figure 2C and 2D). Additionally, the viability of embryos laid 20–30 hours post irradiation increased significantly for both spo-11 and dsb-1 animals, but decreased slightly for wild-type, compared to unirradiated controls (Figure 2E). The ability of exogenous DSBs to rescue the recombination defects of dsb-1 animals indicates that these mutants are specifically defective in meiotic DSB formation.
The defects observed in dsb-1 mutant hermaphrodites are virtually indistinguishable from spo-11(me44) mutants, except that mutations in dsb-1 were associated with reduced brood size (Table 1). Although dsb-1(we11) showed linkage to the middle of Chromosome IV, close to the spo-11 locus, complementation tests revealed that we11 is not an allele of spo-11. Quantitative RT-PCR also indicated that spo-11 mRNA levels were unaffected in dsb-1(we11) mutants (Figure S1).
Whole-genome sequencing of backcrossed dsb-1(we11) animals identified several mutations in annotated coding sequences, including a nonsense mutation in the previously uncharacterized gene F08G5.1 (Figure 3A), which encodes a predicted protein of 385 amino acids and seemed a plausible candidate based on its meiosis-enriched expression pattern [45]. We found that knockdown of F08G5.1 expression via transgene-mediated cosuppression [46] caused embryonic lethality and male progeny, as well as strong reduction of chiasmata, in the oocytes of treated animals (data not shown), supporting the hypothesis that the we11 mutation affects this gene. we11 introduces a premature stop (tac = >taa) after lysine 96 (Figure 3A). A targeted deletion allele (tm5034) removes 290 bp from predicted exons 3 and 4 and the intervening intron (Figure 3A), resulting in a frameshift mutation that introduces a glutamine immediately followed by a stop codon after lysine 96. The phenotype of dsb-1(tm5034) mutants is indistinguishable from dsb-1(we11) (Figure 1 and 2, Table 1). Both are predicted to lack functional protein based on the early stop codons, and this conclusion is supported by immunofluorescence and immunoblotting experiments (below). Based on the evidence described above that mutations disrupting F08G5.1 specifically interfere with meiotic double-strand break formation, we designated F08G5.1 as dsb-1, for double-strand break factor 1.
The DSB-1 protein has no apparent homologs outside of the genus Caenorhabditis, including other nematode genera. Interestingly, the genomes of C. elegans and several other Caenorhabditids each contain 2 predicted paralogs. In an accompanying paper, Rosu et al. show that dsb-1 paralog F26H11.6/dsb-2 is also involved in meiotic DSB formation in C. elegans [47]. DSB-1, DSB-2, and their homologs cluster into two paralogous groups (Figure 3B). Even within Caenorhabditis, members of this protein family are not well conserved (Figure S2).
DSB-1 lacks identifiable domains that might give clues about its function in DSB formation. One notable feature is its high serine content: 60 of 385 amino acids (16%) are serine residues, compared to an average serine content of 8% encoded by all C. elegans ORFs [48]. Protein structure prediction algorithms indicate that each end of DSB-1 may form alpha-helix secondary structures, but the central portion of the protein, which is especially serine-rich, is predicted to be largely unstructured. This central region is also the least conserved portion of the protein (Figure S2). Five serine residues within the central region are followed by glutamine (Q), making them candidate phosphorylation targets for ATM or ATR DNA damage kinases. These clustered ATM/ATR consensus motifs are shared by other DSB-1 homologs, including DSB-2.
To further probe the role of DSB-1 in the formation of meiotic DSBs, we generated an antibody against the full-length protein expressed in E. coli. Immunofluorescence staining revealed that DSB-1 is absent from somatic nuclei, and specifically localizes to chromosomes during early meiotic prophase (Figure 4A and 4B), while dsb-1 mutants showed only background staining (Figure S3). Accumulation of DSB-1 on chromosomes was first observed in nuclei marked by crescent-shaped DAPI-staining morphology, corresponding to the “transition zone” (leptotene/zygotene), and disappeared at mid-pachytene (Figure 4A). Chromosomal localization of DSB-1 preceded the appearance of RAD-51 foci, consistent with an early role for DSB-1 in meiotic recombination (Figure 4A). Thus, the localization of DSB-1 to chromosomes corresponds to the period during which DSBs are likely to be generated.
While most nuclei in the late pachytene region of the germline lacked DSB-1 staining, we consistently observed dispersed nuclei in this region that retained bright fluorescence (Figure 4A and S4). These nuclei also contained abundant RAD-51 foci and frequently displayed compact chromosome morphology resembling that seen in the transition zone, along with evidence of asynapsed chromosomes (Figure S4A and S4B). We tested whether these late DSB-1 positive nuclei might be apoptotic by examining ced-4 mutants, which lack germline apoptosis 49,50, and found that they were still present in similar numbers (data not shown). The persistence of RAD-51 foci and asynapsed chromosomes suggest that these nuclei may be delayed in completing synapsis or other prerequisites for crossover formation, a conclusion reinforced by further analysis of DSB-1 regulation, described below.
DSB-1 was distributed as a network of foci and stretches of staining on meiotic chromosomes (Figure 4B). One chromosome consistently showed weaker DSB-1 staining. This seemed likely to be the X chromosome, which has many unique features in the germline, including distinct chromatin marks 51,52 and genetic requirements for meiotic DSBs 35,36. Co-staining with antibodies against HIM-8, which specifically mark the X chromosome 53, confirmed that the chromosome pair with weaker DSB-1 staining was the X (Figure 4C).
DSB-1 and RAD-51 both localized to chromosomes during early pachytene (Figure 4A). However, we found that RAD-51 did not colocalize with DSB-1 (Figure 4D). Similar findings have been reported for DSB proteins in mice and Schizosaccharomyces pombe 18,54. This could indicate that DSB-1 does not act directly at DSB sites, or that it is removed from DSB sites prior to RAD-51 loading.
Meiotic chromosomes are believed to be organized as chromatin loops tethered at their bases to the proteinaceous chromosome axis 55,56. Based on work from S. cerevisiae, it has been proposed that DSBs occur at sites within chromatin loops that are recruited to the chromosome axis 57–59. Most DSB-1 staining was associated with the periphery of DAPI-stained chromosomes rather than axes (Figure 4E), suggesting that DSB-1 is primarily associated with chromatin loops. This localization pattern is similar to what has been observed for several DSB proteins in S. cerevisiae 6,60–63.
We tested whether DSB-1 localization depends on other factors required for DSB formation. DSB-1 localized to meiotic chromosomes in the catalytically dead spo-11(me44) mutant 25, as well as in spo-11(ok79) mutants, which lack functional protein 23, indicating that DSB-1 localizes to chromosomes independently of DSBs and SPO-11 (Figure 5). DSB-1 localization was also independent of MRE-11 and RAD-50 (Figure 5, data not shown), which are required for DSB formation in C. elegans 24,25. In htp-3 mutants, which lack an essential axial element component that is important for DSB formation 27, DSB-1 was detected on meiotic chromosomes (Figure 5), but the staining appeared somewhat reduced compared to wild-type nuclei.
The CHK-2 kinase is essential for several key events during early meiotic prophase in C. elegans, including DSB formation and homolog pairing 33. We found that nuclear staining of DSB-1 was strongly reduced, albeit still detectable, in chk-2(me64) mutants (Figure 5). Although the intensity of DSB-1 staining was sharply reduced in chk-2 mutants, it appears that faint fluorescence observed upon prolonged exposure reflects DSB-1, because the nuclear staining pattern resembles that seen in wild-type animals, and because chromosomal staining is not detected prior to meiotic entry. Western blot analysis revealed that DSB-1 protein is expressed in chk-2 mutants, although the protein levels appear somewhat reduced compared to wild-type (data not shown). However, the reduction in DSB-1 protein levels in chk-2 mutants does not appear to fully account for the sharply diminished chromosomal localization of DSB-1. These data indicate that DSB-1 localization to chromosomes is largely dependent on the CHK-2 kinase, and suggest that DSB-1 may act downstream of CHK-2 to promote DSBs.
In testing the genetic requirements for DSB-1 localization, we noticed that the zone of DSB-1 staining in the gonad was extended in mutants that disrupt crossover formation. Previous studies have reported a persistence of RAD-51 foci in numerous mutants that are proficient for DSBs but not for crossovers 43,64,65. In wild-type animals, and also in most mutants with extended RAD-51 staining, DSB-1 staining disappeared concomitant with, or slightly before, the disappearance of RAD-51 foci (Figure 4A and S5). Two exceptions were rec-8 and rad-54 mutants, in which DSB-1 staining disappeared by late pachytene, but RAD-51 staining persisted into diplotene (Figure 6, data not shown). Since DSB-1 is required for DSBs and its localization correlates with the timing of DSB formation, its presence on chromosomes may be indicative of proficiency for DSB formation. Although the presence of DSB-1 on chromosomes may not be sufficient for prolonged DSB formation, we interpret extension of the region of DSB-1 staining as evidence of a prolonged DSB-permissive state (see Discussion).
We quantified the extension of DSB-1 localization by comparing the length of the zone of DSB-1-positive nuclei to the total length of the region spanned by the transition zone through late pachytene nuclei, just before oocyte nuclei begin to form a single row near the bend region of the gonad, which coincides with diplotene (Figure 6A). We designated this entire zone as the LZP region (leptotene-zygotene-pachytene), although it also includes a few diplotene nuclei. We found that this metric – the length ratio of the DSB-1-positive region to the LZP region – was consistent across age-matched animals of the same genotype. In wild-type adult hermaphrodites, DSB-1 positive nuclei comprised about 50% of the length the LZP region (Figure 6A and 6B). However, in most mutants that disrupt crossover formation on one or more chromosomes, this zone of DSB-1 staining was significantly extended (Figure 6A and 6B). We saw some variability in this extension, which tended to correlate with the nature of the mutation: Mutations affecting late steps in the crossover pathway, including msh-5(me23) 38, cosa-1(me13) 39, and zhp-3(jf61) 66, extended the DSB-1 zone to ∼75%, of the LZP region (Figure 6A and 6B). Mutations that block earlier steps in homologous recombination, including com-1(t1626) 67, rad-51(Ig8701) 42, and rad-54(tm1268) 65, extended the DSB-1 zone even further, to ∼90% of the LZP region. Mutations that block crossover formation by disrupting synapsis, including syp-1(me17) 68 and syp-2(ok307) 43, also showed an extension of DSB-1 staining to ∼90% (Figure 6B).
Significantly, mutants that lack meiotic DSBs, including spo-11(ok79 or me44) 23,25, mre-11(ok179) 24, and rad-50(ok197) 25, also showed significant extension of the zone of DSB-1 staining to 69–78% of the LZP region (Figure 6A and 6B). Together these findings indicate that the absence of crossovers or crossover precursors, rather than the presence or persistence of earlier recombination intermediates, triggers extension of the DSB-1 zone.
Also of note, in htp-1 and htp-3 mutants 27–29, in which the axial element is disrupted, the region of DSB-1 staining was shorter than in other crossover-deficient mutants (Figure 6B), despite the fact that no crossovers form in these animals and DSBs are either eliminated or reduced 27–29. This suggests that axis structure may play a role in detecting or signaling the absence of crossover precursors to prolong DSB-1 localization, consistent with proposed roles for the axis in other species 32,69–71.
We tested whether irradiation could suppress the extension of the DSB-1 zone seen in spo-11 mutants. Young adult hermaphrodites were irradiated, then fixed and stained 8 hours later. As controls, we included mutants (mre-11 and msh-5) in which crossover defects are not rescued by exogenous DSBs 24,38. Irradiation reduced the zone of DSB-1 staining in spo-11(me44) animals to 56%, compared to 70% for unirradiated controls (Figure 6C). In contrast, the length of the DSB-1 zone in wild-type, mre-11, and msh-5 hermaphrodites was unaffected by irradiation (Figure 6C). These data reinforce the idea that the absence of crossovers or crossover precursors induces prolonged DSB-1 association with chromosomes.
Many mutations that result in extension of the DSB-1 zone also cause elevated oocyte apoptosis, which can be triggered in response to persistent DNA damage or asynapsis 43,50,68,72. We considered the possibility that apoptosis might mediate the observed extension of DSB-1 staining, since this process primarily culls nuclei from the late pachytene, DSB-1 negative region of the gonad (reviewed in 73). To test this idea, a representative subset of meiotic mutations, including spo-11(ok79), msh-5, syp-2, him-8, and zim-2 (see below) were combined with ced-4(n1162), which abrogates germline apoptosis 49. These double mutants displayed extended DSB-1 localization similar to that observed in the corresponding single mutants (Figure S6). We conclude that apoptosis does not account for the extension of DSB-1 staining observed in crossover-defective mutants, nor can it explain the quantitative differences observed among different mutants.
To further characterize the extension of DSB-1 localization that occurs in response to defects in crossover formation, we examined mutant situations in which crossover formation was disrupted on only one chromosome. him-8(tm611) and zim-2(tm574) specifically disrupt homolog pairing and thus crossover formation on chromosomes X and V, respectively 53,74. him-5(ok1896) does not impair pairing or synapsis, but abrogates DSBs on the X chromosome 36. All three of these mutations extended the DSB-1 zone to 83–86% of the LZP region (Figure 6A and 6B). Furthermore, irradiation of him-5 animals, in which the crossover defect can be rescued by exogenous DSBs 36, but not irradiation of him-8, suppressed the extension of DSB-1 localization (Figure 6C). These results indicate that the absence of a crossover precursor on a single chromosome pair is sufficient to prolong DSB-1 association with meiotic chromosomes.
Analysis of mutants with chromosome-specific defects in interhomolog recombination also allowed us to test whether DSB-1 staining is specifically prolonged on crossover-deficient chromosomes. In him-5 and him-8 mutants, the autosomes, but not the X chromosomes, are proficient for crossover formation. X chromosomes can be specifically marked in these mutants using HIM-8 antibodies (in him-5 mutants) or by staining for synaptonemal complex components (in him-8 mutants). In both of these genotypes, we observed persistent DSB-1 staining on all chromosomes throughout the region of extended DSB-1 localization (Figure 7A, 7B, and 7C). As in wild-type nuclei, the X chromosome showed weaker DSB-1 staining than the autosomes (Figure 4C and 7A). These findings indicate that the extension of DSB-1 localization is a genome-wide response affecting all chromosomes within the nucleus.
To test whether the extension of DSB-1 localization is regulated by nuclear-intrinsic or extrinsic signals, we examined animals heterozygous for meDf2, a deficiency of the X chromosome pairing center 40. In meDf2/+ hermaphrodites, X chromosome pairing and synapsis is partially compromised, such that approximately half the nuclei achieve full pairing and synapsis by the end of the pachytene region 75. Nuclei with asynapsed X chromosomes can be recognized by their more condensed, transition zone-like chromosome morphology, or by co-staining for axial element and central region proteins of the synaptonemal complex 75. In the late pachytene region of these animals, we found that DSB-1 staining correlated with the status of individual nuclei: those with asynapsed chromosomes were positive for DSB-1 staining, while fully synapsed nuclei lacked DSB-1 staining (Figure 7D). These results indicate that the extension of DSB-1 localization is a response to a signal intrinsic to individual nuclei, and does not extend to neighboring nuclei within the same region of the gonad. However, as in all mutants examined, DSB-1 disappeared by the end of the pachytene region of the gonad, indicative of an extrinsic, spatially regulated “override” signal that triggers progression to late pachytene and loss of the presumptive DSB-permissive state, even when crossover precursors have not been attained on all chromosomes (see Discussion).
The DSB-1 paralog DSB-2 is also involved in meiotic DSB formation 47. As reported in the accompanying paper by Rosu et al., the two proteins show very similar localization patterns (Figure 8A and 8B, 47). Both localize to nuclei from leptotene/zygotene through mid pachytene, although DSB-1 staining appears slightly earlier than DSB-2 staining (Figure 8A). They also disappear simultaneously from meiotic chromosomes, both in wild-type animals and various mutants that disrupt crossover formation (Figure 8A, data not shown). Additionally, both proteins show similar distributions along meiotic chromosomes (Figure 8B). Intriguingly, however, the two proteins do not extensively colocalize, but instead rarely coincide (Figure 8B).
To probe the functional interactions between DSB-1 and DSB-2, we localized each protein in the absence of the other. We found that DSB-1 localized to chromosomes in dsb-2(me96) mutants, although the fluorescence intensity was reduced relative to wild-type gonads (Figure 9A and 9B; see also 47). The DSB-1 positive region of the gonad was also somewhat shorter (Figure 9A), despite the reduction of crossovers in dsb-2 mutants 47. This suggests that localization of DSB-1 to meiotic chromosomes does not require, but may be reinforced or stabilized by, DSB-2. By contrast, DSB-2 was not detected on meiotic chromosomes in dsb-1 mutants (Figure 9B). Immunoblotting of whole-worm lysates revealed that DSB-1 protein levels are moderately reduced in dsb-2 mutants, while DSB-2 protein levels are severely reduced in dsb-1 mutants (Figure 9C). This parallels our conclusions from in situ localization of these proteins, and suggests that the reduction of staining observed on chromosomes is a consequence of lower protein levels.
We also tested the effect of eliminating both DSB-1 and DSB-2 by constructing a double mutant strain. The phenotypes observed in dsb-1; dsb-2 mutant animals were indistinguishable from dsb-1 mutants (Figure 10A and 10B). This result is consistent with the idea that these proteins collaborate in some way to promote DSB formation, and argues against more complex epistasis scenarios.
We have discovered a novel protein, DSB-1, required for meiotic DSB formation in C. elegans. Our data indicate that DSB-1 acts specifically to promote DSBs, and does not play a major role in DNA repair or other meiotic processes. DSB-1 localizes to chromosomes during meiotic prophase, concomitant with the period of DSB formation. It appears more abundant on the autosomes than the X chromosome. The significance of this finding is unclear, since DSB-1 is clearly required for DSBs on all chromosomes, but it may be related to observations that the X chromosome has distinct chromatin structure and differential genetic requirements for DSB formation 35,36,51,52.
Both DSB-1 and its paralog DSB-2 are required for normal levels of meiotic DSBs. These proteins show a similar temporal and spatial pattern of localization to meiotic chromosomes. The localization of both proteins is also extended to a similar extent in mutants that disrupt crossover formation. In mutants where the localization of both DSB-1 and DSB-2 was assayed simultaneously, as well as in wild-type animals, the proteins localize to the same subset of meiotic nuclei, except that DSB-1 appears slightly earlier, suggesting that they are co-regulated. However, these proteins seem unlikely to act as a complex, since they show little if any colocalization.
Although DSB-1 and DSB-2 appear to play similar roles in meiotic DSB formation, the severity of their mutant phenotypes are not equivalent. As shown by Rosu et al., DSBs are reduced but not eliminated in young dsb-2 mutant hermaphrodites 47, while dsb-1 mutants lack DSBs regardless of age. The less severe defects observed in young dsb-2 mutants likely reflect the presence of substantial residual DSB-1 protein on meiotic chromosomes in dsb-2 mutants, whereas DSB-2 is not detected on chromosomes in dsb-1 mutants, and protein levels are severely reduced. DSB-1 appears to stabilize DSB-2, perhaps by promoting its association with chromosomes, and to a lesser extent is reciprocally stabilized/reinforced by DSB-2.
The CHK-2 kinase promotes the chromosomal association of DSB-1. CHK-2 is also required for DSB-2 localization on meiotic chromosomes 47, although it is not clear whether CHK-2 promotes DSB-2 loading directly, or indirectly through its role in the loading of DSB-1. Our findings suggest a model in which DSB-1 and DSB-2 mutually promote each other's expression, stability, and/or localization, with DSB-2 depending more strongly on DSB-1, to promote DSB formation (Figure 10C).
The number of sites of DSB-1 and DSB-2 localization per nucleus – too many to quantify in diffraction-limited images – appears to greatly exceed the number of DSBs, estimates of which have ranged from 12 to 75 per nucleus 65,76,77. DSB-1 and DSB-2 may each bind to sites of potential DSBs, with only a subset of these sites undergoing DSB formation, perhaps where they happen to coincide. They could also be serving as scaffolds to recruit other factors required for DSB formation to meiotic chromosomes and/or to promote their functional interaction. This idea is currently difficult to test, since we have not yet been able to detect chromosomal association of SPO-11 in C. elegans, and no other proteins specifically required for DSBs have been identified. Alternatively, these proteins may influence DSB formation by modifying chromosome structure. We did not observe overt changes in chromosome morphology in dsb-1 mutants, but further analysis – e.g., mapping of histone modifications – may be necessary to uncover more subtle changes.
DSBs normally occur within a discrete time window during early meiotic prophase. In C. elegans this corresponds to the transition zone and early pachytene, based on RAD-51 localization. As DSB-1 is necessary for DSB formation, and its appearance on meiotic chromosomes coincides with the timing of DSBs, we infer that the chromosomal localization of DSB-1 is indicative of a regulatory state permissive for DSB formation. We observed that when crossover formation is disrupted, this DSB-1-positive region is extended. Rosu et al. report a similar extension of DSB-2 in crossover-defective mutants 47.
Previous work has shown that RAD-51 foci persist longer and accumulate to greater numbers in various mutants that make breaks but not crossovers 43,64,65,67,75. Extended or elevated RAD-51 staining could reflect an extension of the time that DSBs are made, a greater number of DSBs, or a slower turnover of the RAD-51-bound state. However, persistence of DSB-1 and DSB-2 on meiotic chromosomes in these mutants suggests that the period in which nuclei are competent for DSB formation is extended. In support of this idea, in many crossover-defective mutants defective, the number of RAD-51 foci per nucleus not only reaches a far higher level but also peaks later than in wild-type animals and continues to rise even after RAD-51 is normally cleared from meiotic chromosomes 53,75,77, indicating that breaks continue to be generated after their formation would normally cease. Several mutations that impair crossover formation on a limited number of chromosome pairs result in altered crossover distributions on the crossover-competent chromosomes 64,78, which, particularly in light of the current findings, seems likely to reflect changes in DSB activity. In addition, RAD-51 chromatin immunoprecipitation data from our laboratory (C. V. Kotwaliwale and AFD, unpublished) have indicated that the DSB distribution is altered in him-8 mutants, one of the genotypes that show extended DSB-1 staining.
Taken together, these findings strongly suggest that both the temporal and genomic distribution of DSBs is altered in many situations that perturb crossover formation. A similar phenomenon may also account for the “interchromosomal effect” first observed in Drosophila female meiosis 79. We note that this raises caveats about previously published estimates of DSB numbers in C. elegans that have been based on quantification of RAD-51 foci in genotypes that do not complete crossovers, such as rad-54 or syp-1 mutants 65,76,77.
Extension of the DSB-1 zone occurs even when nuclei are unable to initiate meiotic recombination due to an absence of DSBs. This suggests that the extension is not due to the persistence of unresolved recombination intermediates, but is instead a response to the absence of a particular crossover-competent recombination intermediate, or crossover precursor. We found that disruption of crossover formation on a single pair of chromosomes is sufficient to prolong DSB-1 localization on all chromosomes. Based on this result we believe that chromosomes lacking a crossover precursor may emanate a signal that sustains a DSB-permissive state within the affected nucleus. Thus, a single chromosome pair lacking a crossover precursor elicits a genome-wide response that results in extension of DSB-1 localization, which may reflect a modulation of the timing, and perhaps the extent, of DSB formation. Such a mechanism would help to ensure formation of an “obligate” crossover on every chromosome pair.
All mutants that we found to extend the localization of DSB-1 cause a disruption in crossover formation, although they have various primary molecular defects. It is possible that extension of DSB-1 localization occurs in response to distinct molecular triggers in different mutant situations. For example, spo-11 mutants may be responding to an absence of DSBs, rad-54 mutants to unrepaired DSBs, and syp-1 mutants to asynapsed chromosomes. A similar model in which different unfinished meiotic tasks can elicit delays in meiotic progression was proposed in a recent study 80. However, we feel that a parsimonious interpretation of our data is that the absence of a crossover precursor on one or more chromosomes is sufficient to prolong DSB-1/2 localization. The varying degree of extension seen in different mutants could reflect the engagement of additional regulatory mechanisms, such as the synapsis checkpoint and/or DNA damage checkpoint, which might converge with a crossover assurance mechanism to modulate regulators of DSB-1.
We propose that an “obligate crossover” checkpoint mediates the extension of DSB-1 localization (Figure 11). Our data suggest that DSB formation is activated during early meiosis and normally persists long enough for most nuclei to attain crossover precursors on all chromosomes (Figure 11). If interhomolog recombination is impaired on one or more chromosome pairs, individual nuclei can prolong the DSB-permissive state in an attempt to generate a crossover on every chromosome. Our observation that a block to crossover formation on a single pair of chromosomes results in persistent DSB-1 throughout the affected nuclei is reminiscent of the spindle assembly checkpoint (SAC), in which failure of a single pair of sister kinetochores to biorient on the mitotic spindle triggers a cell-autonomous delay in anaphase onset that affects cohesion on all chromosomes 81. Interestingly, a key mediator of the SAC, Mad2, is homologous to the meiotic axis proteins HTP-3 and HTP-1 28,82, which appear be important for the regulatory circuit that mediates prolonged DSB-1 localization in response to crossover defects.
An alternative model would be a negative feedback circuit in which the acquisition of all necessary crossover-intermediates triggers inactivation of DSB formation. According to this view, the presence of crossover precursors generates a signal to exit the DSB permissive state, rather than the absence of precursors extending this period. Such a model would require a ‘counting’ mechanism that enables exit from the DSB permissive state in response to a threshold number of crossover precursors. This seems less likely based on first principles, and also less consistent with our data.
Our observations also suggest that there is a minimum duration of proficiency for DSB formation that does not depend on how rapidly chromosome pairs attain crossover precursors. We would expect meiotic nuclei to achieve crossover precursors on every chromosome in a stochastic manner. If DSB-1 were removed from chromosomes upon reaching this state, we would likely see a patchwork of DSB-1 positive and negative nuclei in the early pachytene region, but instead we observe homogenous staining in this region, and abrupt disappearance of DSB-1 within a narrow zone of the gonad. Additionally, in mutants that appear to be defective in triggering the obligate crossover checkpoint, such as htp-3 and htp-1, a zone of DSB-1-positive nuclei similar in length to that in wild-type animals is observed. Together these observations suggest that there is a preset temporal window for DSB formation that can be extended in individual nuclei but not shortened.
The duration of the DSB-permissive state might be specified by an activity or signal that decays with time and/or distance after meiotic entry. We speculate that the disappearance of DSB-1 may reflect a drop below a threshold level of CHK-2 activity, decay of CHK-2-mediated phosphorylation of DSB-1 or other targets, and/or a rise in an opposing activity – e.g., a phosphatase. Any of these could be inhibited by the putative checkpoint mechanism that prolongs DSB-1 localization in response to impaired crossover formation.
The nature of the recombination intermediate that satisfies the requirement for a crossover precursor on all chromosomes remains unknown. We distinguish “crossover precursors” from “interhomolog recombination intermediates” because components that are specifically required for crossovers, including MSH-5, ZHP-3, and COSA-1 38,39,66,83, are all required for timely disappearance of DSB-1 from chromosomes. However, cytological markers for crossovers, including foci of ZHP-3 and COSA-1, do not appear until the late pachytene region of the gonad 39,66, after DSB-1 and DSB-2 disappear from meiotic chromosomes 47. Thus, it seems likely that crossover precursors, rather than mature crossovers, are sufficient to allow exit from the DSB-permissive state.
Genetic and cytological evidence indicate that nuclei eventually cease to make DSBs, even when crossovers fail to be made on one or more chromosomes. As nuclei approach the bend region of the gonad at the end of pachytene, an “override” signal appears to shut off DSB formation (Figure 11). Unlike in mammals, where crossover failures result in extensive apoptosis 84, C. elegans hermaphrodites produce both sperm and oocytes in roughly normal numbers even when homolog pairing, synapsis, and/or recombination are severely impaired.
Numerous studies have documented a phenomenon known as the “extended transition zone” in mutants with defects in homolog pairing and/or synapsis 30,43,53,68. An extended transition zone has been defined as a longer region of the gonad containing nuclei with crescent-shaped DAPI-staining morphology, multiple patches of the nuclear envelope proteins SUN-1 and ZYG-12, and strong foci of the ZIM proteins 68,74,85. An extended transition zone appears to be a response to asynapsed chromosomes 43,68. Previous work from our lab showed that the extension of the transition zone in synapsis-defective animals such as him-8 hermaphrodites was suppressed by mutations in recombination factors, including spo-11 and msh-5, and we therefore proposed that it might reflect a response to unresolved recombination intermediates 64. However, subsequent work has revealed that these double mutant situations actually resulted in precocious fold-back synapsis of unpaired chromosomes, thereby silencing the asynapsed chromosome response (SER and AFD, unpublished).
Since mutations that abrogate pairing or synapsis also impair interhomolog recombination, it is not surprising that most genotypes with extended transition zones also show persistent DSB-1 localization. However, not all mutants that disrupt crossover formation extend the transition zone. spo-11 and him-5 mutants, for example, are deficient for DSB formation on one or more chromosomes and show extended DSB-1 staining, but do not show typical extended transition zones. Instead, these mutants appear to have extended regions of early pachytene nuclei. Based on these observations, we believe that the obligate crossover checkpoint mechanism is distinct from the response to asynapsed chromosomes. However, these two regulatory circuits serve similar purposes – to enable meiotic nuclei more time to complete synapsis or achieve crossovers on all chromosomes – and they may also involve common molecular components.
Proteins with apparent homology to DSB-1 are restricted to the Caenorhabditis lineage. Even within Caenorhabditids, DSB-1, DSB-2 and their homologs are only weakly conserved. This reinforces abundant evidence from other organisms that apart from Spo11 itself and the Rad50-Mre11 complex, proteins that promote DSB formation diverge rapidly during evolution 17,18,86. This might seem surprising given that meiotic DSB formation is an essential aspect of sexual reproduction in most eukaryotes. However, potent and acute evolutionary pressures act on meiosis. For example, the germline is the site of intense warfare between the host genome and selfish genetic elements, which may contribute to the rapid evolution of meiotic proteins. In addition, the genome-wide distribution of DSBs appears to underlie the strongly biased distribution of crossovers observed in many species 87,88, including C. elegans (C. V. Kotwaliwale and AFD, unpublished). The nature of this biased distribution shows interesting variation among species 89,90. Since crossover number and position have a direct impact on the fidelity of meiotic chromosome segregation, mechanisms governing DSB distribution have likely evolved in concert with changes in chromosome structure and the spindle apparatus to maintain reproductive fitness.
Several features of meiosis in C. elegans distinguish it from other organisms in which DSB-promoting factors have been identified. In particular, DSBs and early recombination steps contribute directly to homolog pairing and synapsis in many species, while in C. elegans homolog pairing and synapsis occur independently of DSBs. Additionally, C. elegans lacks Dmc1, Hop2, and Mnd1, which are thought to function together as an essential meiotic recombination module in most eukaryotes 91. C. elegans also lacks the DSB proteins Mei4 and Rec114, which are conserved between budding yeast and mice 18. A correlation between the absence of DMC1/Hop2/Mnd1 and Mei4/Rec114 has been noted in several other lineages, and has been suggested to reflect a functional link between the formation of DSBs and their subsequent repair 18. Interestingly, Rec114, like DSB-1/2, has several potential target sites for ATM/ATR phosphorylation, and these are important for regulation of DSBs in budding yeast meiosis 14. Thus, the DSB-1/2 family of proteins may play analogous roles to known mediators of DSB formation in other species, despite their lack of apparent sequence similarity.
All C. elegans strains were cultured under standard conditions at 20°C. The wild-type strain was N2 Bristol. The nonsense dsb-1(we11) allele was generated by EMS mutagenesis. The dsb-1 deletion allele (tm5034) was generated by the Japanese National BioResource for the Nematode. Both dsb-1 alleles were extensively outcrossed to wild-type (5-6x), and additionally outcrossed in a directed three-point cross to dpy-20 unc-30 to eliminate most linked mutations. Additional mutants analyzed in this study were: spo-11(me44, ok79), mre-11(ok179), rad-50(ok197), chk-2(me64), com-1(t1626), rad-51(Ig8701), rad-54(tm1268), msh-5(me23), cosa-1(me13), zhp-3(jf61), htp-3(tm3655), htp-1(gk174), syp-1(me17), syp-2(ok307), him-8(tm611), zim-2(tm574), dsb-2(me96), and ced-4(n1162). Strains used in this study were:
L4 hermaphrodites were picked onto individual plates and transferred to new plates every 12 hours, for a total of 6–8 12-hour laying periods, until newly-laid fertilized eggs were no longer observed. Eggs were counted immediately after each 12-hour laying period. Surviving hermaphrodite and male progeny were counted 3 days later.
Polyclonal antibodies against recombinant full-length DSB-1 protein were produced at Pocono Rabbit Farm & Laboratory. 6xHis-DSB-1 was purified from E. coli using Ni beads under denaturing conditions. The protein was resolved on an SDS-PAGE gel and the excised DSB-1 band was used to immunize guinea pigs. Rabbit anti-HTP-3 antibodies were raised against a synthetic peptide (PTEPASPVESPVKEQPQKAPK) by Strategic Diagnostics Inc., SDIX. Additional antibodies used in this study were: guinea pig anti-HTP-3 75, rat anti-HIM-8 53, rabbit anti-RAD-51 92, goat anti-SYP-1 92, and rabbit anti-DSB-2 47.
Immunofluorescence was performed as previously described 93. Briefly, hermaphrodites 24–28 hours post L4 were dissected in egg buffer containing sodium azide and 0.1% Tween 20, fixed for 3 min in 1% formaldehyde in the same buffer between a Histobond slide and coverslip, and frozen on dry ice. The coverslip was removed, and slides were transferred to methanol chilled to −20°C. After 1 min, slides were transferred to PBST (PBS containing 0.1% Tween 20), washed in two further changes of PBST, blocked with Roche blocking agent, and stained with primary antibodies in block for 2 hours at room temperature or overnight at 4°C. Slides were then washed with 3 changes of PBST and stained with secondary antibodies. Secondary antibodies labeled with Alexa 488, Cy3, or Cy5 were purchased from Invitrogen or Jackson ImmunoResearch. Following immunostaining, slides were washed, stained in 0.5 mg/ml DAPI, destained in PBST, and mounted in buffered glycerol-based mounting medium containing 4% n-propyl gallate as an antifading agent.
For quantification of DAPI-staining bodies in oocytes, animals were dissected, fixed, and DAPI-stained as described above, omitting the steps involving immunostaining.
FISH procedures have also been previously described in detail 93. Probes used in this study included the 5S rDNA repeat 23 and a short repeat associated with the right end of the X chromosome 53.
All images were acquired using a DeltaVision RT microscope (Applied Precision) equipped with a 100× 1.40 oil-immersion objective (Olympus) or (for whole gonad images) a 60× 1.40 oil-immersion objective (Olympus). Image deconvolution and projections were performed with the softWoRx software package (Applied Precision). Image scaling, false coloring, and composite image assembly were performed with Adobe Photoshop. All micrographs presented in the figures are maximum-intensity projections of 3D data stacks.
Lysate from 50 young adult hermaphrodites, picked at 24 hours post L4, was used for each lane. Gel electrophoresis was performed using 4–12% Novex NuPage gels (Invitrogen). Proteins were transferred to PVDF membrane. Guinea pig DSB-1 antibodies and rabbit DSB-2 antibodies (see above) were used for immunoblotting, followed by detection with HRP-conjugated secondary antibodies and ECL Western Blotting Substrate (Pierce).
Young adult worms were irradiated with approximately 10 Gy (1000 rad) from a Cs-137 source. For each experiment, unirradiated controls were treated identically to irradiated animals, other than exposure to radiation. For quantification of DAPI-staining bodies at diakinesis, hermaphrodites were irradiated 4–5 hours post L4 and dissected 18 hours post irradiation. To assess progeny survival, animals were irradiated 4–5 hours post L4, eggs laid 20–30 hours post irradiation were quantified, and surviving progeny were quantified 3 days later. For quantification of DSB-1 localization, animals were irradiated 16 hours post L4 and dissected 8 hours post irradiation. For RAD-51 immunofluorescence, animals were irradiated 24 hours post L4 and dissected 1 hour post irradiation.
1000 homozygous we11 animals were picked from an outcrossed, balanced strain. A genomic DNA library was prepared as described in the genomic DNA library protocol from Illumina. Libraries were sequenced using 76-bp single-end Illumina sequencing. MAQGene 94 was used to identify mutations present in the we11 mutant strain.
A 2.1-kb region of genomic DNA including the dsb-1 coding sequence and promoter was amplified by PCR using the following primers: 5′-CCGCTTCCGAATACCGCC-3′ and 5′-GGTGCCGCTGTGTAGAAGAAGC-3′. 100 ng/µl of dsb-1 PCR product was combined with 50 ng/µl of unc-119 rescuing plasmid pMM051 95 and injected into unc-119 animals. Rescued non-Unc F1 animals were picked to individual plates and assayed for embryonic lethality and male progeny. F2 animals were dissected, stained, and observed to quantify the number of DAPI-staining bodies in oocytes at diakinesis.
12 young adult animals, 24 hours post L4, were used for each genotype. RNA was purified from animals and reverse transcribed into cDNA with the SuperScript kit from Invitrogen using poly-A primers. spo-11 mRNA levels were compared by real-time PCR analysis with SYBR Green (Kapa Biosystems). act-1 and htp-3 mRNA levels were used as normalization controls. Primers used were as follows: spo-11 (5′-TGAGCCCGGATCTGTAGAAT-3′, 5′-TAGCTTGTTCCTTCGGTGGT-3′), act-1 (5′-CCCCATCAACCATGAAGATC-3′, 5′-TCTGTTGGAAGGTGGAGAGG-3′), and htp-3 (5′-CGAGTGATGACAGGGCTATATTC-3′, 5′-TGCAAGATAAACGCAGTTGG-3′).
|
10.1371/journal.ppat.1000209 | Role of Amphipathic Helix of a Herpesviral Protein in Membrane Deformation and T Cell Receptor Downregulation | Lipid rafts are membrane microdomains that function as platforms for signal transduction and membrane trafficking. Tyrosine kinase interacting protein (Tip) of T lymphotropic Herpesvirus saimiri (HVS) is targeted to lipid rafts in T cells and downregulates TCR and CD4 surface expression. Here, we report that the membrane-proximal amphipathic helix preceding Tip's transmembrane (TM) domain mediates lipid raft localization and membrane deformation. In turn, this motif directs Tip's lysosomal trafficking and selective TCR downregulation. The amphipathic helix binds to the negatively charged lipids and induces liposome tubulation, the TM domain mediates oligomerization, and cooperation of the membrane-proximal helix with the TM domain is sufficient for localization to lipid rafts and lysosomal compartments, especially the mutivesicular bodies. These findings suggest that the membrane-proximal amphipathic helix and TM domain provide HVS Tip with the unique ability to deform the cellular membranes in lipid rafts and to downregulate TCRs potentially through MVB formation.
| Herpesvirus persists in its host by entering a latent state, periodically reactivating to produce infectious viral particles. Some of the herpesviruses have also been known to be related to cancers. Herpesvirus saimiri (HVS), an oncogenic monkey herpesvirus, persists in the T lymphocytes of its natural host, the squirrel monkey, without any apparent disease symptoms, but infection of other species of New World and Old World primates results in fulminant T cell lymphomas. Two viral oncoproteins, Saimiri Transforming Protein and Tyrosine kinase-interacting protein (Tip), are required for T cell transformation. It has been known that Tip may also play some role in viral persistency within T cells by inhibiting the activation of the host cells upon antigenic stimulation. Here, we have identified a structural domain, a putative amphipathic helical motif, preceding the transmembrane domain of Tip. We also found that the structural motif is essential for Tip's localization on specialized membrane domains, lipid rafts, and selective downregulation of antigen receptors. Furthermore, we could genetically dissect the functional roles of the amphipathic helical motif and transmembrane domain of Tip in membrane deformation and oligomerization, respectively. These findings significantly advanced our understanding of how herpesvirus modulates host lymphocytes for viral persistence and pathogenesis.
| Lipid rafts are membrane microdomains that take part in coordinating cell signalling functions and membrane trafficking. In T cells, upon antigenic stimulation, T cell receptors (TCRs) are recruited to lipid rafts, where they transmit signals via several pathways. The TCR signals induce the anchoring of lipid rafts to the underlying actin cytoskeleton, resulting in the assembly of lipid rafts [1]. Subsequently, clustered lipid rafts, containing TCR/CD3 complexes, are subjected to endocytosis, and the TCR/CD3 complexes are targeted for lysosomal degradation [2]. Thus, current evidence indicates that lipid rafts function as platforms for both the signalling and endocytosis of activated TCRs. Despite the important role of lipid rafts in signalling and membrane trafficking in T cells, the regulatory mechanisms controlling membrane trafficking to lysosomal compartments remain unclear. Several biochemically distinct compartments for membrane trafficking have been identified in other cell types including primary endocytic vesicles, early endosomes, late endosomes, and lysosomes. It has been recently demonstrated that multivesicular bodies (MVB), also known as vesiculated late endosomes, are required for many key trafficking processes such as the downregulation of activated signalling receptors [3]. However, difficulties in elucidating the mechanisms of membrane trafficking have been compounded in T cells, because the fate of endocytic vesicles and the dynamics of transport intermediates remain uncertain.
Herpesvirus persists in its host by entering a latent state, periodically reactivating to produce infectious viral particles. Herpesvirus saimiri (HVS), an oncogenic γ2 herpesvirus, persists in the T lymphocytes of its natural host, the squirrel monkey, without any apparent disease symptoms, but infection of other species of New World and Old World primates results in fulminant T cell lymphomas [4]. In addition, when HVS infects the primary T lymphocytes of humans, Old World primates, New World primates, or rabbits, it can immortalize infected T cells, allowing them to grow independently of IL-2 [5].
Tyrosine kinase-interacting protein (Tip) is encoded in the first open reading frame at the left end of the highly oncogenic strains of HVS. Tip is not required for viral replication, but is required for T cell transformation in cultures, and for lymphoma induction in primates [4]. Tip has multiple binding sites for cellular proteins. The interaction of Tip with Lck kinase, which is mediated by the Src homology 3-binding (SH3B) motif and C-terminal Src-related kinase homology (CSKH) domain of Tip [6],[7], interferes with early events in the TCR signal transduction pathway, resulting in inhibition of immunological synapse formation [8]. Tip also interacts with p80, a novel cellular endosomal protein that contains an N-terminal WD repeat domain and a C-terminal coiled-coil domain [9]. The interaction of Tip with p80, which is mediated by a region containing a serine-rich (SR) motif, facilitates the formation of enlarged lysosomal vesicles, and results in the targeting of Lck and TCR/CD3 complexes for lysosomal degradation. We have previously demonstrated that Tip constitutively localizes in lipid rafts and exploits Lck and p80 to recruit TCR/CD3 complexes, leading to lipid raft aggregation and internalization [10]. Constitutive localization of Tip in lipid rafts depends on the C-terminal transmembrane (TM) domain, but not Lck and p80 interaction, and is also necessary for the efficient downregulation of TCR/CD3 and CD4 surface expression without affecting the inhibition of TCR signal transduction [11].
In this study, we report the presence of a putative amphipathic helical motif preceding the TM domain of Tip. Structural analysis revealed that Tip's amphipathic helical motif is composed of hydrophobic and positively-charged amino acid residues. Recently, the amphipathic helical motif has attracted much attention due to its active role in membrane curvature formation and membrane trafficking [12]. Thus, we investigated roles of the amphipathic helical motif in the molecular functions of Tip, including lipid raft localization and downregulation of TCR/CD3 and CD4. We found that the membrane-proximal amphipathic helical motif is required for the efficient localization of Tip in lipid rafts as well as its selective downregulation of TCR/CD3, potentially through deformation of membrane structures and MVB formation in T cells.
We have recently reported that the TM domain (amino acid residues 229–250) of Tip is required for its association with lipid rafts, while other motifs involved in interactions with Lck and p80 are dispensable for lipid raft targeting [11]. In this study, GFP-Tip fusion proteins carrying deletions from the cytoplasmic region of Tip were generated (Figure S1), and the motif required for lipid raft localization was mapped. The degree to which Tip was associated with lipid rafts, in 293T cells, was estimated by densitometry and represented as the average percentage value from triplicate samples. The position and the integrity of lipid rafts in the discontinuous sucrose gradient were determined by the presence of GM1 ganglioside, which associates reliably with lipid rafts (Figure 1A). We found that the wild type GFP-Tip fusion protein was efficiently associated with lipid rafts (approximately 75%), which is consistent with our previous results [10],[11]. Deletion of Tip's cytoplasmic domain, which contains known protein interaction motifs, had no discernible effect on the association of Tip with lipid rafts (GFP-Tip184-256). Two additional deletion mutants, GFP-Tip197-256 and GFP-Tip211-256, showed a similar Tip distribution; approximately 50% was associated with lipid rafts. A GFP-Tip227-256 mutant carrying only the Tip TM domain was detected primarily in the fractions where cytoplasmic GFP protein localized, and showed only 20% association with lipid rafts. These results suggest that the C-terminal cytoplasmic regions proximal to the TM domain might contribute significantly to Tip's localization to lipid rafts.
To exclude the potential confounding effects by GFP fusion on lipid raft association of Tip, we also constructed flag-tagged version of Tip mutants and examined their localization on lipid rafts (Figure S2). In this independent experiment, similar level of lipid raft association of flag-tagged Tip mutants was observed when compared with those of GFP-fusion proteins, indicating that GFP fusion does not significantly affect the lipid raft association of Tip and its mutants.
We next analyzed the potential structure of the C-terminal regions spanning residues 184 to 256 of Tip using a protein structure prediction server [13]. The secondary structure analysis predicted with high confidence that the amino acid residues from 213 to 250 of Tip would form an α-helix (Figure S3). Notably, the amino acid residues from 213 to 228, proximal to the TM domain, are composed of hydrophobic and positively-charged amino acid residues, and are predicted to form an amphipathic helical structure (Figure 1B). This amphipathic helical motif was highly conserved in Tip from three different strains of HVS, and in Tio (Two-in-one) of Herpesvirus ateles (HVA), a recently identified member of the γ2-herpesvirus family (Figure 1C and Figure S3). Tio, an oncoprotein of HVA, has been shown to induce transformation of T cells in a manner similar to that seen in StpC and Tip of the C488 strain of HVS [14]. These findings suggest that this potential amphipathic helical motif preceding the TM domain might play a role in Tip function.
To evaluate the effect of the amphipathicity of Tip's membrane-proximal helical motif upon its lipid raft localization, this motif was mutated by replacing the hydrophobic and charged residues with lysine and alanine, respectively (Figure 2A). The resulting ability of the Tip mutants to associate with lipid rafts was then assessed. When Tip's four conserved hydrophobic residues (I216, L220, L223, and I227), predicted to form hydrophobic face, were replaced with lysines (Tip amp1), there was a ∼50% reduction in lipid raft association, in comparison to wild type Tip (Figure 2B). The degree of lipid raft association observed in this mutant was similar to that observed in the GFP-Tip227-256 mutant, carrying only the TM domain of Tip (Figure 1A). As such, these data suggest that the four conserved hydrophobic residues are critical for targeting of Tip to the lipid rafts. Sequential replacement of two or three consecutive, topologically-adjacent hydrophobic residues with lysine residues (Tip amp1-2K.1∼Tip amp1-3K.2) resulted in a gradual reduction in Tip's lipid raft association, ranging from 16% to 54% of the degree of association observed in wild type Tip. Both Tip amp1-2K.3 and Tip amp1-3K.2 mutants carrying lysines proximal to the TM domain showed a low degree of lipid raft association, similar to that observed in Tip amp1, indicating that the hydrophobic isoleucine and leucine residues proximal to the TM domain are more critical for the association with lipid rafts than are the more distal residues. Substitution of the positively-charged residues with alanine (Tip amp2: R214, K218, K221, and R222) also resulted in a ∼40% reduction in lipid raft association compared to wild type, demonstrating the significant contribution of these residues to Tip's localization to lipid rafts.
Tip-mediated downregulation of TCR/CD3 and CD4 depends on its ability to associate with lipid rafts [11]. To examine the contribution of amphipathicity of Tip's membrane-proximal helix to this downregulation, levels of TCR, CD3, CD4, and CD45 surface expression were examined in Jurkat T cells stably expressing wild type Tip, Tip amp1, or Tip amp2, using flow cytometry. As shown previously [9], expression of wild type Tip in T cells effectively downregulated the surface expression of TCR/CD3 and CD4 (Figure 3A). In striking contrast, the downregulation of TCR and CD3 surface expression was severely impaired in Jurkat T cells expressing Tip amp1 or Tip amp2 (Figure 3A). However, downregulation of CD4 was not significantly affected by the mutations which abolished the amphipathicity of the membrane-proximal helix. Neither wild type Tip nor its mutants had any significant effect upon the surface expression of CD45, demonstrating the specificity of Tip's effects for TCR and CD3 downregulation. These results indicate that the amphipathicity of Tip's membrane-proximal helix is involved in the downregulation of TCR/CD3, but not CD4 surface expression.
We have previously shown that Tip's targeting to the lysosomal compartments involves its formation of a complex containing Lck and p80 [9]. Tip's formation of this complex is correlated with its lipid raft association and the lysosomal degradation of TCR/CD3 complexes [10],[11]. To examine whether loss of amphipathicity in Tip's membrane-proximal helix might affect the lysosomal localization of the viral proteins and TCR/CD3, Jurkat T cells transiently expressing Tip or Tip amp1 were reacted with antibodies specific to EEA1, an early endosomal marker, LAMP2, a late endosomal/lysosomal marker, or CD3ζ and then examined under a confocal microscope (Figure 3B). To quantitatively compare the degree of colocalization of the proteins in the vesicular compartments, we measured the Pearson correlation coefficient (R) values (see Materials and Methods) for each set of colocalizing proteins in 10 to 20 cells (Figure S4). Vesicles containing wild type Tip were weakly colocalized with EEA1 (average R value = 0.16), but were strongly colocalized with LAMP2 (R = 0.76) or CD3ζ (R = 0.82), as shown previously [10]. In contrast, Tip amp1 displayed partial colocalization with EEA1 (R = 0.51) and CD3ζ (R = 0.60) but did not colocalize with LAMP2 (R = 0.02), indicating that the amphipathicity of Tip's membrane-proximal helix is required for efficient lysosomal targeting of Tip and TCR/CD3.
We have previously shown that Tip's TM domain is required for its lysosomal trafficking [11]. To investigate the role of the membrane-proximal amphipathic helix and TM domain in Tip's lysosomal localization, Jurkat T cells expressing GFP-Tip211-256, GFP-Tip amp1211-256, or GFP-Tip CD71TM211-256 were reacted with antibodies specific to EEA1 or LAMP2 (Figure 4A). Colocalization of the fusion proteins with the endocytic markers were evaluated quantitatively for Pearson correlation coefficient (Figure S5). The intracellular vesicles containing GFP-Tip211-256 colocalized strongly with LAMP2 (R = 0.82), but weakly with EEA1 (R = 0.21), suggesting that the amphipathic helix and TM domain are sufficient for Tip fusion proteins to be delivered into the late endosomes or lysosomes. In fact, almost all of the intracellular vesicles containing GFP-Tip211-256 were costained with LAMP2 in the transfected T cells (data not shown). Unlike GFP-Tip211-256, both GFP-Tip amp1211-256 and GFP-Tip CD71TM211-256, a Tip mutant carrying the TM domain of CD71 in place of that of Tip (Figure S1), were only partially localized in LAMP2-positive vesicles, showing slight preferential colocalization with EEA1 (R values range from 0.34 to 0.46, Figure 4A and S5A). These results were further confirmed in HeLa cells expressing these fusion proteins (Figure 4B and S5B). The colocalization of lysosomes with vesicles containing GFP-Tip211-256 (R = 0.54) was also more prominent than with vesicles containing GFP-Tip amp1211-256 (R = 0.30) or GFP-TipCD71TM211-256 (R = 0.37), in HeLa cells.
To determine in more detail the distribution of the GFP fusion proteins, Jurkat T cells expressing GFP-Tip211-256 or GFP-Tip amp1211-256 were analyzed by immunoelectron microscopy after staining with gold-conjugated anti-GFP antibodies. The gold signal was generally associated with intracellular vesicular compartments. Interestingly, GFP-Tip211-256 was frequently detected in luminal buddings of vesicular membranes, or in membranous complexes within the lumen (Figure 5), reminiscent of the process of MVB formation [15]. MVBs form by budding into the lumen of the vacuolar endosomes, which carry membrane proteins selected for the late endosomal route. They are thought to fuse with late endosomes or, following maturation, directly with lysosomes [15]. In cells expressing GFP-Tip amp1211-256, GFP was generally associated with smaller vesicular compartments, most likely the early endosomes, as shown in Figure 4. Associations of membrane curvature or MVBs with the fusion proteins were barely detectable (Figure 5). Taken together, it appears that the peptide encompassing the membrane-proximal helix and the TM domain of Tip might be involved in MVB formation in late endosomal compartments where Tip and its complex are degraded.
Membrane curvature is an active means for creating membrane domains and organizing trafficking [12]. Several mechanisms have been suggested to constitute active cellular processes for the formation of membrane curvature, and these include changes in lipid composition, oligomerization of curvature scaffolding proteins, and the insertion of amphipathic helices into the lipid bilayer [12],[16]. The possibility that the membrane-proximal amphipathic helix of Tip interacts with lipids was examined using a lipid binding assay. As shown in Figure 6A, a synthetic peptide derived from the membrane-proximal amphipathic helix of Tip (Tip wt211-228) was found to bind to a series of negatively charged lipids including phosphatidic acid (PA), phosphatidylserine (PS), phosphatidylglycerol (PG), cardiolipin, phosphatidylinositide (PtdIns), and sulfatide, but did not bind to other neutral or positively-charged lipids such as triglyceride (TG), diacylglycerol (DAG), phosphatidylenthanolamine (PE), phosphatidylcholine (PC), cholesterol, or sphingomyelin. The binding specificity of the peptide to these lipids was further examined by probing an array of several lipids immobilized on nitrocellulose membranes at concentrations ranging from 100 pmol to 6.2 pmol. As demonstrated in Figure 6B, both wild type (Tip wt211-228) and mutant (Tip amp1211-228) peptides, carrying lysine residues instead of hydrophobic amino acids, were able to bind dose-dependently to PA, PS, and PG, with binding saturation occurring at approximately 50 pmol of lipids (Figure 6B and Figure S6). Interestingly, the mutant peptides derived from Tip amp1211-228 were also capable of binding to PE. The peptide-lipid interactions were further validated in a liposome binding assay, in which liposomes composed of 65% PC, 25% PS and 10% cholesterol were reacted with biotin-conjugated peptides, then cosedimented by ultracentrifugation. The coprecipitated peptides were resolved by gel electrophoresis and were subsequently probed with streptavidin-HRP conjugates. As shown in Figure 6C, approximately 50% of Tip wt211-228 peptides were cosedimented with liposomes, whereas less than 5% precipitated in the absence of liposomes. The Tip amp1211-228 peptides also precipitated after incubation with liposomes, even more efficiently than did the amphipathic wild type peptide. These data suggest that the hydrophobic residues of the amphipathic helix had little effect upon peptide-lipid binding properties. It remains a possibility, however, that these residues might restrict the affinity and preference of Tip for specific lipids.
The influence of the amphipathic helix upon membrane curvature formation was examined using a liposome-based membrane deformation assay [17],[18],[19]. The peptides were incubated with liposomes, which have the same lipid composition as those used in Figure 6, and subsequently examined by electron microscopy (Figure 7). Tip wt211-228 peptides resulted in efficient and robust formation of tubules with diameters of 20–40 nm, whereas Tip amp1211-228 did not. A laser light scattering assay revealed that incubation with Tip wt211-228 peptides dramatically altered the size distribution of the liposomes into ranges of 40–250 nm, whereas no significant changes in liposome size were detected following incubation with Tip amp1211-228 (Figure 7 and Figure S7).
Collectively, the reported results indicate that the positively-charged residues within Tip's amphipathic helix confer a specific affinity for negatively-charged phospholipids, potentially through ionic interactions. The conserved hydrophobic residues enable the amphipathic helix to act like a wedge inserted into the membrane, to induce membrane curvature.
Previously, we demonstrated that Tip can induce the aggregation of lipid rafts and enhance the recruitment of lipid raft-resident proteins, eventually forming large vesicular compartments in T cells [9],[11]. These results suggested that Tip might oligomerize within membrane microdomains, inducing structural changes in the lipid bilayer. As such, the possibility of Tip oligomerization was investigated by coexpressing flag-tagged wild type Tip and GFP-Tip fusion proteins, then immunoprecipitating with an anti-flag antibody (Figure 8A). Immunoblotting with an anti-GFP antibody revealed that the flag-tagged Tip co-precipitated with the GFP-Tip fusion protein, but not with GFP, suggesting that Tip interacts with itself. To determine the region responsible for Tip oligomerization, GFP-Tip mutants were included in the immunoprecipitation assay [11]. GFP-Tip mutants no longer binding with Lck (TipmLBD) or p80 (TipΔ2) formed an immune complex with flag-tagged wild type Tip, indicating that Tip's Lck- and p80-binding motifs do not participate in Tip oligomerization. However, a Tip mutant carrying the TM domain of CD71 (Tip CD71TM) in place of its native one failed to interact with wild type Tip, suggesting that Tip's TM domain mediates its oligomerization.
Recently, Mitchell et al., reported that Tip is present as monomeric form in solution based on hydrogen-exchange mass spectrometry and circular dichroism study [20]. However, the recombinant protein they used contains only cytoplasmic region without transmembrane domain of Tip. Thus, our current result using the full-length Tip protein including transmembrane domain is more appropriate in reflecting natural status of Tip in vivo.
Oligomerization of Tip was further confirmed using blue native polyacrylamide gel electrophoresis with detergent-solubilized 293T cells expressing GFP-Tip211-256 or GFP-Tip CD71TM211-256. Truncated forms of the GFP fusion proteins, with short cytoplasmic domains, were used so as to minimize potential interactions with other cellular proteins, and to facilitate more accurate estimates of the size of the oligomeric protein. As shown in Figure 8B, a GFP fusion protein carrying the TM domain of wild type Tip (GFP-Tip211-256) migrated as a ∼150 kDa protein, whereas GFP-Tip CD71TM211-256 migrated as a ∼40 kD protein, corresponding to the size of the monomeric form of GFP-Tip CD71TM211-256. The size of the ∼150 kD protein complex is suggestive of homo-oligomers of four GFP-Tip211-256 monomers.
Lipid rafts contain proteins that retain their association with membrane lipids. These proteins are mostly GPI-anchored or acylated, but a few are transmembrane proteins, which are targeted to lipid rafts through their TM domain or through membrane-proximal determinants [21]. Here, we have found that the presence of a membrane-proximal amphipathic helix, located in Tip's cytoplasmic face, significantly contributed to Tip's localization in the lipid raft. Extensive mutagenesis analysis revealed that the residues forming both the hydrophobic ridge and the positively-charged face of the helical motif are important for Tip's efficient association with lipid rafts (Figure 2). The segregation of hydrophobic and polar residues into two opposite faces of the helical structure matches well with the chemistry of the membrane interface, and has been suggested to contribute to membrane adsorption [22],[23]. It has also been suggested that the amphipathic helical motif might target caveolin to lipid rafts through partial insertion of the hydrophobic ridge into lipid bilayer, and electrostatic interaction of the charged surface with phospholipids [24]. Another lipid raft-residing protein, α-synuclein [25], is also anchored to membranes by an elongated amphipathic helical structure [26]. Although the specificity of the amphipathic helical motifs for lipid rafts has been poorly defined, binding of α-synuclein to raft-like liposomes was shown to require acidic phospholipids, with a preference for phosphatidylserine [25]. In T cells, cholesterol and negatively-charged phospholipids are concentrated in the ordered raft domains upon antigenic stimulation [1]. Interestingly, a recent report showed that many signaling and transport proteins contain clusters of positively-charged amino acids, suggesting that those clusters could mediate the plasma membrane-targeting of proteins, through interaction with acidic phospholipids [27]. In this study, we showed that the amphipathic helical peptide of Tip specifically interacts with negatively-charged lipids such as PS, PA, PG, and PtdIns rather than neutral or amino phospholipids (Figure 6), suggesting that positively-charged amino acids within the amphipathic helical motif might associate with these negatively-charged lipids. This possibility is consistent with previous findings, which showed that the cytoplasmic leaflet of biological membranes is enriched in negatively-charged lipids, and that lipid rafts are enriched with PS, PA, and PG [28]. As suggested by the results of binding assays utilizing artificial liposomes containing cholesterol (Figure 6C), however, the presence of the hydrophobic ridge might restrict the interaction of the peptide with lipid rafts. In fact, α-synuclein binds more strongly to membranes containing low or no cholesterol [25] and the binding affinity of an amphipathic peptide to unilamellar vesicles is reduced by the presence of cholesterol [29]. The rigidifying effect of cholesterol on phospholipid acyl chains may limit the penetration of the peptide into the bilayer interior. Taken all together, the interaction of the amphipathic helical region of Tip with lipid bilayer might be heavily dependant on the membrane lipid composition.
Previously, we found that the TM domain of Tip is required for association with lipid rafts [11]. In this study, we found that the TM domain alone confers weak (∼20%) lipid raft association (Figure 1A) but is sufficient to mediate oligomerization of Tip (Figure 8). These results suggested that the TM domain might play a cooperative role in lipid raft association together with the lipid binding amphipathic helix; i.e. the both domains are required for the efficient association of Tip with lipid rafts. With regard to the relationship between lipid raft association and protein oligomerization, varying results have been reported. In some cases, self-assembling or ligand-induced oligomerization was required for proteins to associate efficiently with the lipid raft [30],[31], whereas oligomerization and lipid raft formation were independent in other cases [32]. Alanine scan mutagenesis was performed in an effort to elucidate the role of the Tip TM domain in lipid raft localization and oligomerization (Figure S8). A lipid raft fractionation assay using Tip mutants carrying four consecutive alanine residues in the TM domain showed that two mutants, carrying alanine residues in the regions 240–244 or 249–253, associate with lipid rafts to the same degree as wild type Tip, and that other mutants were able to associate with lipid rafts to a limited degree of approximately 40 to 47% (Figure S8B). These observations imply that the amino acids 231–241 and the TVLS motif, which are thought to interact with the inner and outer leaflets of biological membranes respectively, were both required for efficient lipid raft association, suggesting that the specific amino acid sequences comprising the Tip TM domain might contribute to the interaction with lipid raft domains. The immunoprecipitation assay, using cells expressing AU1-tagged wild type Tip and flag-tagged Tip mutants generated by the alanine scan mutagenesis, showed that all mutants tested co-precipitated with wild type Tip (Figure S8C). As such, oligomerization might be mediated by regions longer than four consecutive amino acids, or by multiple contacts within the TM domain, rather than by a single or local contact. Since no motif specific for the oligomerization of Tip could be identified, the precise relationship between protein oligomerization and lipid raft association has yet to be determined, and will be the topic of future studies.
It could be questioned whether the mutations in the amphipathic helix or transmembrane domain of Tip could affect the efficiency of membrane association itself rather than lipid raft localization. However, when we examined the membrane association of Tip and its mutants, Tip amp1 or Tip CD71TM, without detergent treatment during membrane fractionation, there was no significant difference in the degree of membrane association of the Tip proteins (Figure S9). Rather, the degree of membrane association of Tip amp1 or Tip CD71TM was slightly enhanced when compared to that of Tip wt. This result suggested that the mutations in the amphipathic helix or transmembrane domain could change the raft localization property of Tip without significantly affecting the efficiency of membrane association itself.
Vesicle trafficking involves dynamic remodeling of cellular membranes, for which the formation of local membrane curvature is a critical step [12],[33]. It has recently been shown that generation of membrane curvature can be driven by the interplay between lipids and proteins, through several mechanisms [12]. An emerging theme among these mechanisms is the involvement of amphipathic peptides that partially penetrate the lipid bilayer, acting as wedges. Active insertion of helical peptides into the bilayer results in an increase of surface area in one leaflet, possibly generating spontaneous curvature in the bilayer. This local curvature is subsequently sensed and stabilized by other domains of the curvature-forming proteins, or by coat proteins. For example, the N-terminal amphipathic helix found in the BAR domain of amphiphysin and endophilin has been shown to cause local membrane curvature, which is stabilized by a banana-shaped lipid-binding domain [19],[34],[35]. Helical domains found in other proteins such as epsin, Arf, and Sar1 were also shown to generate local membrane curvature in an induced manner and subsequently recruit coat proteins to stabilize the curvature [17],[18],[36]. The power of an amphipathic peptide to generate membrane deformation was previously demonstrated when a designed 18-mer peptide was shown to form extensive, 40–50 nm diameter tubules from liposomes [37]. They showed that that the deformation of liposomes depended on lipid composition and peptide properties such as length and the ratio of hydrophobic to hydrophilic amino acids. In the present study, we showed that an amphipathic 18-mer peptide derived from Tip's membrane-proximal helix can efficiently induce membrane deformation in an in vitro liposome tubulation assay (Figure 7). The point mutation of the conserved hydrophobic residues in this peptide into basic lysine residues improved liposome binding, in comparison to the wild type peptide, but abolished tubulation (Figure 7). These results suggest that Tip's membrane-proximal amphipathic helix is likely to alter membrane structure in a manner similar to that employed by other cellular proteins containing amphipathic helices. In fact, GFP fusion peptides encompassing the helical region and TM domain of Tip were detected in vesicular curvatures and multivesicular structures within endocytic vesicles of Jurkat T cells (Figure 5), whereas the mutant fusion peptides were not. Luminal budding of the limiting membrane and the formation of MVBs in the late endosomal pathway are efficient mechanisms for targeting membrane proteins/receptors to lysosomes for degradation [3]. Through its amphipathic helical motif, Tip might initiate the luminal budding step, which might be further enhanced by oligomerization through the TM domain. Considering that Tip's structural influence on the lipid bilayer is generated in the cytoplasmic face of endocytic vesicles, and that Tip lacks a curvature-sensing/stabilizing domain, the luminal budding would likely be assisted by other cellular proteins. Recently, it was shown that an inverse BAR domain-like mechanism in the proteins IRSp53 and MIM (missing-in-metastasis) induces a membrane curvature opposite to that of BAR domains, and deforms membranes by binding to their interior, resulting in plasma membrane protrusions rather than invaginations [38]. Thus, Tip-associated luminal budding may be facilitated by cellular proteins with an inverse BAR domain-like mechanism, which may be recruited through direct or indirect interactions with Tip. Vps (vaculolar protein sorting) proteins have been shown to be involved in lysosomal degradation of activated receptors through the MVB-sorting pathway in yeast and mammals [3],[39],[40]. Some mutations in those genes resulted in an enlarged late endosomal compartment, presumably because of an inability to invaginate the limiting membrane to form the MVB. It would be interesting to identify whether the protein sorting machineries could mediate inverse BAR domain-like functions in Tip associated-MVB formation.
Expression of Tip in T cells was previously reported to induce clustering of lipid raft domains as well as redistribution of TCR/CD3 complexes into lipid raft domains [9],[10]. Tip expression can also reorganize raft domains and enhance the recruitment of raft-resident components [11]. Similarly, T cell activation leads to the segregation of plasma membrane domains to form TCR signaling clusters, and this is accompanied by the condensation of the plasma membrane, driven by activation-induced protein-protein interactions such as anchorage to the cytoskeleton [41]. The clustered raft domain platforms are subsequently internalized and degraded in the lysosome to attenuate TCR signaling [42]. Previously, we showed that the TM domain is essential for the downregulation of TCR/CD3 complexes and CD4 by Tip [11]. Downregulation of the membrane proteins, however, is mediated through different mechanisms [9],[10]. Downregulation of TCR/CD3 complexes by Tip is dependent on its interaction with and kinase activity of Lck as well as the interaction with p80, whereas downregulation of CD4 by Tip is dependent on the physical association with Lck only. In the present study, we showed that Tip's membrane-proximal amphipathic helix, consisting of 14 amino acids, was essential for the selective downregulation of TCR/CD3 complexes but not for CD4. The effect of this short motif on the receptor trafficking might be linked to the functional properties of raft-targeting and membrane deformation, as mentioned above. Lysosomal targeting through MVB formation by the amphipathic helix and TM domain of Tip suggested how clustered TCR/CD3 complexes in lipid raft domains are targeted for lysosomal degradation in Tip-expressing T cells. In contrast to TCR/CD3 complexes, CD4 surface expression is downmodulated consistently by the Tip mutants lacking amphipathicity in their membrane proximal helix (Figure 3). As such, downregulation of CD4 surface expression could be mediated by different mechanisms. Although the molecular mechanisms of CD4 trafficking in resting and activated T cells are largely unknown, it is interesting to note that human immunodeficiency virus has dual arms, Nef and Vpu, to downregulate CD4 surface expression through distinct mechanisms [43]. Nef links mature CD4 to components of clathrin-dependent trafficking pathways at the plasma membrane, and perhaps also in intracellular compartments, leading to internalization and delivery of CD4 to lysosomes for degradation. Vpu, on the other hand, interacts with newly-synthesized CD4 in the endoplasmic reticulum, linking CD4 to the SCF ubiquitin ligase, and facilitating the entry of CD4 into the endoplasmic reticulum-associated degradation pathway. The Tip-associated molecular mechanisms controlling CD4 expression remain to be elucidated.
The phenotypic resemblance of Tip and TCR activation, leading to Lck activation, recruitment of TCRs to lipid rafts and finally to lysosomal degradation, suggests that HVS Tip may pirate cellular signaling molecules to emulate TCR stimulation for viral persistence and pathogenesis. Epstein-Barr virus LMP2A, a functional homologue of HVS Tip, has also been shown to be mimic the B cell receptor signal transduction to maintain viral latency, allowing long-term survival of infected B cell [44],[45]. This viral protein interacts with B-cell signaling proteins, such as Lyn and Syk, through its N-terminal cytoplasmic tail. LMP2A functions in lipid rafts to block translocation of the B-cell receptor into lipid rafts, which leads to inhibition of the subsequent signaling and accelerated internalization of the BCR-cell receptor upon stimulation. Thus, the study of HVS Tip may provide valuable insight into the conserved mechanisms employed by other γ-herpesvirus signal modulators to regulate lymphocyte functions and may have significant implications for the understanding of viral persistence and pathogenesis.
In summary, we have shown that a potential membrane-proximal amphipathic helix preceding the TM domain of Tip is essential for efficient lipid raft localization and selective downregulation of TCR/CD3, most likely through mechanisms involving membrane curvature and MVB formation in endocytic vesicles. Moreover, we could dissect the functional roles of the amphipathic helix and the TM domain in membrane deformation and oligomerization, respectively. These novel mechanisms of the viral protein could provide valuable insights into the functional relationship between lipid rafts and MVB formation and the molecular details of membrane trafficking of the key receptors in T cells.
Jurkat T cells were grown in RPMI, and 293T and HeLa cells were maintained in DME medium, supplemented with 10% FBS. Jurkat T cells were electroporated using a Bio-Rad electroporator at 260V and 975µF in serum-free RPMI medium. Lipofectamine2000 (Invitrogen) or calcium phosphate (Clontech) was used to induce transient expression of Tip in HeLa and 293T cells. Stable Jurkat T cell lines expressing Tip or its mutants were selected and maintained in the presence of puromycin (5 µg/ml). We tested the expression level of Tip or its mutants in each cell line by semi-quantitative RT-PCR using β-actin gene as internal control [46]and confirmed the similar level of expression in all the established cell lines (data not shown). An anti-GFP antibody (Santa Cruz Biotechnology), a CTB-HRP conjugate (Sigma), an anti-EEA1 antibody, and an anti-LAMP2 antibody (BD Bioscience) were used for immunoassays. DNA fragments encoding Tip and its mutants were cloned into pFJ, pBabe or p3xFlag_CMV vectors (Sigma) using methods described previously [11]. GFP fusion proteins containing Tip or its mutants were made using pEGFP-C2 plasmids (Clontech). PCR-based mutagenesis was performed to create the Tip mutants, using sequences described previously [10],[11]. A peptide corresponding to the amphipathic helical region of Tip (Tip wt211-228, ANERNIVKDLKRLENKIN) and a mutant peptide in which hydrophobic residues were replaced with lysines (underlined; Tip amp1211-228, ANERNKVKDKKRKENKKN) were synthesized by Peptron Inc. A lysine residue was added to the C-terminus of each peptide to allow them to be conjugated with biotin.
Lipid rafts were isolated using a method involving flotation on discontinuous sucrose gradients [10]. Briefly, 5×107 293T cells were washed with ice-cold PBS and lysed for 30 min on ice in 1% Triton X-100 in TNEV buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM EDTA) containing phosphatase inhibitors and protease inhibitor cocktail (Roche). The lysates were further homogenized in a Wheaton loose-fitting Dounce homogenizer. Nuclei and cellular debris were pelleted by centrifugation at 900×g for 10 min. For the discontinuous sucrose gradient, 0.5 ml of cleared cell lysates were mixed with 0.5 ml of 85% sucrose in TNEV and transferred to a Beckman 14×89 mm centrifuge tube. Diluted lysates were overlaid with 4 ml of 35% sucrose in TNEV and finally 1 ml 5% sucrose in TNEV. Samples were then centrifuged in an SW41 rotor at 200,000×g for 20 h at 4°C, and 0.5 ml fractions were collected from the top of the gradient.
Membrane-enriched fraction was prepared to examine the efficiency of membrane association of Tip and its mutants as described elsewhere with slight modification [47]. In brief, 293 T cells expressing GFP fusion proteins containing Tip or its mutants were harvested and resuspended in lysis buffer (50 mM Tris, pH 7.8, 250 mM Sucrose, and 2 mM EDTA) with protease inhibitor cocktail (Roche). After incubation on ice for 10 min, cells were lysed by 30 strokes using Dounce homogenizer at 4°C. Cellular debris and nuclei were removed by centrifugation at 1000×g for 10 min at 4°C. The postnuclear supernatant was layered onto a 60% sucrose cushion and centrifuged at 160,000×g for 1 h at 4°C. The membrane fraction on top of the sucrose cushion was collected, diluted 1∶2 with cold phosphate-buffered saline (PBS; 100 mM phosphate, 150 mM NaCl, pH 7.2) and pelleted at 100,000×g for 1 h at 4°C. The supernatant was discarded and the membrane pellet was rinsed twice with cold PBS, and pelleted at 20,000×g for 30 min at 4°C. The enriched membrane fraction was further used for SDS-PAGE and subsequent immunoblot assay.
Blue Native PAGE was performed as described previously [48] with slight modifications. Homogenized cells were solubilized by adding Triton X-100 to a final concentration of 2.5%. After removing cellular debris by centrifugation, the whole-cell lysates were collected and resolved by native gel (10%) electrophoresis. Resolved proteins were transferred to a PVDF membrane and detected by immunoblot assay. Aldorase from rabbit muscle (∼160 kDa, Sigma) and bovine serum albumin (monomer: ∼66 kDa, dimer ∼132 kDa, Sigma) were used as molecular weight markers.
Cells (5×105) were washed with RPMI medium containing 10% fetal calf serum, and incubated with fluorescein isothiocyanate-conjugated or phycoerythrin-conjugated monoclonal antibodies for 30 min at 4°C. After washing, each sample was fixed with 4% paraformaldehyde solution and flow cytometric analysis was performed with a FACScan (Becton Dickinson Co.). Antibodies against CD3 (SK7), CD4 (Leu-3a), CD45 (HI30), and αβTCR were purchased from BD Pharmingen.
Cells were fixed with 4% paraformaldehyde for 15 min, permeabilized with 0.2% Triton X-100 for 15 min, and reacted with primary antibodies in PBS for 30 min at room temperature. Alexa 488- or Alexa 594-conjugated anti-rabbit or anti-mouse antibodies (Molecular Probes) were used as secondary antibodies. Confocal microscopy was performed using an Olympus FV1000 laser-scanning microscope (Olympus) fitted with a 60× Olympus objective. Images were collected at 512×512 pixel resolution using Olympus imaging software. The stained cells were optically sectioned in the z-axis, and the images in the different channels (photo multiplier tubes) were collected sequentially. The images were rendered using Olympus Fluoview v1.6b or Adobe Photoshop software. To quantify the degree of relative colocalization, we obtained the Pearson correlation coefficient (R) values, which are standard measures of colocalization [49]. The R values were calculated using the Olympus Fluoview v1.6b colocalization module which generates a “colocalized” image from two channels.
For immunoprecipitation, cells were harvested and resuspended in lysis buffer (150 mM NaCl, 0.5% Nonidet P-40, and 50 mM HEPES buffer, pH 7.4) containing protease inhibitors. Immunoprecipitated proteins from precleared cell lysates were used for immunoblot. For immunoblot, polypeptides were resolved by SDS-PAGE and transferred to a PVDF membrane. Immunoblot detection was performed with a 1∶1000 or 1∶3000 dilution of primary antibody and an enhanced chemiluminescence system (Pierce).
Membrane lipid strips and arrays (Echelon Biosciences) were used for peptide-lipid binding assays according to the manufacturer's instructions. Peptides (0.4 µM) were incubated overnight at 4°C and detected with streptavidin-HRP conjugates. Densitometric analysis was applied to determine the relative affinity of peptide binding to the various lipids. After subtracting background values, numerical densitometric values were attributed to each of the five concentrations measured. The highest value, for binding of peptides to 100 pmol of lipids, was arbitrarily assigned “100% binding” and all other lipids were normalized in comparison to that maximum binding value. Synthetic liposomes were made using phosphatidylcholine (65% mol/mol), phosphatidylserine (25% mol/mol), and cholesterol (10% mol/mol; Avanti Polar Lipids Inc.), as described previously [19]. To achieve desired diameters, the liposomes were extruded more than 10 times through a polycarbonate membrane (Avanti). The size of the liposomes was measured by laser light scattering analysis (Brookhaven Instruments Co.). For the peptide binding assays, liposomes were diluted in 100 µl of binding buffer (20 mM HEPES, pH 7.4, 150 mM NaCl) at a final lipid concentration of 2 mM and incubated for 10 min at room temperature with peptides (4 µM). Liposome-protein complexes were recovered by centrifugation (100,000×g) at room temperature for 20 min, the supernatant was completely removed, and sedimented liposomes were solubilized in SDS sample buffer. The peptides in the supernatant and pellet were subjected to SDS-PAGE using 16% tricine gels, and analyzed as described above.
Jurkat T cells expressing GFP fusion proteins were fixed in 0.5% glutaraldehyde and 4% paraformaldehyde in 0.05 M sodium cacodylate buffer (pH 7.2) at 4°C for 2 h. Ultrathin sections (50 nm in thickness) were cut using an ultramicrotome (MT-X; RMC) and stained with an anti-GFP primary antibody and an anti-rabbit IgG secondary antibody conjugated with 10 nm gold particles (Sigma). Sections were then stained with 2% uranyl acetate and Reynolds' lead citrate, and examined by transmission electron microscopy (LIBRA 120; Carl Zeiss) at an accelerating voltage of 120 kV. Negative control experiments were also performed to ensure the specificity of the labeling by replacing the primary antibody with rabbit preimmune serum. Liposomes incubated with peptides as described in liposome binding assays were adsorbed onto carbon-coated copper grids, stained with uranyl acetate, and then observed by electron microscopy.
The GenBank (http://www.ncbi.nlm.nih.gov/Genbank) accession numbers for Tip and Tio proteins used in this paper are Tip C488 (AAA72928), Tip C484 (P88825), Tip C484-77 (P25049), and Tio (AAC95538).
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.